WO2011021911A2

WO2011021911A2 - Method and apparatus for encoding/decoding images using adaptive motion vector resolution

Info

Publication number: WO2011021911A2
Application number: PCT/KR2010/005568
Authority: WO
Inventors: 김수년; 임정연; 문주희; 이영렬; 김해광; 전병우; 이주옥; 김동원
Original assignee: 에스케이텔레콤 주식회사
Priority date: 2009-08-21
Filing date: 2010-08-21
Publication date: 2011-02-24
Also published as: WO2011021911A3

Abstract

The present invention relates to a method and an apparatus for encoding/decoding images using adaptive motion vector resolution. The method for encoding the images of the present invention comprises the steps of: determining motion vector resolution in each region or each motion vector; and performing inter-prediction encoding using the motion vector according to the motion vector resolution determined in each region or each motion vector. The present invention enables efficient encoding of images by adaptively changing the motion vector resolution in predetermined region units of the images and performing the inter-prediction encoding for the images.

Description

Method and apparatus for image coding / decoding using adaptive motion vector resolution

The present invention relates to a method and apparatus for image encoding / decoding using adaptive motion vector resolution. More specifically, the present invention relates to a method and apparatus for improving coding efficiency by adaptively changing the resolution of a motion vector in inter-prediction encoding and inter-prediction decoding of an image.

The contents described in this section merely provide background information on the embodiments of the present invention and do not constitute a prior art.

The encoding of data for a video is made up of intra prediction encoding and inter prediction encoding. Such intra prediction encoding or inter prediction encoding is widely used for compressing various data in an effective way to reduce correlation between data. In particular, since the motion vector of the current block determined by estimating the motion of the current block to be currently encoded by inter prediction encoding is closely correlated with the motion vector of the neighboring block, the motion vector of the neighboring block is determined from the motion vector of the current block. After calculating the predicted motion vector (PMV: Predicted Motion Vector), the differential value of the predicted value (DMV: Differential Motion Vector) is calculated without encoding the value of the motion vector of the current block itself. By coding only '), the amount of bits to be encoded can be significantly reduced, thereby improving the coding efficiency.

That is, in the case of performing inter prediction encoding, the encoder encodes and transmits only a difference vector, which is a difference between the predicted motion vector and the current motion vector determined by estimating the motion of the current block in a previously encoded, decoded and reconstructed reference frame. . The decoder also predicts the motion vector of the current block by using the motion vector of the neighboring block previously decoded, and reconstructs the current motion vector by adding the transmitted difference vector and the predicted motion vector.

In addition, when performing inter prediction encoding, the resolution may be collectively increased by interpolating a reference frame, and then encoded by a difference vector, which is a difference between the prediction motion vector determined by estimating the motion of the current block and the current motion vector, and then transmitted. have. In this case, when the resolution of the reference frame image, that is, the reference image, is increased, more accurate inter prediction is possible, thereby reducing the amount of bits generated by encoding the residual signal between the original image and the predicted image, but increasing the resolution vector of the motion vector. The amount of bits generated by encoding also increases. On the contrary, when the resolution of the reference picture becomes smaller, the bit amount generated by encoding the residual signal becomes larger, but since the resolution of the motion vector becomes smaller, the bit amount generated by encoding the difference vector also becomes smaller.

As described above, in the conventional inter prediction encoding, since the interpolation is performed with the same resolution and the motion vector having the same resolution is interpolated with the same resolution for all blocks, slices, pictures, and the like of the image, efficient encoding is difficult. There is a problem that the efficiency is lowered.

In order to solve the above-described problem, the present invention has an object of improving the coding efficiency by adaptively changing the resolution of a motion vector in inter prediction encoding and inter prediction decoding of an image.

In order to achieve the above object, the present invention provides a video encoding / decoding apparatus for determining a motion vector resolution for each region or motion vector and inter-prediction encoding using a motion vector according to the motion vector resolution determined for each region or motion vector. Image encoder; And an image decoder for extracting resolution information from the bitstream, restoring the resolution, and performing inter-prediction decoding using a motion vector according to the motion vector resolution of each reconstructed region or motion vector. to provide.

According to another object of the present invention, there is provided an apparatus for encoding an image, comprising: a resolution determiner for determining a motion vector resolution for each region or motion vector; And an inter prediction encoder configured to perform inter prediction encoding using a motion vector according to a motion vector resolution determined for each region or motion vector.

According to still another object of the present invention, there is provided an apparatus for decoding an image, comprising: a resolution change flag extractor for extracting a resolution change flag from a bitstream; A resolution decoder configured to extract and decode a coded resolution identification flag from the bitstream according to the extracted resolution change flag to restore a motion vector resolution for each region or motion vector; And an inter prediction decoder configured to inter-prediction-decode using a motion vector according to the motion vector resolution of each reconstructed region or motion vector.

According to another object of the present invention, there is provided an apparatus for decoding an image, comprising: a resolution decoder for reconstructing the resolution of each region or motion vector by using information received from a bitstream; And an inter prediction decoder configured to perform inter prediction decoding using a motion vector according to the resolution of each of the reconstructed regions.

According to still another object of the present invention, in a method of encoding / decoding an image, the motion vector resolution is determined for each region or motion vector, and the inter prediction is performed by using the motion vector according to the motion vector resolution determined for each region or motion vector. An image encoding step of encoding; And an image decoding step of extracting resolution information from the bitstream, restoring the resolution, and inter-prediction decoding using a motion vector according to the motion vector resolution of each reconstructed region or motion vector. To provide.

According to still another object of the present invention, there is provided a method of encoding an image, the method comprising: determining a motion vector resolution for each region or motion vector; And inter-prediction encoding using a motion vector according to a motion vector resolution determined for each region or motion vector.

According to still another object of the present invention, there is provided a method of decoding an image, comprising: extracting a resolution change flag from a bitstream; Extracting and decoding the encoded resolution identification flag from the bitstream according to the extracted resolution change flag to restore motion vector resolution for each region or motion vector; And inter-prediction decoding using a motion vector according to the motion vector resolution of each reconstructed region or motion vector.

According to still another object of the present invention, there is provided a method of decoding an image, comprising: a resolution decoding step of reconstructing a resolution of each region or motion vector by using information received from a bitstream; And an inter prediction decoding step of performing inter prediction decoding by using a motion vector according to the resolution of each of the reconstructed regions.

As described above, according to the present invention, the image can be efficiently encoded by inter-prediction coding by adaptively changing the motion vector resolution in predetermined area units.

1 is a block diagram schematically illustrating a video encoding apparatus;

2 is an exemplary diagram for explaining a process of interpolating a reference picture;

3 is an exemplary diagram for explaining a process of determining a predicted motion vector;

4 is an exemplary diagram showing a cutting type unary code having a maximum value T of 10, and FIG. 5 is an exemplary diagram illustrating an exponential golem code of 0, 1, and 2;

5 is an exemplary diagram showing exponential Gollum codes of 0, 1, and 2,

6 is an exemplary view showing a cut-unary / K-order index gollum combined code when the maximum value T is 9 and K is 3;

7 is an exemplary diagram illustrating a zigzag scan order;

8 is a block diagram schematically illustrating a video decoding apparatus;

9 is a block diagram schematically illustrating a video encoding apparatus according to an embodiment of the present invention;

10 is an exemplary diagram hierarchically expressing a motion vector resolution in a quad tree structure according to an embodiment of the present invention;

FIG. 11 is an exemplary diagram for describing a process of encoding a motion vector resolution hierarchically represented in a quad tree structure according to an embodiment of the present invention; FIG.

12 is an exemplary diagram illustrating a motion vector resolution of a region determined according to an embodiment of the present invention.

13 is an exemplary diagram hierarchically expressing a motion vector resolution in a tag tree structure according to an embodiment of the present invention;

14 is an exemplary diagram illustrating a result of encoding a motion vector resolution expressed hierarchically expressed in a tag tree structure according to an embodiment of the present invention;

15 is an exemplary diagram for describing a process of determining a motion vector resolution using pixels around a region according to an embodiment of the present invention;

16 is an exemplary diagram for explaining a process of predicting a predicted motion vector according to an embodiment of the present invention;

17 is a flowchart illustrating a video encoding method using an adaptive motion vector resolution according to an embodiment of the present invention.

18 is a block diagram schematically illustrating an apparatus for decoding an image using adaptive motion vector resolution according to an embodiment of the present invention;

19 is a flowchart illustrating an image decoding method using adaptive motion vector resolution according to an embodiment of the present invention.

20 is an exemplary diagram illustrating another method of dividing a node into lower layers according to an embodiment of the present invention;

21 is an exemplary diagram illustrating a binarization bit string allocated to each symbol according to a motion vector resolution according to an embodiment of the present invention;

22 is a diagram illustrating an optimal motion vector of a neighboring block and a current block to explain a process of determining a resolution of a motion vector;

23 is a diagram illustrating a conversion equation according to a motion vector resolution;

24 is a diagram illustrating a table obtained by converting a resolution of a motion vector of a neighboring block into a resolution of a block to be currently encoded;

25 is a diagram of a code number table of differential motion vectors according to motion vector resolution;

FIG. 26 is a diagram illustrating an optimal motion vector of a neighboring block and a current block to explain a process of determining a resolution of a differential motion vector; FIG.

27 is a diagram illustrating a differential motion vector code number table according to differential motion vector resolution;

28 is a diagram illustrating a motion vector of a current block and a reference motion vector of a neighboring block;

29 is a diagram illustrating a code number table of a differential reference motion vector according to a differential reference motion vector resolution;

30 is a diagram illustrating an example of indexing and encoding a reference picture based on a distance between a current picture and a reference picture;

31 is a diagram illustrating a reference picture index according to a reference picture number and a resolution;

32 is a diagram schematically illustrating an image encoding apparatus using an adaptive motion vector according to another embodiment of the present invention;

33 is a diagram illustrating a resolution identification flag when the specified resolutions are 1/2 and 1/4;

34 is a diagram illustrating a neighboring block of a current block;

35 is a diagram illustrating a probability model according to a condition;

36 is a diagram illustrating a resolution identification flag when the specified resolution is 1/2, 1/4, 1/8;

37 is a view showing another example of a probability model different from a condition;

38 and 39 are diagrams illustrating the degree of adaptability according to the distance between a current picture and a reference picture;

40 is a diagram showing an example of storing a different reference picture index number for each predetermined resolution set;

41 is a diagram illustrating a structure in which reference pictures are encoded;

42 is a diagram illustrating a resolution set for each reference picture when the resolution set is specified as 1/2 and 1/4;

43 is a diagram illustrating a resolution of a current block and a resolution of a neighboring block;

44 is a diagram illustrating another example of a resolution of a current block and a resolution of a neighboring block;

45 is a diagram illustrating a resolution identification flag according to a resolution;

46 is a diagram illustrating a resolution of a current block and a resolution of a neighboring block;

47 is a view showing a video encoding method using an adaptive motion vector resolution according to another embodiment of the present invention;

48 schematically illustrates an image decoding apparatus using adaptive motion vector resolution according to an embodiment of the present invention;

49 is a diagram illustrating a peripheral motion vector of the current block;

50 is a diagram illustrating a converted value of a peripheral motion vector according to a current resolution;

51 is a flowchart illustrating a method of decoding an image using adaptive motion vector resolution according to another embodiment of the present invention.

Hereinafter, some embodiments of the present invention will be described in detail through exemplary drawings. In adding reference numerals to the components of each drawing, it should be noted that the same reference numerals are assigned to the same components as much as possible even though they are shown in different drawings. In addition, in describing the present invention, when it is determined that the detailed description of the related well-known configuration or function may obscure the gist of the present invention, the detailed description thereof will be omitted.

In addition, in describing the component of this invention, terms, such as 1st, 2nd, A, B, (a), (b), can be used. These terms are only for distinguishing the components from other components, and the nature, order or order of the components are not limited by the terms. If a component is described as being "connected", "coupled" or "connected" to another component, that component may be directly connected or connected to that other component, but between components It will be understood that may be "connected", "coupled" or "connected".

A video encoding apparatus (Video Encoding Apparatus), a video decoding apparatus (Video Decoding Apparatus) to be described below is a personal computer (PC), notebook computer, personal digital assistant (PDA), portable multimedia player (PMP) : Portable Multimedia Player (PSP), PlayStation Portable (PSP: PlayStation Portable), Wireless Communication Terminal (Wireless Communication Terminal), TV (Television), etc., Communication device for communication with various devices or wired or wireless communication network, etc. The present invention refers to various devices including various programs for encoding or decoding an image, a memory for storing data, and a microprocessor for executing and operating a program.

In addition, the image encoded in the bitstream by the video encoding apparatus is real-time or non-real-time through the wired or wireless communication network, such as the Internet, local area wireless communication network, wireless LAN network, WiBro network, mobile communication network, or the like, or a cable, universal serial bus (USB: Universal) The image decoding apparatus may be transmitted to a video decoding apparatus through a communication interface such as a serial bus, decoded by the video decoding apparatus, reconstructed, and played back.

In addition, although the image encoding apparatus and the image decoding apparatus may be provided with functions for performing intra prediction as well as inter prediction, detailed descriptions will be omitted in order to prevent confusion because they are not directly related to the embodiment of the present invention.

In general, a moving picture is composed of a series of pictures, and each picture is divided into a predetermined area such as a block. When a region of an image is divided into blocks, the divided blocks are largely classified into intra blocks and inter blocks according to encoding methods. An intra block refers to a block that is encoded by using an intra prediction coding method. An intra prediction coding is performed by using pixels of blocks that are previously encoded, decoded, and reconstructed in a current picture that performs current encoding. A prediction block is generated by predicting pixels of a block, and a difference value with pixels of the current block is encoded. An inter block refers to a block that is encoded using inter prediction coding. Inter prediction coding generates a prediction block by predicting a current block within a current picture by referring to one or more past or future pictures, and then generates a current block. This is a method of encoding a difference value with. Here, a picture referred to for encoding or decoding the current picture is referred to as a reference picture.

Hereinafter, an apparatus for encoding and decoding an image in units of blocks will be described with reference to FIGS. 1 through 8. Here, the block may be a macroblock of MxN size or a subblock or subblock of OxP size. For example, encoding and decoding in units of blocks is an example, and an image is encoded into a standardized area or an unstructured area such as a block. You may be able to decrypt it.

1 is a block diagram schematically illustrating a video encoding apparatus.

The image encoding apparatus 100 may include a predictor 110, a subtractor 120, a transformer 130, a quantizer 140, an encoder 150, an inverse quantizer, and an inverse quantizer. 160, an inverse transformer 170, an adder 180, and a memory 190.

The predictor 110 inter-predictions the current block to generate a predicted block. That is, the predictor 110 determines the original pixel values of the pixels of the current block through motion estimation when the block to be currently encoded, that is, the current block, is input from the input picture. A prediction block having predicted pixel values is generated and output by predicting using a motion vector of the current block.

The subtractor 120 subtracts the current block and the prediction block to generate a residual block of the current block. Here, the output residual block includes a residual signal, which has a value obtained by subtracting the prediction pixel value of the prediction block from the original pixel value of the current block.

The transformer 130 transforms the residual block to generate a transformed block. That is, the transformer 130 converts the residual signal of the residual block output from the subtractor 120 into a frequency domain to generate and output a transform block having a transform coefficient. Here, as a method of transforming the residual signal into the frequency domain, a Discrete Cosine Transform (DCT) -based transform or a Hadamard Transform (DCT) may be used, but the present invention is not limited thereto. Various transformation techniques such as the technique may be used, and the residual signal is transformed into a frequency domain and transformed into transformation coefficients.

The quantizer 140 quantizes the transform block to generate a quantized transform block. That is, the quantizer 140 quantizes the transform coefficients of the transform block output from the transformer 130 to generate and output a quantized transform block having quantized transform coefficients. Here, dead zone uniform threshold quantization (DZUTQ) or quantization weighted matrix (DZUTQ) may be used as the quantization method, but various quantization methods such as quantization improved therefrom may be used.

The encoder 150 may output a bitstream by encoding the quantized transform block. That is, the encoder 150 uses various encoding techniques, such as entropy encoding, on a frequency coefficient string scanned by various scan methods such as zigzag scan of the quantized transform coefficients output from the quantizer 140. To generate and output a bitstream including additional information necessary for encoding and decoding the corresponding block (for example, information on a prediction mode, a quantization coefficient, a motion vector, etc.). can do.

Inverse quantizer 160 inverse quantizes the quantized transform block. That is, the inverse quantizer 160 inversely quantizes and outputs quantized transform coefficients of the quantized transform block output from the quantizer 140.

Inverse transformer 170 inversely transforms an inverse quantized transform block. That is, the inverse transformer 170 inversely transforms the transform coefficients inversely quantized by the inverse quantizer 160 to restore a residual block having the reconstructed residual coefficients.

The adder 180 reconstructs the current block by adding the prediction block predicted by the predictor 110 and the residual block inversely transformed and reconstructed by the inverse transformer 170. The reconstructed current block is stored in the memory 190, and the reconstructed current block stored in the memory 190 is accumulated in units of blocks or in units of pictures and transferred to the predictor 110 in units of pictures, and then other blocks such as the next block or the next picture are stored. Can be used to predict blocks.

Meanwhile, the predictor 110 determines the motion vector of the current block by estimating the motion of the current block by using the reference picture stored in the memory 190. The resolution of the reference picture is interpolated by interpolating the reference picture stored in the memory 190. Can be used to perform motion estimation.

2 is an exemplary diagram for describing a process of interpolating a reference picture.

2 illustrates pixels of a reference picture stored in the memory 190 and pixels interpolated with subpixels. Filtering the previously reconstructed pixels A to U of the reference picture with an interpolation filter can generate subpixels a to s, and the subpixels a to s are interpolated between the already reconstructed pixels as shown. The resolution of the reference picture may be four times higher.

Motion Estimation is a process of finding a part most similar to the current block in the interpolated reference picture and outputting a block of the corresponding part and a motion vector indicating the corresponding block. The prediction block found in this process is subtracted from the current block in the subtractor 120, through which a residual block having a residual signal is generated. In addition, the motion vector is encoded by the encoder 160.

When encoding the motion vector, the encoder 160 may predict the motion vector of the current block by using the motion vectors of blocks adjacent to the current block and encode the motion vector by using the predicted motion vector, that is, the predicted motion vector. .

3 is an exemplary diagram for describing a process of determining a predicted motion vector.

Referring to FIG. 3, the current block is X, the motion vector of the adjacent block A on the left of the current block is MV_A (x component: MVx_A, y component: MVy_A), and the motion of the adjacent block B above the current block. Assuming that the vector is MV_B (x component: MVx_B, y component: MVy_B) and that the motion vector of the adjacent block C on the upper right side of the current block is MV_C (x component: MVx_C, y component: MVy_C), Each component of the predicted motion vector MV_pred_X (x component: MVx_pred_X, y component: MVy_pred_X) may be determined as a median value for each component of the motion vector of the neighboring block of the current block, as shown in Equation 1. Meanwhile, the motion vector prediction method in the present invention is not limited to the method introduced herein.

Equation 1

The encoder 160 may encode a difference vector having a difference between the predicted motion vector and the motion vector. As an encoding method for encoding the difference vector, for example, Universal Variable Length Coding (UVLC) Entropy Coding (UVLC ') and Context-Adaptive Binary Arithmetic Coding (CABAC) may be used. In the present invention, the encoding method in the encoder 160 is not limited to the method introduced here.

When the difference vector is encoded using UVLC, the difference vector may be encoded using a K-th Order Exp-Golomb Code. At this time, the value of K may be '0' or any other value. The prefix of the Kth order Gollum code

It has a truncated unary code corresponding to, and the suffix can be represented as a binary bit string of x + 2 ^k (1-2 ^{l (x)} ) values having a length of k + l (x). have.

4 is an exemplary diagram showing a cutting type unary code having a maximum value T of 10, and FIG. 5 is an exemplary diagram illustrating an exponential Golem code of 0, 1, and 2.

In addition, when the difference vector is encoded using CABAC, the difference vector is to be encoded using the sign bit of the concatenated Truncated Unary / K-th Order Exp-Golomb Code. Can be.

In a cut unary / K-order index golem combined code, the maximum value T is 9 and K can be 3. 6 is an exemplary view showing a cut-unary / K-order index gollum combined code when the maximum value T is 9 and K is 3. FIG.

7 is an exemplary diagram illustrating a zigzag scan sequence.

The quantized frequency coefficients quantized by the quantizer 140 may be scanned by the encoder 160 and encoded in the form of quantized frequency coefficient sequences. The quantized frequency coefficients in the block form may be scanned in a zigzag order as shown in FIG. 7. However, it may be scanned in various other scanning orders.

8 is a block diagram schematically illustrating a video decoding apparatus.

The image decoding apparatus 800 may include a decoder 810, an inverse quantizer 820, an inverse transformer 830, a predictor 840, an adder 850, and a memory 860.

The decoder 810 extracts a quantized transform block by decoding the bitstream. That is, the decoder 810 decodes and inversely scans the bit string extracted from the input bitstream to restore the quantized transform block having the quantized transform coefficients. In this case, the decoder 810 may decode using an encoding technique such as entropy encoding used by the encoder 160 of the image encoding apparatus 100.

In addition, the decoder 810 extracts and decodes the encoded difference vector from the bitstream, reconstructs the difference vector, predicts the motion vector of the current block, adds the predicted motion vector and the reconstructed difference vector, and adds the motion vector of the current block. Can be restored.

Inverse quantizer 820 inverse quantizes the quantized transform block. That is, the inverse quantizer 820 inverse quantizes the quantized transform coefficients of the quantized transform block output from the decoder 810. In this case, the inverse quantizer 820 performs inverse quantization by inversely performing the quantization technique used by the quantizer 140 of the image encoding apparatus 100.

The inverse transformer 830 inversely transforms the inverse quantized transform block output from the inverse quantizer 820 to restore the residual block. That is, the inverse transformer 830 restores the residual block having the reconstructed residual signal by inversely transforming the inverse quantized transform coefficients of the inverse quantized transform block output from the inverse quantizer 820. Inverse conversion is performed by performing the conversion technique used in the converter 130 of the inverse.

The predictor 840 generates a prediction block by predicting the current block by using the motion vector of the current block extracted from the bitstream and decoded.

The adder 850 reconstructs the current block by adding the prediction block and the reconstructed residual block. That is, the adder 850 adds the predicted pixel value of the predicted block output from the predictor 840 and the reconstructed residual signal of the restored residual block output from the inverse transformer 830 to add the reconstructed pixel value of the current block. Restore the current block.

The current block reconstructed by the adder 850 is stored in the memory 860 and stored in block units or as reference pictures in picture units, so that the predictor 840 can be used to predict the next block.

As described above, the image encoding apparatus 100 and the image decoding apparatus 800 described above with reference to FIGS. 1 through 8 interpolate the reference picture in subpixel units to increase the resolution of the reference picture and the motion vector, thereby inter prediction encoding and inter prediction decoding. The reference picture may be interpolated at the same resolution in units of pictures or in groups of pictures (GOPs) and the resolution of the motion vector may be increased.

However, in this case, the prediction is more accurate by increasing the resolution of the reference picture and the bit amount generated by encoding the residual signal can be reduced accordingly. However, the resolution of the motion vector is also increased to encode the motion vector. Since the amount of bits to be collectively increased, as a result, even in the case of performing inter prediction encoding by increasing the resolution of the reference picture, the encoding efficiency may not be greatly increased or the encoding efficiency may be lowered depending on the image.

Hereinafter, inter prediction encoding is performed by adaptively increasing the resolution of a reference picture in an area unit having an arbitrary size, such as a picture unit, a slice unit, a block unit, etc. according to characteristics of an image, thereby increasing the resolution or complexity of an image. A method and apparatus for inter prediction encoding and decoding by increasing the resolution in a small region and lowering the resolution in a region where the image is not complicated or large in motion are described.

9 is a block diagram schematically illustrating a video encoding apparatus according to a first embodiment of the present invention.

The image encoding apparatus 900 using the adaptive motion vector according to the first embodiment of the present invention includes an inter prediction encoder (910), a resolution change flag encoder (920), and a resolution determiner ( Resolution Determiner (930), a resolution encoder (Resolution Encoder, 940) and a differential vector encoder (Differential Vector Encoder, 950). The resolution change flag generator 920, the resolution encoder 940, and the difference vector encoder 950 are not necessarily all included in the image encoding apparatus 900, but may be selectively included according to an implementation method.

The inter prediction encoder 910 inter-prediction encodes an image in units of regions using a motion vector according to a motion vector resolution determined for each region or motion vector of the image. The inter prediction encoder 910 may be implemented by the image encoding apparatus 100 described above with reference to FIG. 1. Here, when one or more components of the resolution encoder 940 and the differential vector encoder 950 of FIG. 9 are additionally included, the functions of the additional components and the encoder 150 in the inter prediction encoder 910 are included. In case of overlapping, the overlapping function may be omitted in the encoder 150. In addition, when there is a part where the function of the predictor 110 and the function of the resolution determiner 930 overlap in the inter prediction encoder 910, the overlapping function may be omitted in the predictor 110.

In addition, one or more components added among the resolution encoder 940 and the differential vector encoder 950 may be configured separately from the inter prediction encoder 910 as shown in FIG. 9, but may be configured as an encoder in the inter prediction encoder 910. It may be configured integrally with 150. In addition, although the flag information generated by the resolution change flag generator 920 may be converted into a bitstream by the resolution change flag generator 920, the function of converting the bitstream into a bitstream is performed by the encoder 150 of the inter prediction encoder 910. It may be.

However, the image encoding apparatus 100 described above with reference to FIG. 1 has been described as encoding an image in units of blocks. However, the inter prediction encoder 910 uses a variety of blocks such as macroblocks or subblocks, slices, pictures, and the like. It may be divided into regions of shape and size and encoded in an area unit having an arbitrary size. The predetermined area may be a macroblock of 16 × 16 size, but may be a block of various sizes and shapes such as a 64 × 64 size block and a 32 × 16 size block.

In addition, while the image encoding apparatus 100 described above with reference to FIG. 1 inter-predictively encodes motion blocks having the same motion vector resolution for all blocks of an image, the inter prediction encoder 910 is a motion vector resolution determined differently for each region. Each region may be inter prediction encoded by using a motion vector having. In this case, an image region in which the motion vector resolution may be determined may be a picture (frame or field), a slice, or an image block having a predetermined size.

That is, when inter prediction encoding a predetermined region, the inter prediction encoder 910 increases the resolution by interpolating a reference picture that is already encoded, decoded, and reconstructed, and then performs motion estimation. As an interpolation filter for interpolating a reference picture, various filters such as a Winner filter, a bilinear filter, a Kalman filter, and the like can be used. The resolution for interpolation is 2/1 pixel. The resolution of various fractional and integer pixel units such as 1/1 pixel, 1/2 pixel, 1/4 pixel, 1/8 pixel unit, etc. may be applied. In addition, different filter coefficients may be used or the number of filter coefficients may be changed according to the various resolutions. For example, when the resolution is 1/2 pixel unit, interpolation may be performed using a Wiener filter, and when the resolution is 1/4 pixel unit, interpolation may be performed using a Kalman filter. In addition, the number of filter taps may be changed when interpolating each resolution. For example, when the resolution is 1/2 pixel unit, the interpolation may be performed using an 8 tap winner filter, and when the resolution is 1/4 pixel unit, the interpolation may be performed using a 6 tap winner filter.

In addition, the inter prediction encoder 910 may encode the filter coefficients by determining the optimum filter coefficient for each motion vector resolution that minimizes the error between the current picture to be encoded and the reference picture. In this case, the filter may use any filter such as a Wiener filter or a Kalman filter, and the number of filter taps may be several, and the number of filters and filter taps may be different for each resolution.

In addition, the inter prediction encoder 910 may inter predict using a reference picture interpolated using a different filter according to the resolution of a region or a motion vector. For example, in order to calculate an optimal filter that minimizes a sum of squared difference (SSD) between a current picture to be encoded and a reference picture, as shown in Equation 2, an optimal filter tap is selected for each resolution using the Wiener filter. Can be calculated

Equation 2

In Equation 2, S denotes a pixel of the current picture, h ^sp denotes a filter coefficient of the pixel domain, P denotes a pixel of the reference picture, e ^SP denotes an error, and x and y denote a position of the current pixel. Indicates.

That is, the inter prediction encoder 910 calculates filter coefficients for each resolution by using a Wiener hop equation as shown in Equation 2, encodes the optimal filter coefficients for each resolution, and includes them in the bitstream, and performs interpolation filtering on the reference picture. After the operation, reference pictures for each resolution may be generated and encoded. At this time, the filter coefficient of the 6-tap Winner filter at 1/2 resolution, the 8-tap Kalman filter at 1/4 resolution, the filter coefficient of the linear filter at 1/8 resolution are calculated, and the filter coefficients are encoded and included in the bitstream. The reference picture for each resolution can be interpolated and encoded. In this case, the inter prediction encoder 910 encodes using a reference picture interpolated by a 6-tap winner filter when the resolution of the current region or motion vector is 1/2 resolution, and interpolates by an 8-tap Kalman filter when 1/4 resolution. The encoded reference picture may be used.

The resolution change flag generator 920 may generate and include a resolution change flag indicating whether the motion vector resolution and / or the resolution of the differential motion vector is determined for each region or motion vector of the image. The area for changing the motion vector resolution and / or the resolution of the differential motion vector by the resolution change flag may be a block, a macroblock, a group grouping blocks, a group grouping macroblocks, or an area having an arbitrary size such as MxN. have. That is, the resolution change flag generator 920 inter-predictively encodes a motion vector having a fixed motion vector resolution for all regions or sub-regions of an image or motion vector resolution for each region (or motion vector). The inter prediction encoding may be performed using a motion vector having a determined motion vector resolution, and a resolution change flag indicating whether to generate a differential motion vector having a fixed resolution may be generated and included in the bitstream. The resolution change flag may be determined and generated according to setting information input by a user, or may be determined and generated according to a predetermined criterion by analyzing an image to be encoded. Such a resolution change flag may be included in a header of the bitstream. The header may be a picture parameter set, a sequence parameter set, a slice header, or the like.

If the resolution change flag generated by the resolution change flag generator 920 indicates that the motion vector resolution and / or the resolution of the differential motion vector is fixed, the inter prediction encoder 910 is fixed for the subregion defined in the header. Inter-prediction encoding of each region is performed by using the motion vector of each region having a set motion vector resolution. For example, if the resolution change flag included in a slice header of a slice indicates that the motion vector resolution is fixed, the motion vector resolution having the least rate-distortion cost is determined for the image of the slice and all regions of the slice are determined. For, the inter prediction encoding may be performed using a motion vector of each region having a corresponding motion vector resolution.

In addition, when the resolution change flag indicates that the motion vector resolution and / or the resolution of the differential motion vector is adaptively changed for each region or motion vector, the inter prediction encoder 910 may determine each region determined by the resolution determiner 930. Each region is inter prediction encoded by using the motion vector of each region having a motion vector resolution of. For example, if the resolution change flag included in the slice header of a slice indicates that the motion vector resolution and / or the resolution of the differential motion vector is adaptively changed for each region or motion vector, the slice is determined by the resolution determiner 910. An area of each slice may be inter prediction encoded by using a motion vector of each area having a motion vector resolution determined for each area within. As another example, when the resolution change flag included in the slice header of a slice indicates that the motion vector resolution and / or the resolution of the differential motion vector are adaptively changed for each motion vector, the slice is determined by the resolution determiner 930. Inter prediction encoding may be performed using a motion vector resolution determined for each motion vector in the frame.

When the resolution determiner 930 generates a flag indicating that the resolution of the motion vector resolution and / or the differential motion vector is adaptively changed for each area or motion vector by the resolution change flag generator 920, the resolution determiner 930 generates a unit of motion or unit of motion vector. Optimum motion vector resolution and / or differential motion vector resolution using some cost function, such as RD Cost (Rate-Distortion Cost) while changing the resolution of the motion vector resolution and / or differential motion vector. Determine. Here, the optimal motion vector resolution and / or the resolution of the differential motion vector only represent the motion vector resolution and / or the resolution of the differential motion vector determined using a predetermined cost function, and the motion vector resolution and / or are determined in this way. Or does not mean that the resolution of the differential motion vector always has optimal performance. When the predetermined cost function is the rate-distortion cost, the motion vector resolution and / or the resolution of the differential motion vector, where the rate-distortion cost is minimal, may be the optimal motion vector resolution and / or the resolution of the differential motion vector.

The resolution encoder 940 may encode a motion vector resolution and / or a resolution of a differential motion vector determined for each region or motion vector. That is, the resolution encoder 940 encodes a motion vector resolution identification flag indicating the motion vector resolution of each region determined by the resolution determiner 930 and / or a resolution identification flag of the differential motion vector indicating the resolution of the differential motion vector. It can be included in the bitstream. The implementation of the motion vector resolution identification flag and / or the resolution identification flag of the differential motion vector is possible in a variety of ways. Here, the resolution indicated by the resolution identification flag may be the resolution of the motion vector resolution or the differential motion vector, or may be used simultaneously for the motion vector resolution and the resolution of the differential motion vector.

The differential vector encoder 950 may encode a differential motion vector that is a difference between a motion vector and a predicted motion vector according to a motion vector resolution determined for each region or motion vector. Here, the differential motion vector may be encoded according to the resolution of the differential motion vector.

The resolution identification flag indicating the motion vector resolution may indicate both the x component and the y component of the motion vector or may indicate respective resolutions. That is, when the camera photographing the image moves or an object in the image moves, the resolution determiner 930 may determine different resolutions of the x component and the y component of the motion vector for motion estimation. For example, the resolution determiner may determine resolution of 1/8 pixel unit for the x component of a motion vector of a certain area and resolution of 1/2 pixel unit for the y component. The 910 may inter-prediction the corresponding region by determining a motion vector with different resolutions for each of the x component and the y component for the motion vector of the corresponding region, and using the determined motion vector to perform motion estimation and motion compensation.

FIG. 22 is a diagram illustrating an optimal motion vector of a neighboring block and a current block in order to explain a process of determining the resolution of a motion vector in the resolution determiner 930.

When a flag indicating that the resolution of the motion vector resolution and / or the differential motion vector is changed by the resolution changing flag generator 920 adaptively for each region or motion vector (in the second embodiment, the resolution designation flag generator 3220) Can be set whether or not the motion vector resolution and / or the resolution of the differential motion vector are changed or fixed by the resolution designation flag generated by the controller), and the types of resolutions of the current block and neighboring blocks are 1/2, 1/4, Assume that 1/8 and the optimal resolution is determined as shown in FIG. In this case, the resolution of block A is 1/2 and the motion vector is (4/2, -8/2), the resolution of block B is 1/4 and the motion vector is (36/4, -28/4), The resolution of block C is 1/8 and the motion vector is (136/8, -104/8), the current block X is 1/4 and the motion vector is (16/4, 20/4). In this case, the predicted motion vector may depend on the resolution of the current motion vector. In this case, a resolution conversion process may be performed such that the resolution of the neighboring motion vector is the same as the resolution of the current motion vector to calculate the predicted motion vector.

FIG. 23 is a diagram illustrating a conversion equation according to a motion vector resolution, and FIG. 24 is a diagram illustrating a table obtained by converting a resolution of a motion vector of a neighboring block to a resolution of a block X to be currently encoded.

The predicted motion vector can be obtained using the neighboring motion vectors. If the neighboring motion vectors are stored according to their resolutions, they can be converted using multiplication and division when the neighboring motion vectors are different from the current motion vectors. . In this case, the resolution conversion process may be performed when a prediction motion vector is obtained. Alternatively, if the neighboring motion vector is stored based on the highest resolution, the current motion vector may be converted using division if the current motion vector is not the highest resolution. In this case, the resolution conversion process may perform a resolution conversion process of converting the encoded motion vector to the highest resolution if it is not the highest resolution. Alternatively, if the peripheral motion vector is stored based on an arbitrary resolution, the current motion vector may be converted using multiplication and division when the current motion vector is different from the reference resolution in which the peripheral motion vectors are stored. In this case, the resolution conversion process may perform a resolution conversion process of converting the motion vector to the reference resolution when the motion vector resolution is not the reference resolution when the motion vector is stored. In the case of division, rounding, rounding, and rounding can be performed. As shown in Figs. 23 and 24, rounding is used in the example. In addition, the example uses a peripheral motion vector stored according to each resolution.

A predicted motion vector can be obtained with reference to the table of FIG. 23. Here, the predicted motion vector can be obtained using median, and median can be obtained for each component.

MVPx = median (16/4, 36/4, 32/4) = 32/4

MVPy = median (-32/4, -28/4, -28/4) = -28/4

Thus, the predicted motion vector is (32/4, -28/7). The differential motion vector is obtained using the predicted motion vector thus obtained. The differential motion vector can be obtained by the difference between the motion vector and the predicted motion vector as shown in Equation (3).

Equation 3

Therefore, the differential motion vector is (-16/4, 48/4) = (-4, 12).

25 is a diagram illustrating a code number table of differential motion vectors according to motion vector resolution.

When the differential vector encoder 950 encodes the differential motion vector, the differential motion vector may be encoded for each motion vector value by using the code number table of the differential motion vector according to the motion vector resolution as shown in FIG. 25.

In addition, the prediction motion vector may be obtained as follows using the example of FIG. 22. Here, the median for each component can be obtained without converting the surrounding motion vector according to the current resolution.

MVPx = median (4/2, 36/4, 136/8) = 36/4

MVPy = median (-8/2, -28/4, -104/8) = -104/8

Thus, the predicted motion vector is (36/4, -104/8). The differential motion vector is obtained using the predicted motion vector thus obtained. The differential motion vector may be obtained as a difference between the motion vector and the predicted motion vector as shown in Equation 4.

Equation 4

Thus, the differential motion vector is (-20/4, 72/4) = (-5, 18).

When the differential vector encoder 950 encodes the differential motion vector, the differential motion vector may be encoded with respect to the motion vector value for each resolution by using a code number table of the differential motion vector according to the motion vector resolution as shown in FIG. 25.

In addition, the prediction motion vector may be obtained as follows using the example of FIG. 22. Here, instead of converting the surrounding motion vector according to the current resolution, the median may be obtained for each component and then converted according to the current resolution.

MVPx = median (4/2, 36/4, 136/8) = 36/4

MVPy = median (-8/2, -28/4, -104/8) = -104/8

Therefore, the predicted motion vector becomes (36/4, -52/4) with reference to FIG. The differential motion vector is obtained using the predicted motion vector thus obtained. The differential motion vector may be obtained as a difference between the motion vector and the predicted motion vector as shown in Equation 5.

Equation 5

Thus, the differential motion vector is (-20/4, 72/4) = (-5, 18).

In addition, the prediction motion vector may be obtained as follows using the example of FIG. 22. The median may also be obtained by using only motion vectors around the same resolution as the current motion vector. In FIG. 22, since only the neighboring motion vector having the same resolution as the current motion vector is block B, the predicted motion vector becomes (36/4, -28/4). The differential motion vector is obtained using the predicted motion vector thus obtained. The differential motion vector may be obtained as a difference between the motion vector and the predicted motion vector as shown in Equation 6.

Equation 6

Thus, the differential motion vector is (-20/4, 48/4) = (-5, 12).

In addition, when the neighboring motion vector is stored based on the resolution of 1/8, the predicted motion vector may be obtained as follows using the example of FIG. 22. Referring to FIGS. 23 and 24, the predicted motion vectors are (32/4, -28/4). The differential motion vector is obtained using the predicted motion vector thus obtained. The differential motion vector can be obtained by the difference between the motion vector and the predicted motion vector as shown in Equation (3). Therefore, the differential motion vector is (-16/4, 48/4) = (-4, 12).

Meanwhile, the resolution encoder 940 may encode the resolution change flag and the type of resolution in the header (meaning the resolution designation flag in the second embodiment), and at this time, the optimally determined resolution identification flag is encoded by 1/4. The differential vector encoder 950 may encode the differential motion vector obtained by using the predicted motion vector calculated using the peripheral motion vector converted according to the resolution determined by the resolution determiner 930.

FIG. 26 is a diagram illustrating an optimal motion vector of a neighboring block and a current block in order to explain a process of determining a resolution of a differential motion vector in the resolution determiner 930.

As shown in FIG. 26, when the motion vector resolutions of the neighboring block and the current block are 1/8, the predicted motion vector may be calculated as shown in Equation 7 below.

Equation 7

Thus PMV = (2/8, -2/8) = (1/4, -1/4). The differential motion vector can be obtained as shown in Equation (8).

Equation 8

Therefore, the MVDx can be encoded with 1/8 and the MVDy can be encoded with 1/4.

FIG. 27 is a diagram illustrating a differential motion vector code number table according to differential motion vector resolution.

As shown in Fig. 27, the code numbers of the differential motion vectors become (1, 1) according to the code number table of the differential motion vectors. Therefore, the resolution encoder 940 encodes the differential motion vector resolution identification flag by (1/8, 1/4) for each of the x and y components, encodes the code number of the differential motion vector as (1, 1), and performs the differential motion. Signs can be separately encoded for x and y components of the vector.

On the other hand, when the differential vector encoder 950 encodes a differential motion vector, a reference resolution is determined, and a motion vector having a resolution other than the reference resolution is converted into a motion vector having a reference resolution and obtained from a reference motion vector of a neighboring block. The differential motion vector is calculated using the predictive motion vector. If the motion vector has a resolution other than the reference resolution, a method of additionally encoding the reference resolution flag may be used. The reference resolution flag may include data indicating whether the resolution is the same as the same reference resolution and data indicating the position of the actual motion vector.

The reference resolution may be defined in a header, and may be a picture parameter set, a sequence parameter set, a slice header, or the like.

28 is a diagram illustrating a motion vector of a current block X and a reference motion vector of a neighboring block.

Indicates that the resolution change flag (meaning the resolution designation flag in the second embodiment) is not a single resolution, the resolution type is 1/2, 1/4, 1/8, and the reference resolution is 1/4, as shown in FIG. When the optimal resolution is determined, the motion vectors 4/8 and 5/8 of the current block are converted into Equation 9 by the reference motion vector Ref_MV using the reference resolution 1/4.

Equation 9

If the resolution of the motion vector of the current block is different from the resolution of the reference motion vector, it may be converted using multiplication and division. In the case of division, rounding, rounding, rounding, etc. can be used, and rounding is used in this embodiment. Accordingly, the reference motion vector becomes (2/4, 3/4), and the position of the actual motion vector having a resolution other than the reference resolution may be represented by the reference resolution flag. Here, the difference between the motion vector of the current block and the reference motion vector is (0, 1/8), and the value of the reference resolution flag may include location information such as (0, 1), for example. In the illustrated position information (0, 1), 0 indicates that the reference motion vector is the same as the motion vector, and 1 indicates the motion vector as small as -1/8 in the reference motion vector.

Meanwhile, the difference vector of the reference motion vector is calculated using the prediction reference motion vector, which is a median value of the reference motion vector of the neighboring block. The prediction reference motion vector can be obtained by using Equation 10.

Equation 10

Therefore, when the prediction reference motion vector (Ref_PMV) is (1, -1) and the differential reference motion vector (Ref_MVD) is (Ref_MV (2/4, 3/4)-Ref_PMV (1, -1)), 2/4, 7/4). Therefore, the encoder encodes Ref_MVD (-2/4, 7/4) and encodes a reference resolution flag (0, 1) indicating the position of the motion vector.

FIG. 29 is a diagram illustrating a code number table of a differential reference motion vector according to differential reference motion vector resolution.

Therefore, referring to FIG. 29, when the reference resolution is 1/4, the value of the differential reference motion vector is 2/4, and the code number is 2, and the value of the differential reference motion vector is 3/4 as in the present embodiment. The code number is 3, and the code number for each component of the differential reference motion vector is included in the reference resolution flag.

The resolution encoder 940 may variously encode a motion vector resolution and / or a resolution of a differential motion vector determined for each region or motion vector. Hereinafter, various examples of encoding the motion vector resolution or the resolution of the differential motion vector will be described with reference to FIGS. 10 to 14. In the following description, only an example of encoding a motion vector resolution will be described. However, since the resolution of the differential motion vector may be encoded in the same manner as the motion vector resolution, an example of encoding the resolution of the differential motion vector is omitted.

The resolution encoder 940 may integrate the motion vector resolution determined for each region or motion vector between adjacent regions having the same motion vector resolution and generate a resolution identification flag in units of the integrated region. For example, the resolution identification flag may be generated hierarchically in a quadtree structure. In this case, the resolution encoder 940 may encode an identifier indicating the maximum number of quad tree layers and the area size indicated by the lowest node of the quad tree layer in the header for the corresponding area of the bitstream.

10 is an exemplary diagram hierarchically expressing a motion vector resolution in a quad tree structure according to an embodiment of the present invention.

10A shows regions having various motion vector resolutions in one picture. Each region in 10A may be macroblocks of size 16 × 16, and the number indicated in each region represents the motion vector resolution of each region. 10B is a group showing regions having the same motion vector resolution among the regions shown in 10A. 10C is a hierarchical representation of the motion vector resolution of regions bounded by 10B in a quad tree structure. In 10C, since the size of the region indicated by the lowest node is a 16 × 16 macroblock and the maximum number of hierarchies in the quad tree is 4, such information is encoded and included in the header for the region.

FIG. 11 is an exemplary diagram illustrating a result of encoding a motion vector resolution expressed hierarchically in a quad tree structure according to an embodiment of the present invention. FIG.

By encoding the motion vector resolution of the quadree structure shown in 10C, the final bit as shown in Fig. 11 can be obtained. Encodes whether a node is divided into one bit. For example, if the bit value is '1', the node is divided into lower layers. If the bit value is '0', the node is not split into lower layers. It is not shown.

At 10C, the node at level 0 is divided into lower layers, so that it is encoded with bit value '1'. Since the first node of the divided level 1 is no longer divided into 1/2 resolutions, the first node is encoded with a bit value '0' and the motion vector resolution 1/2 is encoded. Since the second node of level 1 is divided into lower layers, it is encoded with bit value '1'. Since the third node of level 1 is not divided into lower layers, the third node is encoded with bit value '0' and the motion vector resolution 1/4 is encoded. Since the last fourth node of level 1 has been divided into lower layers, it is encoded with bit value '1'. In the same way, each node of level 2 is encoded. Since the maximum number of layers is designated as 3 in the header at level 3, since there are no lower layers, only motion vector resolution is encoded. Thus, the final bit hierarchically encoded with the quadtree structure of the motion vector resolution for the region shown in 10A may be generated as shown in FIG. 11.

Motion vector resolutions such as 1/2, 1/4, and 1/8 shown in the last bit mean that they are encoded with bits representing them, but are not represented as bit values for convenience of description. The motion vector resolution may be expressed as a bit value in various ways depending on the implementation manner. For example, if the available motion vector resolution is two, it may be indicated by a 1-bit flag. If the available motion vector resolution is four or less, it may be indicated by a 2-bit flag.

If the number of maximum layers and the size of the region indicated by the lowest node are specified in the slice header, the resolution identification flag generated as described above may be included in the field of the slice data. The image decoding apparatus to be described later may extract and decode the resolution identification flag from the bitstream to restore the motion vector resolution of each region.

In addition, in the example illustrated in FIG. 10, only two examples of the node being divided into lower layers (that is, divided into four zones) or not divided are illustrated. However, as shown in FIG. 20, the lower layer is illustrated. Nodes can be subdivided into lower layers in various ways, such as not divided into two regions, divided into two horizontally long regions, divided into two vertically long regions, and divided into four regions.

In addition, the resolution encoder 940 may generate a resolution identification flag by encoding the motion vector resolution determined for each region or motion vector using the motion vector resolution predicted using the motion vector resolution of the peripheral region of each region. For example, assuming that an area is a block of size 64 × 64, the motion vector resolution of the corresponding area may be estimated using the motion vector resolutions of the left area and the upper area of the area. In this case, when the predicted motion vector resolution is the same as the motion vector resolution of the corresponding region, the resolution encoder 940 encodes the resolution identification flag for the corresponding region as a bit value '1'. When the predicted motion vector resolution and the motion vector resolution of the corresponding area are not the same, the resolution identification flag for the corresponding area may be encoded into a bit value '0' and a bit value representing the motion vector resolution of the corresponding area. For example, if the resolution of the upper region and the left region of the region is 1/2 resolution and the resolution of the region is also 1/2 resolution, the resolution identification flag for the region is encoded with the bit value '1' and the region If the resolution of the motion vector is not encoded and the resolution of the upper and left regions of the region is 1/2 resolution and the resolution of the region is 1/4 resolution, the resolution identification flag for the region is set to the bit value '0'. Can be encoded and additionally encoded the motion vector resolution of the region.

In addition, the resolution encoder 940 may generate a resolution identification flag by encoding the motion vector resolution determined for each region or motion vector by using the run and length of the motion vector resolution of each region or motion vector. have.

In FIG. 12, regions in one picture are represented by macroblocks having a size of 16 × 16, and motion vector resolutions of the regions are illustrated in each region. For example, the motion vector resolution of the regions shown in FIG. 12 is encoded using a run and a length. For example, when the motion vector resolutions of the regions shown in FIG. 12 are arranged in the raster scan direction, a motion vector resolution of 1/2 is obtained. 4 times in succession, 1/4 motion vector resolution comes out 1 time, 1/8 motion vector resolution comes out 2 times in succession, 1/2 motion vector resolution comes out 4 times in succession (after motion Vector resolution is omitted), when the motion vector resolution of each region is expressed by the run and length, (½, 4), (¼, 1), (⅛, 2), (½, 4),... It can be expressed as Accordingly, the resolution encoder 940 may generate a resolution identification flag by encoding the motion vector resolution of each region represented by the run and the length by representing the bit value.

In addition, the resolution encoder 940 may generate a resolution identification flag by hierarchically encoding a motion vector resolution determined for each region or motion vector using a tag tree. In this case, the resolution encoder 940 may include an identifier indicating the maximum number of hierarchies of the tag tree and the size of the region indicated by the lowest node in the header.

13 is an exemplary diagram hierarchically expressing a motion vector resolution in a tag tree structure according to an embodiment of the present invention.

FIG. 13 hierarchically shows a motion vector resolution determined for each region within a portion of an image in a tag tree structure. Here, it is assumed that each region is a macroblock of size 16 × 16.

In FIG. 13, since the minimum value of the motion vector resolutions of the first four regions of level 3 is 1/2, the motion vector resolution of the first region of level 2 is 1/2. In this way, regions are hierarchically grouped by the number of layers, and then code bits are generated and encoded from the upper layer to the lower layer.

14 is an exemplary diagram illustrating a result of encoding a motion vector resolution expressed hierarchically expressed in a tag tree structure according to an embodiment of the present invention.

In the method for generating the sign bit, the difference between the number of motion vector resolutions of the upper layer and the current layer is expressed as a sequence of '0', and the bit value of the last bit is expressed as '1'. In the case of the uppermost layer, the motion vector resolution number of the upper layer is '0', the motion vector resolution number of 1/2 is '1', and the motion vector resolution number of 1/4 is '2', 1 / Assume that the number of motion vector resolutions of 8 is '3', and if the motion vector resolutions of the regions shown in FIG. 13 are hierarchically encoded in a tag tree structure, a resolution identification flag may be generated as shown in FIG. 14. Here, the number assigned to each motion vector resolution may be changed.

In FIG. 14, (0,0), (0,1), etc., expressed in each region are reference marks for identifying each region, and '0111', '01', etc., are encoded with motion vector resolution of each region. Indicates the bit value of the resolution identification flag.

In the case of the resolution identification flag of the region identified by (0,0), the motion vector resolution number of the upper layer at level 0 is '0' and the motion vector resolution number at 1/2 is number '1'. Therefore, the difference between the number of level 1 and level 0 is '1', and when '1' is converted into coded bits, it is '01'. In addition, since the difference value of the number of motion vector resolutions from the level 1 to the higher layer (level 0) becomes '0', the encoding bit becomes '1'. In addition, since the difference value of the number of the motion vector resolution with respect to the upper layer (level 1) at level 2 becomes '0', the encoding bit becomes '1'. Again, at level 3, the difference value of the number of the motion vector resolution with the higher layer (level 2) becomes '0', so the encoding bit becomes '1', and finally the motion vector resolution of the region identified by (0,0). The coded bit of becomes '0111'.

Looking at the case of the resolution identification flag of the area identified by (0,1), level 0, level 1, and level 2 are already reflected in the resolution identification flag of the area identified by (0,0), so they are higher in level 3. '1', which is a difference value of the number of motion vector resolutions with respect to the layer (level 2), may be encoded, and the corresponding encoding bit becomes '01'. Therefore, in the region identified by (0, 1), only '01' is generated as the resolution identification flag of the region.

Looking at the case of the resolution identification flag of the region identified by (0,4), level 0 is already reflected in the resolution identification flag of the region identified by (0,0), so that the level 1, By encoding only level 2 and level 3, the coded bit becomes '0111'.

In addition, the resolution encoder 940 may generate a resolution identification flag by changing and encoding the number of bits allocated to the motion vector resolution according to the frequency of the motion vector resolution determined for each region or motion vector. To this end, the resolution encoder 940 may change and encode the number of bits allocated to the motion vector resolution of the corresponding area according to the frequency of occurrence of the motion vector resolution up to the immediately preceding area in units of areas, or include a plurality of areas. The number of bits allocated to the motion vector resolution of the partial region may be encoded according to the frequency of occurrence of the motion vector resolution of the immediately preceding partial region or the frequency of occurrence of the motion vector resolution to the immediately preceding partial region on an area basis. To this end, the resolution encoder 940 calculates the frequency of the motion vector resolution in units of regions or partial regions, assigns small numbers in order of increasing frequency of the calculated motion vector resolution, and assigns the small numbers of motion vector resolutions. A small number of bits can be allocated to encode the motion vector resolution of each region.

For example, when the resolution encoder 940 changes the number of bits according to the frequency of occurrence of the motion vector resolution up to the previous region in the area unit, the motion vector resolution of 1/2 is 10 times in all the regions up to the previous region. Is generated, 1/4 motion vector resolution is generated 15 times, and 1/8 motion vector resolution is generated 8 times, the smallest number (e.g., 1) is assigned to 1/4 motion vector resolution. Assign the next smaller number (e.g., 2) to 1/2 motion vector resolution, give the largest number (e.g. 3) to 1/8 motion vector resolution, After allocating small bits in numerical order, if the motion vector resolution of the region to be encoded with motion vector resolution is 1/4 pixel unit, the smallest bit is allocated to encode 1/4 motion vector resolution for the region. Can be.

In addition, when the resolution encoder 940 changes the number of bits according to the frequency of occurrence of the motion vector resolution of the previous region group for each region group unit, the resolution encoder 940 generates the motion vector resolution of each region of the previous region group. The frequency vector may be updated to allocate small numbers in order of occurrence frequency and to allocate small bits in order of number order to encode motion vector resolution of each region of the region group to be encoded. An area group may be a quad tree, a quad tree bundle, a tag tree, a tag tree bundle, a macroblock, a macroblock bundle, and an area of any size. For example, if two macroblocks are designated as two macroblocks, the frequency of occurrence of the motion vector resolution is updated by every two macroblocks, and the number of bits of the motion vector resolution is assigned, or if four region groups are designated as quadtrees. The frequency of occurrence of the motion vector resolution is updated every four to allocate the number of bits of the motion vector resolution.

In addition, the resolution encoder 940 may change the encoding method of the resolution identification flag according to the distribution of the motion vector resolutions of the surrounding areas of each region based on the motion vector resolution determined for each region or motion vector. That is, the shortest number of bits is allocated to the resolution having the highest probability of being the resolution of the corresponding area according to the distribution of the motion vector resolution of the surrounding area or the area group. For example, if the motion vector resolution of the left region of the region is 1/2 and the motion vector resolution of the upper region is 1/2, the motion vector resolution of the region is likely to be 1/2, so 1 / The shortest number of bits is allocated to two motion vector resolutions and encoded. As another example, the motion vector resolution of the left region of the region is 1/4, the motion vector resolution of the upper region is 1/2, the motion vector resolution of the upper region is 1/2 and the motion vector resolution of the upper right region is 1 In the case of / 2, the number of short bits is allocated and encoded in the order of the highest probability, such as the order of 1/2, 1/4, 1/8, etc., of the motion vector resolution of the corresponding region.

Further, when entropy encoding by arithmetic coding, the resolution encoder 940 generates a binary coded bit string of the resolution identification flag based on the distribution of the motion vector resolution determined for each region or motion vector according to the distribution of the motion vector resolution of the surrounding region of each region. In addition, the arithmetic coding and the probability update are performed by applying the probability model differently according to the distribution of the motion vector resolution of the surrounding area and the probability of the motion vector resolution which has occurred up to now. In addition, arithmetic coding and probability updating are performed using different probability models according to bit positions.

Referring to FIG. 21, for example, assuming that only three motion vector resolutions, such as 1/2, 1/4, and 1/8, are used for entropy encoding by CABAC and the encoding is used. If the motion vector resolution of the left region is 1/2 and the motion vector resolution of the upper region of the region is 1/2, then assign the shortest binarized bit string to 1/2 motion vector resolution, and match the remaining 1/4 motion vector resolution. The short binary bits are allocated to the 1/8 motion vector resolution in the order of the highest probability of occurrence. At this time, when the probability that the 1/8 motion vector resolution has occurred until now is higher than the 1/4 motion vector resolution, the bit string having '00' is allocated to the 1/8 motion vector resolution as shown in FIG. 21, and the bit having '01'. Arithmetic coding is performed by assigning columns to half motion vector resolution.

In addition, when encoding the first binarization bit, the probability model may be divided into four probability models and encoded. The four probabilistic models are, firstly, a probability model in which both the left and top regions have the same resolution and are the same as the highest probability resolution so far; second, the left and the top regions all have the same resolution and the most Probability model when the resolution of the high probability differs from the third. Third, the probability model when the resolution of the left and upper regions are all different, and the resolution of the highest probability so far and the resolution of the left or upper region is the fourth. In this case, the resolution of the left region and the upper region and the highest probability resolution so far can be divided into probability models. When the second binarization bit is encoded, the probability model may be divided into two probability models and encoded. The two probabilistic models are the first when the motion vector resolutions of both the left and top regions of the corresponding region are different and have the highest probability motion vector resolution so far and the motion vector resolution such as the motion vector resolution of the left or upper region. Probability model can be classified into a probability model in which the motion vector resolution of the left region and the upper region of the corresponding region and the resolution of the highest probability so far are different.

In another example, the CABAC is used for encoding, and the motion vector resolution used for encoding uses only three motion vector resolutions, such as 1/2, 1/4, and 1/8, and has the highest probability of occurrence. If the motion vector resolution is 1/4, assign the shortest binarization bit string '1' to 1/4 motion vector resolution, and set '00' to the remaining 1/2 motion vector resolution and 1/8 motion vector resolution, respectively. Assign '01'. In addition, when encoding the first binarization bit, a probability model may be divided into three probability models and encoded. The three probabilistic models are: first, a probability model where the motion vector resolution of the left region and the upper region of the region is the same as the highest resolution that has occurred so far; and second, only one of the resolutions of the left region and the upper region of the region. Probability model in which the probability that occurred so far is the same as the highest resolution, and third, the probability model when the motion vector resolution of the left and upper regions of the corresponding region differs from the motion vector resolution of the highest probability so far. Can be. When encoding the second binarization bit, the probability model may be divided into six probability models and encoded. The six probabilistic models are, firstly, probabilistic models when the resolution of the left and top regions of the region is 1/8 motion vector resolution, and second, the motion vector resolution of the left and upper regions of the region is 1/1. Probabilistic model with 2 motion vector resolutions, Third, Probabilistic model with motion vector resolutions of the left and upper regions of the corresponding region being 1/4 motion vector resolution, and Fourth, resolution of the left and upper regions of the region Probability model when one is 1/8 motion vector resolution and the other is 1/4 motion vector resolution.Fifth, one of the motion vector resolutions of the left and upper regions of the region is 1/2 motion vector resolution and the other Probabilistic model for one motion of 1/4 motion vector resolution, and sixth for motion vector resolutions of the left and upper areas of the area. Im a vector resolution and the other may be separated by one-half if the probability model of the motion vector resolution. The resolution with the highest probability so far may be the probability of the resolution encoded up to the previous region, the probability of any region, or the predetermined fixed resolution.

In addition, the resolution encoder 940 determines whether the motion vector resolution determined for each region or motion vector can be estimated according to a pre-determined estimation method in the image decoding apparatus, and thus, for the region where the motion vector resolution can be estimated. A resolution identification flag may be generated by encoding an identifier indicating that it is possible to estimate, and a resolution identification flag may be generated by encoding an identifier indicating that the motion vector resolution cannot be estimated and a motion vector resolution of the corresponding area. .

That is, when the encoder 940 wants to encode a motion vector resolution for each region or motion vector, the motion encoder calculates a motion vector and a predicted motion vector of the corresponding region by applying a plurality of motion vector resolutions, and encodes the difference vector. After decoding the difference vector, the reconstructed difference vector is assumed to be the optimum resolution, and the motion vector is decoded for each resolution. Assuming that each resolution is an optimal resolution, when the motion of the pixels around the region is compensated for by using the reconstructed motion vector, the motion vector resolution that minimizes the cost according to a predetermined cost function is determined. The resolution encoder 940 determines that the motion vector resolution thus determined is the motion vector resolution of the region to be originally encoded (the motion vector resolution determined as the optimal motion vector resolution of the region, except that the optimal motion vector resolution always provides the best performance. In the case where the resolution is equal to the motion vector resolution that is optimally determined under the conditions for determining the motion vector resolution, the identifier indicating that the motion vector resolution of the corresponding area can be estimated by the image decoding apparatus (eg, '1' ') May be generated as a resolution identification flag of the corresponding area, in which case the motion vector resolution of the corresponding area is not encoded. If the determined motion vector resolution is not the same as the motion vector resolution of the region to be originally encoded, the identifier (eg, '0') indicating that the motion vector resolution of the region cannot be estimated by the image decoding apparatus and the corresponding The original motion vector resolution of the region may be encoded and generated as a resolution identification flag of the region. Here, as the predetermined cost function, various distortion functions such as Mean Squre Error (MSE) or Sum of Absolute Transformed Differences (SATD) may be used.

In addition, when each component of the difference vector is '0', the resolution encoder 940 may not encode the resolution of the region or the motion vector. When each component of the difference vector is '0', the motion vector resolution does not need to be encoded because the predictive motion vector is encoded as a motion vector.

FIG. 15 is an exemplary diagram for describing a process of determining a motion vector resolution using pixels around a region according to an embodiment of the present invention.

Referring to FIG. 15, an optimal motion vector resolution determined by the motion encoder 940 performing motion estimation on a region to which motion vector resolution is to be encoded is a motion vector resolution of 1/2, and a motion vector is (4,10). If the predicted motion vector is (2,7), the difference vector is (2,3). At this time, the resolution encoder 940 assumes that the image decoding apparatus can decode and reconstruct only the difference vector, and changes the motion vector resolution to several to predict the predicted motion vector according to each motion vector resolution, thereby resolving each motion vector resolution. After reconstructing the motion vector according to the motion vector resolution, the motion vector resolution at which the distortion between the surrounding pixels of the current area and the surrounding pixels of the area indicated by the motion vector according to the reconstructed motion vector resolution is minimized is determined.

Once the motion vector resolution is 1/4 pixel unit, if the predictive motion vector is (3,14), the difference vector reconstructed by the image decoding apparatus is (2,3), and thus the motion vector of the corresponding region to be reconstructed is (5,17). ) In addition, when the motion vector resolution is 1/2 pixel unit and the predicted motion vector is (2,7), since the difference vector reconstructed by the image decoding apparatus is (2,3), the motion vector of the corresponding region to be reconstructed is (4, 10). In this manner, the motion vector of the corresponding region restored by the image decoding apparatus when the motion vector resolution is 1/8 pixel units is also calculated.

By using the motion vector of the corresponding region reconstructed according to the resolution of each motion vector, the resolution at which the distortion ratio between the surrounding pixels of the region compensating for the movement of the corresponding region in the reference picture and the surrounding pixels of the region is minimum When the resolution is equal to the optimal motion vector resolution determined in advance, the resolution encoder 940 encodes only an identifier indicating that the motion vector resolution can be estimated by the image decoding apparatus, and generates the resolution vector as a resolution identification flag of the corresponding region. The resolution is not encoded.

The resolution determiner 930 determines a motion vector resolution determined for each region or motion vector when the size of the predicted motion vector or the difference vector of the motion vector according to the motion vector resolution determined for each region or motion vector is larger than a threshold. Can be determined. For example, when the magnitude of the prediction motion vector or the difference vector of a certain region is larger than the threshold, the motion vector resolution of the region or motion vector is determined as a predetermined value or a random value. May not be encoded. In addition, when the size of the motion vector of the region or the surrounding vector of the motion vector is large or the size of the motion vector of the region or the motion vector is larger than the threshold, the motion vector resolution of the region is set to a predetermined value or a random value. In this case, the motion vector resolution of the corresponding region may not be encoded. In this case, the motion vector resolution of the region or motion vector may be changed to any resolution without any flag. The threshold may be a predetermined value or any value input and may be calculated from the motion vectors of the surrounding blocks.

The resolution determiner 930 may encode information about the resolution by encoding the reference picture index without generating a resolution identification flag when the resolution of the current block is known using the reference picture index.

For example, the reference picture may be indexed and encoded based on the distance between the current picture and the reference picture as shown in FIG. 30. For example, if four reference pictures are written, candidates of indexable reference pictures can be indexed as shown in FIG. 31 when the current picture is five times.

31 is a diagram illustrating a reference picture index according to a reference picture number and a resolution.

When the resolution used is 1/4 and 1/8, as shown in FIG. 31, if the optimal reference picture is 3 times and the resolution is 1/8, the reference picture index can be encoded with 3, and the decoding apparatus In FIG. 3, reference picture number 3 is extracted from a bitstream, and reference picture is 3 and resolution is 1/8 using the same table as the encoder.

The difference vector encoder 950 may encode the difference vector differently according to the motion vector resolution. In other words, as the motion vector resolution increases, the size of the motion vector increases, so that the amount of bits required for encoding the difference vector also increases. Accordingly, the amount of bits can be reduced by encoding the difference vector in different ways according to the size of the motion vector resolution.

For example, when the difference vector encoder 950 encodes the difference vector using UVLC, the difference vector encoder 950 may encode using the K-th order exponential golem code, and according to the motion vector resolution determined for each region, The order K can be changed. For example, when encoding a difference vector using UVLC, if the motion vector resolution is 1/4 pixel unit, the order K of the code may be set to '0', and if the motion vector resolution is 1/8 pixel unit, The order K of may be set to '1'.

In addition, when the difference vector encoder 950 encodes the difference vector using CABAC, the difference vector encoder 950 may encode using the unary / K-order exponential golem combined code, wherein the unary / The order (K) and the maximum value (T) of the K-order exponential golem combined code may be changed and encoded. For example, when encoding a differential vector using CABAC, if the motion vector resolution is 1/4 pixel unit, the order K of the code may be set to '3' and the maximum value T may be set to '6' and the motion If the vector resolution is 1/8 pixel units, the order K of the code may be set to '5' and the maximum value T may be set to '12'.

In addition, when the difference vector encoder 950 encodes the difference vector using CABAC, the cumulative probability may be differently calculated and encoded according to the motion vector resolution determined for each region. For example, each time the difference vector of the regions is encoded, the probability model is updated according to the motion vector resolution determined for each region, and when the difference vector of the other regions is encoded, the probability model according to the updated motion vector resolution is updated. It is available. That is, when the motion vector resolution of a region is 1/2 pixel unit, the difference vector is encoded using a probability model of 1/2 pixel unit, the probability model of 1/2 pixel unit is updated, and the motion vector of a region When the resolution is 1/8 pixel units, the differential vector may be encoded using a 1/8 pixel unit probability model and the probability model may be updated in 1/8 pixel unit.

In addition, the difference vector encoder 950 may predict the predicted motion vector for the motion vector of each region by using the motion vector of the surrounding region of each region, in order to calculate the difference vector of each region. When the vector resolution and the motion vector resolution of each region are not the same, the motion vector resolution of the surrounding region may be converted into the motion vector resolution of each region to be predicted. When converting to motion vector resolution, rounding, rounding, or rounding may be used. Here, the surrounding area should be understood as a concept including an adjacent area.

16 is an exemplary diagram for describing a process of predicting a predicted motion vector according to an embodiment of the present invention.

Referring to the example shown in FIG. 16, if the motion vectors of the surrounding area of the region to be predicted are (4,5), (10,7), (5,10) and rounding is used when converting the predicted motion vector, When the resolution of each motion vector is 1/2, 1/8, or 1/8, if the resolution of the motion vector of the region to be predicted is 1/4, the predicted motion vector may be (5,5). If the motion vector resolution of the desired region is 1/8, the predictive motion vector may be (10, 10).

In addition, when the block mode of one or more regions of each region is the skip mode (SKIP mode), the difference vector encoder 950 is an area to predict the motion vector at the highest resolution among the motion vector resolutions of the surrounding regions of the region. The motion vector resolution of can be converted and predicted. Referring to the example illustrated in FIG. 16, when the region to be predicted is the skip mode, since the highest resolution among the resolutions of the motion vectors of the surrounding region is 1/8, it is assumed that the resolution of the region to be predicted is 1/8. Predicting the predicted motion vector may be (10, 10).

In addition, when predicting the predicted motion vector of the region to be predicted using the motion vector of the surrounding region, the differential vector encoder 950 may convert and predict the motion vector of the surrounding region at an arbitrary resolution. In this case, when the arbitrary motion vector resolution and the motion vector resolution of the region to be predicted are not the same, the final predicted motion vector may be obtained by converting the predicted motion vector to the motion vector resolution of the region to be predicted. Referring to the example shown in FIG. 16, if the arbitrary motion vector resolution is 1/2 pixel unit, the predicted motion vector is converted to (3,3), and if the motion vector resolution of the region to be predicted is 1/8 pixel unit Since it is not the same as any motion vector resolution, the predicted motion vector (3,3) may be converted in units of 1/8 pixels so that the final predicted motion vector may be (12,12). In the same way, if the motion vector resolution of the region to be predicted is 1/4 pixel unit, the final predicted motion vector may be (6,6).

17 is a flowchart illustrating a video encoding method using adaptive motion vector resolution according to the first embodiment of the present invention.

In an image encoding method using adaptive motion vector resolution according to the first embodiment of the present invention, the motion vector resolution is determined for each region or motion vector, and the image is encoded using the motion vector according to the motion vector resolution determined for each region or motion vector. Inter prediction coding is performed on an area basis. To this end, the image encoding apparatus 900 using the adaptive motion vector resolution according to the first embodiment of the present invention determines whether to change the motion vector resolution for each region or motion vector of the image (S1710), and moves for each region. When the vector resolution is changed, the motion vector resolution is determined for each region or motion vector (S1720), and the image is inter prediction encoded by region using a motion vector according to the motion vector resolution determined for each region or motion vector (S1730). In the case of fixing the motion vector resolution without changing the resolution for each region, the inter prediction encoding is performed on a region-by-region basis by using a motion vector according to a fixed motion vector resolution for all regions or subregions of a portion of the image (S1740). .

Here, the motion vector resolution determined for each region may be different for each x component and y component of the motion vector.

Also, the image encoding apparatus 900 may generate a resolution change flag indicating whether the motion vector resolution is determined for each region or motion vector. For example, when it is determined in step S1710 that the motion vector resolution is changed, the image encoding apparatus 900 generates a resolution change flag (eg, '1') indicating that the motion vector resolution is changed for each region. When it is determined in S1710 that the motion vector resolution is fixed without being changed, a resolution change flag (eg, '0') indicating that the motion vector resolution is fixed for each region may be generated. In contrast, the image encoding apparatus 900 generates a resolution change flag according to the setting information input from the user or the outside, and determines whether to determine the motion vector resolution for each region in step S1710 based on the bit value of the generated resolution change flag. You can also judge.

Also, the image encoding apparatus 900 may encode a motion vector resolution determined for each region or motion vector. For example, the image encoding apparatus 900 hierarchically encodes a motion vector resolution determined for each region or motion vector in a quad tree structure by tying regions having the same motion vector resolution, or a motion vector determined for each region or motion vector. The resolution is encoded using a motion vector resolution predicted using the motion vector resolution of the surrounding area of each region, or the motion vector resolution determined for each region or motion vector is encoded using run and length, or the motion vector determined for each region. Resolution is encoded hierarchically using tag tree, or it is encoded by changing the number of bits allocated to motion vector resolution according to the frequency of motion vector resolution determined for each region or motion vector, or motion vector determined for each region or motion vector. sea It is determined whether or not the image can be estimated according to a pre-scheduled estimation method by the image decoding apparatus, and an identifier indicating that the motion vector resolution can be estimated is encoded, and the motion vector resolution cannot be estimated. An identifier indicating that the region cannot be estimated and a motion vector resolution may be encoded together. Here, when the image encoding apparatus 900 hierarchically encodes the quad tree structure or hierarchically encodes using the tag tree, the maximum number of quad tree layers and the area size indicated by the lowest node of the quad tree layer respectively. Alternatively, an identifier indicating the maximum number of tag tree hierarchies and the area size indicated by the lowest node of the tag tree hierarchies may be encoded and included in the header.

In addition, the image encoding apparatus 900 may determine a motion vector resolution determined for each region if the size of the predicted motion vector or the difference vector of the motion vector according to the motion vector resolution determined for each region is larger than a threshold. Can be determined by value. In addition, when each component of the difference vector is '0', the resolution of the region or the motion vector may not be encoded.

Also, the image encoding apparatus 900 may encode a difference vector that is a difference between a motion vector and a predicted motion vector according to a motion vector resolution determined for each region or motion vector. In this case, the image encoding apparatus 900 may encode the difference vector differently according to the motion vector resolution. To this end, when encoding the difference vector using UVLC, the image encoding apparatus 900 may encode using the K-th order exponential golem code, and in this case, the image of the exponential golem code according to the motion vector resolution determined for each region. The order K can be changed. When the image encoding apparatus 900 encodes the difference vector using CABAC, the image encoding apparatus 900 may encode using the unary / K-order exponential golem combined code, in which case the unary / K according to the motion vector resolution determined for each region. The order K and the maximum value T of the difference exponential golem combined code may be changed and encoded. When encoding a difference vector using CABAC, the cumulative probability may be calculated and encoded differently according to a motion vector resolution determined for each region.

In addition, the image encoding apparatus 900 may predict the predicted motion vector of the motion vector of each region by using the motion vector of the surrounding region of each region. In this case, the motion vector resolution of the surrounding region and the motion vector resolution of each region may be predicted. If is not the same, the motion vector resolution of the surrounding area may be converted to the motion vector resolution of each area to be predicted.

In addition, the image encoding apparatus 900 may change the method of encoding the resolution identification flag according to the distribution of the motion vector resolutions of the surrounding areas of each region based on the motion vector resolution determined for each region or motion vector.

In addition, when the image encoding apparatus 900 performs entropy encoding by arithmetic coding, a method of generating a binarization bit string of a resolution identification flag based on a distribution of a motion vector resolution determined for each region or motion vector according to a distribution of motion vector resolutions of surrounding regions of each region. Are applied differently according to the distribution of the motion vector resolution of the surrounding area and the probability of the motion vector resolution that has occurred so far to perform arithmetic coding and probability update. In addition, arithmetic coding and probability updating are performed using different probability models according to bit positions.

In addition, when the block mode of one or more regions of each region is the skip mode, the image encoding apparatus 900 may convert and predict the motion vector resolution of the region to be predicted to the highest resolution among the motion vector resolutions of the surrounding region. .

32 is a diagram illustrating an image encoding apparatus 3200 using an adaptive motion vector, according to a second embodiment of the present invention.

The image encoding apparatus 3200 using the adaptive motion vector according to the second embodiment of the present invention includes an inter prediction encoder 3210, a resolution change flag encoder 3220, and a resolution determiner ( Resolution Determiner 3230, Resolution Encoder 3240, Differential Vector Encoder 3250, and Resolution Conversion Flag Generator 3260. The resolution designation flag generator 3220, the resolution encoder 3240, the differential vector encoder 3250, and the resolution conversion flag generator 3260 are not necessarily all included in the image encoding apparatus 3200. May optionally be included.

The inter prediction encoder 3210 inter-prediction encodes an image in units of regions by using a motion vector according to a motion vector resolution determined for each region or motion vector of the image. The inter prediction encoder 3210 may be implemented with the image encoding apparatus 100 described above with reference to FIG. 1. Here, when one or more components among the resolution encoder 3240 and the difference vector encoder 3250 of FIG. 32 are additionally included, the functions of the components included in the inter prediction encoder 3210 and the functions of the encoder 150 in the inter prediction encoder 3210 may be included. In case of overlapping, the overlapping function may be omitted in the encoder 150. In addition, when the function of the predictor 110 in the inter prediction encoder 3210 and the function of the resolution determiner 3230 overlap, the overlapping function may be omitted in the predictor 110.

In addition, although one or more components added among the resolution encoder 3240 and the differential vector encoder 3250 may be configured as components separate from the inter prediction encoder 3210 as shown in FIG. 32, the encoder in the inter prediction encoder 3210 may be used. It may be configured integrally with 150. Also, the flag information generated by the resolution specifying flag generator 3220 and the resolution switching flag generator 3260 may be converted into a bitstream by the resolution specifying flag generator 3220 and the resolution switching flag generator 3260, respectively. The function may be performed by the encoder 150 in the inter prediction encoder 3210.

The functions of the inter prediction encoder 3210, the resolution encoder 3240, and the difference vector encoder 3250 are the same as the functions of the inter prediction encoder 910, the resolution encoder 940, and the differential vector encoder 950 in FIG. 9. Since each may be the same or similar, detailed descriptions of the inter prediction encoder 3210, the resolution encoder 3240, and the differential vector encoder 3250 are omitted.

The resolution designation flag generator 3220 may designate an adaptive degree of resolution differently for each region or motion vector of the image. The resolution designation flag generator 3220 may generate and include a resolution designation flag for designating a set of the motion vector resolution and / or the resolution of the differential motion vector for each region or motion vector of the image. The area for specifying the motion vector resolution and / or the differential motion vector resolution by the resolution specification flag may be a block, a macroblock, a group grouping blocks, a group grouping macroblocks, or an area having an arbitrary size such as MxN. have. That is, the resolution designation flag generator 3220 may generate and include a resolution designation flag indicating a resolution available for all regions or sub-regions in a partial region of the image in the bitstream. The resolution designation flag may be determined and generated according to setting information input by a user, or may be determined and generated according to a predetermined criterion by analyzing an image to be encoded. Such a resolution designation flag may be included in a header of the bitstream. The header may be a picture parameter set, a sequence parameter set, a slice header, or the like.

If the resolution designation flag specifies the resolution type as 1/2 or 1/4, the optimal resolution determined by the resolution determiner 3230 and the resolution identification flag encoded by the resolution encoder 3240 are 1/2 and 1/4. The resolution identification flag can be encoded in any manner. 33 is a diagram illustrating a resolution identification flag when the designated resolutions are 1/2 and 1/4.

In addition, the resolution identification flag may be encoded using Unary, CABAC, quad tree encoding, or the like. For example, when encoding with CABAC, a binary bit string may be generated with reference to the table of FIG. 33, and then arithmetic and probability encoding may be performed. For example, the probabilistic model can be divided into three cases by using the resolution of the surrounding motion vector or block. 34 illustrates A, B, and C, which are neighboring blocks of the current block X, and FIG. 35 is a diagram illustrating a probability model according to a condition.

If the resolution designation flag specifies the resolution type as 1/2, 1/4, or 1/8, the resolution identification flag to be encoded is selected from resolutions of 1/2, 1/4, and 1/8, and the resolution identification flag is It can be encoded in any way. 36 is a diagram illustrating a resolution identification flag when the designated resolutions are 1/2, 1/4, and 1/8. Referring to FIG. 36, the resolution identification flag may be 0, 10, or 11.

In this case, the resolution identification flag may be encoded using Unary, CABAC, quad tree encoding, or the like. For example, when CABAC is encoded, a binary bit string may be generated with reference to the table of FIG. 36 and then arithmetic and probability encoded. For example, the probabilistic model can be divided into six cases using a binary bit string index (binidx) and the surrounding motion vector or block resolution. Here, as shown in FIG. 34, when A, B, and C, which are neighboring blocks of the current block X, are shown, FIG. 37 is a diagram illustrating a probability model according to a condition.

Meanwhile, the resolution specifying flag generated by the resolution specifying flag generator 3220 may indicate a single resolution. For example, when the resolution is fixed to 1/2 without adaptively applying the resolution, the resolution designation flag may be encoded to indicate that the resolution of the corresponding area is fixed at 1/2 resolution.

In the case where a plurality of reference pictures are used, the adaptive degree of the resolution (that is, the resolution set) can be changed for each reference picture according to an arbitrary criterion without encoding the resolution designation flag. For example, the degree of adaptability of the resolution may vary according to the distance between the reference picture and the current picture.

38 and 39 are diagrams illustrating the degree of adaptability (ie, a resolution set) according to a distance between a current picture and a reference picture. As shown in FIG. 38, when the distance between the current picture and the reference picture is the closest among the plurality of reference pictures, the resolution selects an optimal resolution from a set of resolutions of 1/1, 1/2, 1/4, and 1/8. And encoding a resolution identification flag, and if the distance between the current picture and the reference picture is the furthest of the plurality of reference pictures, the resolution selects an optimal resolution from a resolution set of 1/2, 1/4, and encodes the resolution identification flag. If the distance between the current picture and the reference picture is not the furthest or closest among the plurality of reference pictures, the resolution can be set to select and encode the optimal resolution among the 1/2, 1/4, and 1/8 resolution sets. Can be. Referring to FIG. 39, it can be seen that a single resolution may be used.

In addition, the degree of adaptability of the resolution may be changed by using error measuring means such as a sum of squared difference (SSD) between the resolutions when generating the reference picture. For example, when the available resolutions are 1/1, 1/2, 1/4, and 1/8, when interpolating a reference picture, SSD, etc., for the resolution of 1/1 and 1/2 are first used. If the error value obtained using the error measuring means does not exceed a certain threshold, the resolution 1/2 may not be used, and if the error value exceeds the threshold, the resolution 1/2 may be set to be used. When the resolution of 1/2 is set not to be used, 1/4 is used when the error value obtained by using an error measuring means such as SSD for the resolution of 1/1 and 1/4 does not exceed an arbitrary threshold value. If the error value exceeds a certain threshold, set both 1/1 and 1/4 resolutions. When the resolution of 1/4 is set, do not use 1/8 when the error value obtained by using an error measuring means such as SSD for the resolution of 1/4 and 1/8 does not exceed an arbitrary threshold value. If the error value exceeds a certain threshold, all resolutions of 1/1, 1/4, and 1/8 are used. Here, the threshold may be different or the same by resolution or by quantization parameter.

In addition, it is possible to encode that the degree of adaptability of the resolution is different for each reference picture. For example, when three reference pictures are used, as shown in FIG. 40, a reference picture index number (the resolution set index in FIG. 9, which may be a reference picture number) for each predetermined resolution set is stored in a header to the decoding apparatus. Can transmit

41 is a diagram illustrating a structure in which reference pictures are encoded.

Meanwhile, the resolution designation flag generator 3220 may set the resolution set differently in case of a picture to be used as a reference picture and when it is not used as a reference picture. For example, it is assumed that reference pictures are encoded in the structure as shown in FIG. 41 and according to a time layer, TL0, TL1, and TL2 are pictures used as reference pictures, and pictures of TL3 are pictures not used as reference pictures. . In this case, when the resolution set is set to 1/2 and 1/4 by the resolution designation flag generator 3220, the resolution set for each reference picture may be set as shown in FIG. 42.

Referring to FIG. 42, the resolution sets at the time of encoding picture 6 are 1/2 and 1/4, and the resolution identification flag or the resolution designation flag is encoded in an arbitrary region or motion vector unit with respect to the resolution set thus determined. Can be. When picture 9 is encoded, since the resolution is a single resolution, the resolution identification flag or the resolution designation flag may not be encoded.

Meanwhile, the resolution designation flag generator 3220 may include all the functions of the resolution change flag generator 920 mentioned in the description of FIG. 9.

The resolution switch flag generator 3260 generates a resolution switch flag indicating a change from the surrounding or previous value of the area to encode the current resolution.

43 is a diagram illustrating a resolution of a current block and a resolution of a neighboring block.

For example, if the resolution set is 1/2, 1/4, 1/8, and the resolution of the neighboring block and the current optimal resolution are as shown in FIG. 43, the resolution of the neighboring block is (1/8, 1/4, 1 / 4, 1/4), so the highest resolution is 1/4. In addition, since the optimal resolution of the current block X is 1/4, the resolution conversion flag is encoded as 0. In this case, the decoder extracts the resolution switch flag from the bitstream, and when the resolution switch flag is 0, it can be seen that 1/4, the resolution having the highest frequency among the resolutions of the neighboring blocks, is the resolution of the current block X.

FIG. 44 is a diagram illustrating another example of the resolution of the current block and the resolution of the neighboring block, and FIG. 45 is a diagram illustrating a resolution identification flag according to the resolution.

In FIG. 44, since the optimal resolution of the current block X is not 1/4 of the highest frequency among the resolutions of the neighboring blocks, the resolution conversion flag is coded as 1 to indicate that the resolution is different from the neighboring value. The resolution identification flag is encoded as 1 using the table. If the current resolution is 1/4, it is unlikely that 1/4 will be selected as the resolution to be switched. Therefore, resolution 1/4 is not given a resolution identification flag.

46 is a diagram illustrating a resolution of a current block and a resolution of a neighboring block.

For example, when the resolution set is 1/2, 1/4, and the encoding is performed as shown in FIG. 46, the previous block of the current block X is A, and thus the resolution switch flag may indicate whether the resolution of the block A and the current block X are the same. Therefore, in the above case, since the resolutions of the block A and the current block X are not the same, the resolution switching flag may be 1. In addition, since the resolution sets are 1/2 and 1/4, it is understood that the resolution of the current block is 1/4 using only the resolution conversion flag without encoding the resolution identification flag.

FIG. 47 is a diagram illustrating a video encoding method using adaptive motion vector resolution, according to a second embodiment of the present invention.

As shown in FIG. 47, in the video encoding method using the adaptive motion vector resolution according to the second embodiment of the present invention, a resolution designation flag generation step (S4702), a resolution determination step (S4704), and an inter prediction encoding step (S4706) , Differential vector encoding step S4708, resolution encoding step S4710, and resolution conversion flag generation step S4712.

The resolution designation flag generation step S4702 is performed by the operation of the resolution designation flag generator 3220, the resolution determination step S4704 is performed by the resolution determiner 3230, and the inter prediction encoding step S4706 is the inter prediction encoder 3210. In operation S4708, the difference vector encoding operation S4708 operates in the operation of the difference vector encoder 3250, the resolution encoding operation S4710 operates in the operation of the resolution encoder 3240, and the resolution conversion flag generating operation S4712 includes the resolution switching flag. Since the operation corresponds to the operation of the generator 3260, detailed description thereof will be omitted.

In addition, there may be a step omitted from the image encoding method using the adaptive motion vector resolution according to the second embodiment of the present invention, depending on whether each component of the image encoding apparatus 3200 exists.

18 is a block diagram schematically illustrating an apparatus for decoding an image using adaptive motion vector resolution according to a first embodiment of the present invention.

The image decoding apparatus 1800 using the adaptive motion vector resolution according to the first embodiment of the present invention includes a resolution change flag extractor 1810, a resolution decoder 1820, a differential vector decoder 1830, and inter prediction decoding. It can be configured to include a group (1840).

The resolution change flag extractor 1810 extracts a resolution change flag from the bitstream. That is, the resolution change flag extractor 1810 extracts a resolution change flag indicating whether the motion vector resolution is fixed or changed for each region in the header of the bitstream. When the resolution change flag extractor 1810 indicates that the motion change resolution is fixed, the resolution change flag extractor 1810 extracts and decodes the encoded motion vector resolution from the bitstream to inter-predict the fixed motion vector resolution or the preset motion vector resolution. The decoder 1840 controls inter prediction decoding on all sub-regions defined in the header, and controls the difference vector decoder 1830 to reconstruct the motion vector of each region at a fixed motion vector resolution. When the resolution change flag extractor 1810 indicates that the resolution change flag indicates that the motion vector resolution is changed for each region or motion vector, the resolution decoder flag 1820 causes the motion decoder 1820 to move the motion vector of each region or motion vector of the lower region defined in the header. Control to reconstruct the resolution, and control the inter prediction decoder 1840 to inter predict and decode the reconstructed motion vector resolution for the lower angular region or motion vector defined in the header, and cause the differential vector decoder 1830 to Control to restore the motion vector of each region to the reconstructed motion vector resolution.

Further, the resolution change flag extractor 1810 may determine the motion vector resolution determined for each region or motion vector when the size of the predicted motion vector or the difference vector of the motion vector according to the motion vector resolution determined for each region or motion vector is larger than the threshold. May be determined as a predetermined value. For example, when the magnitude of the prediction motion vector or the difference vector of a certain region is larger than the threshold, the motion vector resolution of the region or motion vector is determined as a predetermined value or a random value. It may not be decoded. In addition, when the size of the motion vector of the region or the surrounding vector of the motion vector is large or the size of the motion vector of the region is larger than the threshold value, the motion vector resolution of the region or the motion vector is set to a predetermined value or any value. In this case, the motion vector resolution may not be decoded. In this case, the motion vector resolution of the region or motion vector may be changed to any resolution without any flag. The threshold may be a predetermined value or any value input and may be calculated from motion vectors of surrounding blocks.

The resolution decoder 1820 extracts and decodes the resolution identification flag encoded from the bitstream according to the resolution change flag extracted by the resolution change flag extractor 1810 to restore the motion vector resolution for each region. In the following description, although the resolution decoder 1820 decodes the motion vector resolution, this is for convenience of description, and in reality, the resolution of the motion vector resolution or the differential motion vector or both resolutions are decoded. You may. Accordingly, the resolution indicated by the resolution identification flag may be the resolution of the motion vector resolution or the differential motion vector, or may be used simultaneously for the motion vector resolution and the resolution of the differential motion vector.

To this end, the resolution decoder 1810 may reconstruct the motion vector resolution for each region or motion vector by decoding resolution identification flags hierarchically encoded in a quad tree structure by tying regions having the same motion vector resolution.

10 to 12, the resolution decoder 1820 reconstructs a motion vector resolution by decoding a resolution identification flag by dividing a quad tree form as shown in FIG. 10 into respective regions as shown in FIG. 12. For example, when decoding the resolution identification flag generated by encoding as shown in FIG. 11, the first bit is '1' and thus the layer is divided. The second bit is '0', so the first node of level 1 Since it is known that is not divided, the next bits are decoded to restore the motion vector resolution of 1/2. In this manner, the same as the encoding method described above with reference to FIGS. 10 and 11, but in reverse order, the resolution identification flags for level 1 and level 2 are decoded to restore motion vector resolution of the corresponding regions or motion vectors. In level 3, since the maximum number of layers is defined as three by the identifier indicating the maximum number of layers included in the header and the size of the area indicated by the lowest node, it is determined that there are no lower layers, so only the motion vector resolution of each area is determined. Restore To this end, the resolution decoder 1810 decodes an identifier indicating the maximum number of quad tree layers and the area size indicated by the lowest node of the quad tree layer, which is included in the header of the bitstream.

In addition, in the above-described example, only two examples of nodes are divided into lower layers (that is, divided into four zones) or not divided are shown. However, as shown in FIG. 20, the nodes are not divided into lower layers. Well, a node can be divided into lower layers in various ways: divided into two horizontally long regions, divided into two vertically long regions, and divided into four regions.

In addition, the resolution decoder 1820 decodes the resolution identification flag encoded by using the motion vector resolution predicted by using the motion vector resolution of each area or the surrounding area of the motion vector to determine the motion vector resolution for each area or motion vector. Can be restored For example, the resolution decoder 1820 indicates that the resolution identification flag extracted for each region or motion vector from the bitstream is the same as the motion vector resolution predicted using the motion vector resolution of the surrounding region (e.g., For example, when the bit value of the resolution identification flag is '1'), the motion vector resolution predicted using the motion vector resolution of the surrounding area of the corresponding region may be restored without reading the next resolution identification flag from the bitstream. If the resolution identification flag indicates that it is not the same as the motion vector resolution predicted using the motion vector resolution of the surrounding area (for example, when the bit value of the resolution identification flag is '0'), the corresponding resolution is identified from the bitstream. Solution of motion vector restored by reading and decoding the next bit value of the flag You can restore the top view.

In addition, the resolution decoder 1820 may restore the motion vector resolution for each region or motion vector by decoding the resolution identification flag in which the run and the length of the motion vector resolution for each region or the motion vector are encoded. For example, the resolution decoder 1820 decodes the resolution identification flag encoded together with the motion vector resolution and / or the differential motion vector resolution for some regions including several regions to restore the run and length of the motion vector resolution. The motion vector resolution of the regions shown in FIG. 12 may be restored using the reconstructed run and length of the motion vector resolution.

In addition, the resolution decoder 1820 may reconstruct the motion vector resolution for each region by decoding the resolution identification flag hierarchically coded using the tag tree for the motion vector resolution for each region or motion vector. For example, with reference to FIGS. 13 and 14, the resolution decoder 1820 may know that the bit corresponding to the level 0 is '01' because the bit of the first region illustrated in FIG. 14 is '0111'. Since the number of motion vector resolutions of the upper layer from 0 is assumed to be '0', it is possible to restore the motion vector resolution of 1/2 in which the difference value of the number is increased to '1'. In addition, since the next bit is '1', the difference between the number and the upper layer is '0', so that the level 1 is also restored to the motion vector resolution of 1/2. Since the next bit is also '1', level 2 and level 3 may be restored to a motion vector resolution of 1/2 as well. The bit of the second region shown in FIG. 14 is '01', and since it has already been decoded for level 0, level 1, and level 2 in the first region, only level 3 needs to be decoded. Since the difference is '1', the motion vector resolution of 1/4 can be restored. In this way, the motion vector resolution of the remaining areas can also be restored.

The resolution decoder 1820 may change and decode the number of bits allocated to the resolution identification flag according to the frequency of the motion vector resolution for each region or motion vector. For example, the resolution decoder 1820 may calculate the frequency of the motion vector resolution reconstructed up to the last region, number the motion vector resolution according to the calculated frequency, and allocate and decode the number of bits according to the number.

An area group may be a quad tree, a quad tree bundle, a tag tree, a tag tree bundle, a macroblock, a macroblock bundle, and an area of any size. For example, when the area group is designated as two macroblocks, the frequency of occurrence of the motion vector resolution is updated by every two macroblocks, and the number of bits of the motion vector resolution is allocated and decoded, or the area group is designated as four quad trees. The frequency of occurrence of the motion vector resolution is updated every four quad trees, and the number of bits of the motion vector resolution is allocated and decoded.

In addition, the resolution decoder 1820 may vary the decoding method of the resolution identification flag based on the distribution of the motion vector resolution of the surrounding area of each region with the motion vector resolution determined for each region or motion vector. According to the distribution of the motion vector resolution of the surrounding area or region group, the shortest number of bits is allocated and decoded to the motion vector resolution having the highest probability of being the resolution of the corresponding area. For example, if the motion vector resolution of the left region of the region is 1/2 and the motion vector resolution of the upper region is 1/2, the motion vector resolution of the region is most likely to be 1/2, so 1 / The shortest number of bits is allocated to two motion vector resolutions and decoded. As another example, the motion vector resolution of the left region of the region is 1/4, the motion vector resolution of the upper region is 1/2, the motion vector resolution of the upper region is 1/2 and the motion vector resolution of the upper right region is 1 In case of / 2, the probability of motion vector resolution of the corresponding region is decoded by allocating short bits in order of high probability such as 1/2, 1/4, 1/8, and so on.

In addition, when the resolution decoder 1820 entropy decodes by arithmetic decoding, a method of generating a binarization bit string of the resolution identification flag based on a distribution of motion vector resolutions of the surrounding regions of each region according to a motion vector resolution determined for each region or motion vector. In this case, the arithmetic decoding and the probability update may be performed by differently applying the probabilistic model according to the distribution of the motion vector resolution of the surrounding area and the probability of the motion vector resolution generated up to now. In addition, arithmetic decoding and probability updating may be performed using different probability models according to bit positions. For example, assuming that only three motion vector resolutions such as 1/2, 1/4, and 1/8 are used for decoding with CABAC, and the motion vector resolution used for decoding is 1/2, 1/4, 1/8, If the resolution is 1/2 and the motion vector resolution of the upper region is 1/2, assign the shortest binarization bit string of '0' to 1/2 motion vector resolution as shown in Fig. 21, and the remaining 1/4 motion vector resolution and 1 The / 8 motion vector resolutions are assigned short binarization bits in the order of the highest probability of occurrence.

At this time, when the probability that the 1/8 motion vector resolution has occurred until now is higher than the 1/4 motion vector resolution, the bit string having '00' is allocated to the 1/8 motion vector resolution as shown in FIG. 21, and the bit having '01'. Assign columns to 1/2 motion vector resolution. In addition, when decoding the first binarization bit, the probability model may be divided into four probability models to be decoded. The four probabilistic models are, firstly, probabilistic models where both the left and top regions have the same resolution and are the same as the highest probabilities resolution so far. Second, the left and the top regions have the same resolution and the highest Probabilistic model when the resolution of the probability is different from the third. Third, the probability model when the resolution of the left region and the upper region are all different and the resolution of the highest probability up to now and the resolution of the left region or the upper region is the fourth. Probability models in the case where the resolution of the left region and the upper region and the resolution of the highest probability up to now are all different. When decoding the second binarization bit, the probability model may be divided into two probability models to be decoded. The two probabilistic models are the first when the motion vector resolutions of both the left and top regions of the corresponding region are different and have the highest probability motion vector resolution so far and the motion vector resolution such as the motion vector resolution of the left or upper region. Probability model can be classified into a probability model in which the motion vector resolution of the left region and the upper region of the corresponding region and the resolution of the highest probability so far are different.

As another example, assuming that only motion vector resolutions such as 1/2, 1/4, and 1/8 are used for decoding with CABAC and decoding, the most probable motion so far If the vector resolution is 1/4, assign the shortest binarization bit string '1' to the 1/4 motion vector resolution, and '00' and 'for the remaining 1/2 motion vector resolution and 1/8 motion vector resolution, respectively. Assign 01 '.

In addition, when encoding the first binarization bit, the probability model may be divided into three probability models and decoded. The three probabilistic models are: first, a probability model where the motion vector resolution of the left region and the upper region of the region is the same as the highest resolution that has occurred so far; and second, only one of the resolutions of the left region and the upper region of the region. Probability model in which the probability that occurred so far is the same as the highest resolution, and third, the probability model when the motion vector resolution of the left and upper regions of the corresponding region differs from the motion vector resolution of the highest probability so far. Can be. When decoding the second binarization bit, the probability model may be divided into six probability models to be decoded. The six probabilistic models are, firstly, probabilistic models when the resolution of the left and top regions of the region is 1/8 motion vector resolution, and second, the motion vector resolution of the left and upper regions of the region is 1/1. Probabilistic model with 2 motion vector resolutions, Third, Probabilistic model with motion vector resolutions of the left and upper regions of the corresponding region being 1/4 motion vector resolution, and Fourth, resolution of the left and upper regions of the region Probability model when one is 1/8 motion vector resolution and the other is 1/4 motion vector resolution.Fifth, one of the motion vector resolutions of the left and upper regions of the region is 1/2 motion vector resolution and the other Probabilistic model for one motion of 1/4 motion vector resolution, and sixth for motion vector resolutions of the left and upper areas of the area. Im a vector resolution and the other may be separated by one-half if the probability model of the motion vector resolution. The resolution with the highest probability so far may be the probability of the resolution encoded up to the previous region, the probability of any region, or the predetermined fixed resolution.

In addition, when the resolution decoder 1820 is a flag indicating that a resolution identification flag decoded for each region or motion vector can be estimated, the resolution decoder 1820 estimates the motion vector resolution according to a pre-determined estimation method and uses the estimated motion vector resolution as the region or the motion vector resolution. In the case of a flag indicating that the resolution identification flag decoded for each region or motion vector cannot be estimated, the motion vector resolution indicated by the decoded resolution identification flag can be restored as a motion vector of the corresponding region. have.

For example, when the resolution identification flag decoded for each region indicates that the motion vector resolution can be estimated, the resolution decoder 1820 may perform each motion vector resolution decoded by the same or similar method as that of the image encoding apparatus 900. The predicted motion vector is predicted by using a change, and the motion vector is reconstructed using the predicted predictive motion vector and the difference vector reconstructed by the difference vector decoder 1830. First, assuming that the motion vector resolution of a region is 1/4 pixel unit, when the prediction motion vector is (3, 14), the difference vector is (2, 3), and thus the reconstructed motion vector of the region is (5, 17). ) Assuming that the motion vector resolution of the region is 1/2 pixel, if the predicted motion vector is (2, 7), then the reconstructed difference vector is (2, 3), so the reconstructed motion vector of the region is (4, 10). ) Using the reconstructed motion vector for each resolution, the resolution at which the distortion is minimized between the surrounding pixels of the region whose motion is compensated in the reference picture and the surrounding pixels of the region is the optimal motion vector resolution. Therefore, when the distortion of the surrounding pixels of the motion-compensated region in the unit of 1/2 pixel is the smallest, the motion vector resolution of 1/2 becomes the optimal motion vector resolution.

In addition, when the resolution identification flag decoded for each region or motion vector indicates that the motion vector resolution is estimated, the resolution decoder 1820 may additionally decode and reconstruct the motion vector resolution of the corresponding region or motion vector in the resolution identification flag. Can be.

In addition, the resolution decoder 1820 may restore the motion vector resolution for each region or motion vector only when each component of the difference vector is not '0'. That is, when the component of the difference vector of the specific region is '0', since the prediction motion vector is decoded into the motion vector, the motion vector resolution of the corresponding region may not be restored.

The difference vector decoder 1830 extracts and decodes the encoded difference vector from the bitstream. The difference vector decoder 1830 reconstructs the difference vector for each region or motion vector by decoding according to the motion vector resolution for each reconstructed region or motion vector. A motion vector for each area may be reconstructed by predicting the prediction motion vector for each area and using the reconstructed difference vector and the predictive motion vector.

To this end, the difference vector decoder 1830 can decode the difference vector using UVLC, and in this case, decode using the K-order exponential golem code, but in accordance with the motion vector resolution for each reconstructed region. You can change the order (K) of the code. In addition, the difference vector decoder 1830 can decode the difference vector using CABAC. In this case, the vector is decoded using the unary / K-order exponential golem combined code, but the motion vector resolution for each reconstructed region or motion vector. The decoding can be performed by changing the order (K) and the maximum value (T) of the unary / K-order exponential golem combined code. In addition, when the difference vector decoder 1830 decodes the difference vector using the CABAC, the difference vector decoder 1830 may calculate and decode the cumulative probability differently according to the motion vector resolution for each reconstructed region or motion vector.

In addition, the difference vector decoder 1830 may predict the predicted motion vector for each region or the motion vector by using the motion vector of the region or the surrounding region of the motion vector. When the motion vector resolutions of Δ are not the same, the motion vector resolution of the surrounding area may be converted into the motion vector resolution of each area to be predicted. The predictive motion vector may be obtained in the same manner in the image encoding apparatus and the image decoding apparatus. Accordingly, various embodiments of deriving a predictive motion vector by the image encoding apparatus and embodiments of motion vector resolution conversion with reference to FIGS. 22 to 26 may be similarly implemented in the embodiment of the image decoding apparatus according to the present invention. have.

In addition, when one or more of each region is a block and the block mode of the block is the skip mode, the differential vector decoder 1830 converts the motion vector resolution of the surrounding region to the highest resolution among the motion vector resolutions of the surrounding region. It can be predicted.

In addition, the resolution identification flag indicating the motion vector resolution decoded by the resolution decoder 1820 may represent both the x component and the y component of the motion vector, or may indicate respective resolutions. That is, when the camera photographing the image moves or an object in the image moves, the resolution decoder 1820 may decode the resolutions of the x component and the y component of the motion vector for motion estimation differently. For example, the x component of a motion vector of a certain region may be decoded at a motion vector resolution of 1/8 pixel unit, and the y component may be decoded at a motion vector resolution of 1/2 pixel unit, thereby inter prediction. The decoder 1840 may inter-prediction-decode a corresponding region by motion estimation and motion compensation with a motion vector having different motion vector resolutions for each x component and y component.

The inter prediction decoder 1840 inter-predicts and decodes each region by using motion vectors for each region according to the motion vector resolution of each reconstructed region or motion vector. The inter prediction decoder 1840 may be implemented by the image decoding apparatus 800 described above with reference to FIG. 8. Here, when the functions of the resolution change flag extractor 1810, the resolution decoder 1820, and the differential vector decoder 1830 of FIG. 18 overlap with the functions of the decoder 810 in the image decoding apparatus 800, the decoder. In 810, the overlapping function may be omitted. In addition, when the operation of the resolution decoder 1820 and the operation of the predictor 850 in the image decoding apparatus 800 overlap, the overlapping operation may be omitted in the predictor 850.

In addition, although the resolution change flag extractor 1810, the resolution decoder 1820, and the differential vector decoder 1830 may be configured as separate components from the inter prediction decoder 1840 as shown in FIG. 18, the inter prediction decoder It may be integrated with the decoder 810 in 1800.

However, although the above-described image decoding apparatus 800 described with reference to FIG. 8 encodes an image in units of blocks, the inter prediction decoder 1840 uses various regions such as macroblocks, blocks, subblocks, slices, pictures, and the like. It can be divided into areas of shape and size and decoded by area having an arbitrary size. The predetermined area may be a macroblock of 16 × 16 size, but may be a block of various sizes and shapes such as a 64 × 64 size block and a 32 × 16 size block.

In addition, although the image decoding apparatus 800 described above with reference to FIG. 8 performs inter prediction decoding on motion blocks and / or differential motion vectors having the same motion vector resolution and / or differential motion vector resolution for all blocks of the image, inter prediction is performed. The decoder 1840 may inter-prediction and decode each region with a motion vector and / or a differential motion vector having a motion vector resolution and / or a differential motion vector resolution determined differently for each region or motion vector. That is, when the inter prediction decoder 1840 inter predicts and decodes a predetermined region, the inter prediction decoder 1840 is already encoded according to the motion vector resolution and / or differential motion vector resolution of each region or motion vector reconstructed by the resolution decoder 1820. The interpolated, decoded and reconstructed reference picture to increase the resolution, and then the motion vector and / or differential according to the motion vector resolution and / or differential motion vector resolution of the corresponding region or motion vector reconstructed by the differential vector decoder 1830. Motion estimation is performed using the motion vector. As an interpolation filter for interpolating a reference picture, various filters such as a Wiener filter, a bilinear filter, a Kalman filter, and the like can be used. The resolution for interpolation is 2 pixel unit, 1 pixel unit, 1/2 pixel unit, 1 / Resolution of various fractional pixel units and integer pixel units such as 4 pixel units, 1/8 pixel units, and the like may be applied. In addition, different filter coefficients may be used or the number of filter coefficients may be changed according to the various resolutions. For example, when the resolution is 1/2 pixel unit, interpolation is performed using a Wiener filter, and when the resolution is 1/4 pixel unit, interpolation is performed using a Kalman filter. In addition, the number of filter taps may be changed when interpolating each resolution. For example, when the resolution is 1/2 pixel unit, interpolation may be performed using an 8 tap winner filter, and when the resolution is 1/4 pixel unit, an interpolation may be performed using a 6 tap winner filter.

In addition, the inter prediction decoder 1840 may decode filter coefficients for each motion vector resolution in the bitstream and interpolate the reference picture with optimal coefficients for each motion vector resolution. In this case, the filter may use any filter such as a Wiener `Kalman filter, and the number of filter taps may be several. In addition, the number of filters and filter taps may be changed for each motion vector resolution. In addition, inter prediction decoding may be performed using a reference picture interpolated using a different filter according to the motion vector resolution of the region or the motion vector. For example, in a bitstream, a filter with a 6-tap Winner filter for 1/2 motion vector resolution, an 8-tap Kalman filter for 1/4 motion vector resolution, and a filter coefficient of a linear filter for 1/8 motion vector resolution are decoded. The reference picture may be interpolated and decoded for each motion vector resolution. In this case, when the motion vector resolution of the current region or the motion vector is 1/2 motion vector resolution, it is decoded using the reference picture interpolated by the 6-tap winner filter, and when the 1/4 motion vector resolution is interpolated by the 8-tap Kalman filter. The reference picture can be used to decode.

48 is a block diagram schematically illustrating an apparatus for decoding an image using adaptive motion vector resolution according to a second embodiment of the present invention.

The video decoding apparatus 4800 using the adaptive motion vector resolution according to the second embodiment of the present invention includes a resolution designation flag extractor 4810, a resolution decoder 4820, a differential vector decoder 4830, and inter prediction decoding. 4840 and resolution switch flag extractor 4850. In this case, the resolution designation flag extractor 4810, the resolution decoder 4820, the differential vector decoder 4830, and the resolution change flag extractor 4850 are not necessarily included in the image decoding apparatus 4800. It may be selectively included according to the encoding method of the video encoding apparatus for generating the bitstream.

In addition, the inter prediction decoder 4840 inter-predicts and decodes each region by using motion vectors for each region according to the motion vector resolution of each reconstructed region or motion vector. The inter prediction decoder 4840 may be implemented by the image decoding apparatus 800 described above with reference to FIG. 8. Here, at least one of the functions of the resolution designation flag extractor 4810, the resolution decoder 4820, the difference vector decoder 4830, and the resolution change flag extractor 4850 of FIG. 48, and the decoding in the image decoding apparatus 4800. When the functions of the device 810 overlap, the overlapping function may be omitted in the decoder 810. In addition, when the operation of the resolution decoder 4820 and the operation of the predictor 850 in the inter prediction decoder 4840 overlap, the overlapping operation may be omitted in the predictor 850.

In addition, the resolution designation flag extractor 4810, the resolution decoder 4820, the differential vector decoder 4830, and the resolution conversion flag extractor 4850 are separate components from the inter prediction decoder 4840 as shown in FIG. 48. It may be configured, but may be integrated with the decoder 810 in the inter prediction decoder 4800.

The resolution specifying flag extractor 4810 extracts the resolution specifying flag from the input bitstream. The resolution designation flag is a flag indicating that it is fixed at a single resolution or a resolution set having one or more resolutions.

The resolution specification flag extractor 4810 extracts the resolution specification flag from the bitstream. That is, the resolution designation flag extractor 4810 extracts a resolution change flag indicating whether the motion vector resolution is fixed to an arbitrary value or there is a resolution set changed for each region in the header of the bitstream. If the resolution designation flag extractor 4810 indicates that the resolution designation flag indicates that the motion vector resolution and / or the differential motion vector resolution is fixed to an arbitrary resolution, the resolution specification flag extractor 4840 and the differential vector encoder In operation 4830, the difference vector encoder 4830 encodes the difference vector at the received resolution and transmits the difference vector to the inter prediction decoder 4840. In this case, the inter-prediction decoder 4840 performs inter-prediction decoding using the received difference vector and the resolution received from the resolution designation flag extractor 4810 and the received bitstream.

The resolution designation flag extractor 4810 causes the resolution decoder 4820 to perform motion vector resolution and / or differential motion vector of each region or motion vector of the lower region defined in the header when the extracted resolution designation flag is a predetermined resolution set. Control to reconstruct the resolution, and control the inter prediction decoder 4840 to inter predict and decode the reconstructed motion vector resolution for the lower angular region or motion vector defined in the header, and cause the differential vector decoder 4830 to Control to restore the motion vector of each region with the reconstructed differential motion vector resolution.

In addition, when using a plurality of reference pictures, when the resolution designation flag is not extracted from the bitstream, an adaptive degree of resolution (that is, a resolution set) may be calculated for each reference picture according to an arbitrary reference. For example, the degree of adaptability of the resolution may vary according to the distance between the reference picture and the current picture. Since the description thereof has been mentioned in the descriptions of FIGS. 38 and 39, a detailed description thereof will be omitted.

In addition, the resolution set may be calculated by using an error measuring means such as a sum of squared difference (SSD) between the resolutions when generating the reference picture. For example, when the available resolutions are 1/1, 1/2, 1/4, and 1/8, when interpolating a reference picture, SSD, etc., for the resolution of 1/1 and 1/2 are first used. If the error value obtained using the error measuring means does not exceed a certain threshold, the resolution 1/2 may not be used, and if the error value exceeds the threshold, the resolution 1/2 may be set to be used. When the resolution of 1/2 is set not to be used, 1/4 is used when the error value obtained by using an error measuring means such as SSD for the resolution of 1/1 and 1/4 does not exceed an arbitrary threshold value. If the error value exceeds a certain threshold, set both 1/1 and 1/4 resolutions. When the resolution of 1/4 is set, do not use 1/8 when the error value obtained by using an error measuring means such as SSD for the resolution of 1/4 and 1/8 does not exceed an arbitrary threshold value. If the error value exceeds a certain threshold, all resolutions of 1/1, 1/4, and 1/8 are used. Here, the threshold may be different or the same by resolution or by quantization parameter.

In addition, when encoding that the degree of adaptability of the resolution is different for each reference picture, the resolution designation flag extractor 4810 extracts the reference picture index number from the bitstream instead of the resolution designation flag, and for each predetermined resolution set as shown in FIG. A resolution set can be extracted by storing and referring to a corresponding reference picture index number.

In addition, when the resolution set is designated by the resolution designation flag, the resolution set that is actually used may be set differently according to whether the reference picture is used by setting the resolution set differently in case of a picture to be used as a reference picture and when it is not used as a reference picture. . Accordingly, the table shown in FIG. 42 is also stored in the image decoding apparatus 4800 and may be used by the resolution decoder 4820 to decode the resolution. Since the description thereof has been mentioned in the descriptions of FIGS. 41 and 42, a detailed description thereof will be omitted.

In addition, the resolution designation flag extractor 4810 may determine the area or the motion vector when the predicted motion vector or the differential motion vector of the motion vector according to the motion vector resolution and / or the differential motion vector resolution determined for each area or motion vector is larger than the threshold. The motion vector resolution and / or differential motion vector resolution determined for each motion vector may be determined as a predetermined value. Details of this have been mentioned in the description of the resolution change flag extractor 1810 of the image decoding apparatus 1800 according to the first embodiment, and thus detailed descriptions thereof will be omitted.

The resolution decoder 4820 extracts and decodes the resolution identification flag encoded from the bitstream according to the resolution designation flag extracted by the resolution designation flag extractor 4810 to restore the motion vector resolution for each region.

To this end, the resolution decoder 4810 may reconstruct the motion vector resolution for each region or motion vector by decoding regions of the same region having the same motion vector resolution and decoding the resolution identification flag hierarchically encoded in a quad tree structure. Since details thereof have been mentioned in the description of the resolution decoder 1810 of the image decoding apparatus 1800 according to the first embodiment, detailed descriptions thereof will be omitted.

In addition, the resolution decoder 4820 decodes the resolution identification flag encoded by using the motion vector resolution predicted by using the motion vector resolution of each region or the surrounding region of the motion vector, thereby obtaining the motion vector resolution for each region or motion vector. Can be restored Since details thereof have been mentioned in the description of the resolution decoder 1810 of the image decoding apparatus 1800 according to the first embodiment, detailed descriptions thereof will be omitted.

In addition, the resolution decoder 4820 may restore the motion vector resolution for each region or motion vector by decoding the resolution identification flag in which the run and the length of the motion vector resolution for each region or the motion vector are encoded. Since details thereof have been mentioned in the description of the resolution decoder 1810 of the image decoding apparatus 1800 according to the first embodiment, detailed descriptions thereof will be omitted.

In addition, the resolution decoder 4820 may restore the motion vector resolution for each region by decoding the resolution identification flag hierarchically encoded using the tag tree. Since details thereof have been mentioned in the description of the resolution decoder 1810 of the image decoding apparatus 1800 according to the first embodiment, detailed descriptions thereof will be omitted.

In addition, the resolution decoder 4820 may change the number of bits allocated to the resolution identification flag according to the frequency of the motion vector resolution for each region or the motion vector, and decode. For example, the resolution decoder 4820 may calculate the frequency of the motion vector resolution reconstructed up to the last region, number the motion vector resolution according to the calculated frequency, and allocate and decode the number of bits according to the number. Since details thereof have been mentioned in the description of the resolution decoder 1810 of the image decoding apparatus 1800 according to the first embodiment, detailed descriptions thereof will be omitted.

In addition, the resolution decoder 4820 may vary the decoding method of the resolution identification flag according to the distribution of the motion vector resolutions of the surrounding areas of each area. According to the distribution of the motion vector resolution of the surrounding area or region group, the shortest number of bits is allocated and decoded to the motion vector resolution having the highest probability of being the resolution of the corresponding area. Since details thereof have been mentioned in the description of the resolution decoder 1810 of the image decoding apparatus 1800 according to the first embodiment, detailed descriptions thereof will be omitted.

In addition, when the resolution decoder 4820 performs entropy decoding by arithmetic decoding, a method of generating a binary coded bit stream of the resolution identification flag based on the distribution of the motion vector resolution of each region or the motion vector resolution determined for each region or motion vector is determined. In this case, the arithmetic decoding and the probability update may be performed by differently applying the probabilistic model according to the distribution of the motion vector resolution of the surrounding area and the probability of the motion vector resolution generated up to now. In addition, arithmetic decoding and probability updating may be performed using different probability models according to bit positions. Since details thereof have been mentioned in the description of the resolution decoder 1810 of the image decoding apparatus 1800 according to the first embodiment, detailed descriptions thereof will be omitted.

In addition, when the resolution decoder 4820 is a flag indicating that a resolution identification flag decoded for each region or motion vector can be estimated, the resolution decoder 4820 estimates the motion vector resolution according to a pre-determined estimation method and uses the estimated motion vector resolution as the region or the motion vector resolution. In the case of a flag indicating that the resolution identification flag decoded for each region or motion vector cannot be estimated, the motion vector resolution indicated by the decoded resolution identification flag can be restored as a motion vector of the corresponding region. have. Since details thereof have been mentioned in the description of the resolution decoder 1810 of the image decoding apparatus 1800 according to the first embodiment, detailed descriptions thereof will be omitted.

When the resolution identification flag decoded for each region or motion vector indicates that the motion vector resolution is estimated, the resolution decoder 4820 may additionally decode and reconstruct the motion vector resolution of the corresponding region or motion vector in the resolution identification flag. Can be. In addition, the resolution decoder 4820 may restore the motion vector resolution for each region or motion vector only when each component of the difference vector is not '0'. That is, when the component of the difference vector of the specific region is '0', since the prediction motion vector is decoded into the motion vector, the motion vector resolution of the corresponding region may not be restored. Since details thereof have been mentioned in the description of the resolution decoder 1810 of the image decoding apparatus 1800 according to the first embodiment, detailed descriptions thereof will be omitted.

In addition, the resolution decoder 4820 extracts a resolution identification flag according to the resolution type of the resolution change flag extracted and decoded from the header, and by using the difference vector decoder 4830 to the motion vector resolution as shown in FIG. The value of the differential motion vector corresponding to the resolution is extracted by referring to the extracted code number by storing the differential motion vector code number table according to the same content as the video encoding apparatus. When the decoded resolution identification flag is 1/4, when the motion vector is decoded, it may be decoded using the prediction motion vector obtained by converting the motion vector extracted from the bitstream and the resolution 1/4 of the neighboring motion vector. The predicted motion vector can be obtained by taking a median using a neighboring motion vector transformed using multiplication and division similarly to the encoder, but the present invention is not limited thereto.

FIG. 49 is a diagram illustrating a peripheral motion vector of the current block X, and FIG. 50 is a diagram illustrating a converted value of the peripheral motion vector according to the current resolution.

In addition, the resolution decoder 4820 may calculate the resolution using the reference picture index without extracting the resolution identification flag. Since this is described in the description of FIGS. 30 to 31 in the image encoding apparatus 3200, detailed descriptions thereof will be omitted.

The inter prediction decoder 4840 may obtain a motion vector and perform decoding using the prediction motion vector obtained using the table shown in FIG. 50 and the difference vector calculated from the difference vector decoder 4830.

If the differential motion vector resolution is encoded and transmitted in the bitstream as described in the description of FIGS. 26 and 27, the resolution decoder 4820 extracts and decodes the resolution identification flag of the differential motion vector in the bitstream. The code number of the differential motion vector and the sign of the differential motion vector are decoded. If the code number table of the differential motion vector according to the differential motion vector resolution as shown in FIG. 27 is stored and the code number extracted from the bitstream is (1, 1), the differential motion vector code number table according to the differential motion vector resolution of FIG. By reference, the differential motion vectors (-1/8, -1/4) can be obtained. In this case, the inter prediction decoder 4840 may calculate the predictive motion vector in the same manner as the image encoding apparatus as follows.

PMVx = median (7/8, 1/8, 2/8) = 2/8

PMVy = median (-6/8, 1/8, -2/8) = -2/8

Thus PMV = (2/8, -2/8) = (1/4, -1/4).

So the motion vector is

MV (1/8, -4/8) = MVD (-1/8, -1/4)-PMV (1/4, -1/4)

And thus the decoded motion vector is (1/8, -4/8).

On the other hand, when the differential vector decoder 1830 receives the reference resolution flag, the differential vector decoder 1830 reconstructs the differential motion vector and decodes it at the reference resolution. In this case, the code number included in the reference resolution flag is extracted, the differential reference motion vector is decoded by referring to the code number table according to the differential reference motion vector resolution as shown in FIG. 29, and the difference is obtained by using the position information included in the reference resolution flag. Restore the motion vector. That is, when the differential vector decoder 1830 decodes the differential motion vector, the differential vector decoder 1830 extracts a reference resolution from the bitstream and calculates a differential reference motion vector using a code number table according to the differential reference motion vector resolution as shown in FIG. 29. do. At this time, the reference resolution flag representing the reference resolution and the position coordinate of the motion vector is extracted from the bitstream, and the differential motion vector is decoded using the reference resolution flag from the differential reference motion vector.

If the motion vector has a resolution other than the reference resolution, a method of additionally encoding the reference resolution flag may be used. The reference resolution flag may include data indicating whether the resolution is the same as the same reference resolution and data indicating the position of the actual motion vector.

Meanwhile, another function of the difference vector decoder 1830 may include a function of the difference vector decoder 1830 described in the image decoding apparatus 1800 according to the first embodiment. Since the detailed description has described the function of the differential vector decoder 1830 in the first embodiment, the detailed description thereof will be omitted.

According to the value of the resolution change flag extracted from the bitstream (for example, 1), the resolution identification flag is extracted and decoded from the bitstream using a resolution identification flag table excluding the resolution having the highest frequency among the same peripheral resolutions of the encoder.

Meanwhile, the resolution switching flag extractor 4850 extracts the resolution switching flag from the bitstream and excludes the resolution having the highest frequency among the peripheral resolutions according to the value of the resolution switching flag (for example, 1). Table), the resolution identification flag can be extracted and decoded in the bitstream. In addition, when using the resolution of the neighboring block, the resolution conversion flag is encoded as 0 when the resolution of the current block is the same as the resolution among the A and B resolutions as well as the highest frequency resolution, and the resolution conversion flag is set to 1 when the resolution is different. It can also be encoded. When the resolution switch flag is encoded as 1, the flag may be excluded from the resolution identification flag table.

Meanwhile, the resolution switch flag extractor 4850 calculates the resolution of the current block by extracting a resolution switch flag value (for example, 1) from the bitstream so that the resolution is different from the previous block A. FIG. For example, if the resolution set is 1/2, 1/4, the resolution to be converted is 1/4 if the resolution of the previous block is 1/2.

Meanwhile, the image encoding / decoding apparatus according to an embodiment of the present invention may be implemented by connecting the bitstream output terminal of the image encoding apparatus of FIG. 9 or 32 and the bitstream input terminal of the image decoding apparatus of FIG. 18 or 48.

An image encoding / decoding apparatus according to an embodiment of the present invention determines a motion vector resolution for each region or motion vector and inter-predictively encodes an image in units of regions using a motion vector according to the motion vector resolution determined for each region or motion vector. Image encoder; And an image decoder extracting the resolution information from the bitstream to restore the resolution and inter-prediction-decoding each region by using the motion vector according to the reconstructed motion vector resolution of each region or motion vector.

19 is a flowchart illustrating an image decoding method using adaptive motion vector resolution according to the first embodiment of the present invention.

In the image decoding method using the adaptive motion vector resolution according to the first embodiment of the present invention, a resolution change flag is extracted from a bitstream, and a resolution identification flag encoded from the bitstream is extracted and decoded according to the extracted resolution change flag. The motion vector resolution for each region or motion vector is reconstructed, and each region is inter predicted and decoded using the motion vector for each region according to the reconstructed each region or motion vector resolution.

To this end, the image decoding apparatus 1800 extracts a resolution change flag from the bitstream (S1910), and determines whether the extracted resolution change flag changes the motion vector resolution for each region (S1920). If the vector resolution is changed for each region or motion vector, the resolution identification flag is extracted and decoded from the bitstream to restore the motion vector resolution of each region or motion vector (S1930), and the motion vector of each restored region or motion vector. Inter-prediction decoding is performed by reconstructing the motion vector of each region or the motion vector at the resolution (S1940). In addition, when the resolution change flag indicates that the motion vector resolution is fixed without changing for each region or motion vector, the image decoding apparatus 1800 extracts, decodes, and restores the motion vector resolution flag from the bitstream (S1950). Inter-prediction decoding is performed on each region by reconstructing the motion vector according to the fixed motion vector resolution for the lower region defined in the header according to the motion vector resolution (S1960). Here, the motion vector resolution decoded for each region or motion vector may be different for each x component and y component of the motion vector.

Here, the image decoding apparatus 1800 decodes the resolution identification flag hierarchically coded in a quad tree structure by tying regions or motion vectors having the same motion vector resolution to restore motion vector resolution for each region or motion vector, or Decode the resolution identification flag encoded using the motion vector resolution predicted using the motion vector resolution of the surrounding area of the area to restore the motion vector resolution for each area, or the run and length of the motion vector resolution for each area or motion vector. Decodes the encoded resolution identification flag to restore the motion vector resolution for each region or motion vector, or decodes the resolution identification flag hierarchically encoded using the tag tree by using the tag tree.Restores the motion vector resolution for each direct vector, or decodes by changing the number of bits allocated to the resolution identification flag according to the frequency of the motion vector resolution for each region or motion vector, or the resolution identification flag decoded for each region or motion vector is estimated. In the case of the flag indicating that it is possible, the motion vector resolution is estimated according to a pre-determined estimation method, and the estimated motion vector resolution is restored as the motion vector resolution of the corresponding region or motion vector, and the resolution identification flag decoded for each region or motion vector In the case of a flag indicating that the estimation is impossible, the motion vector resolution indicated by the decoded resolution identification flag may be restored. In this case, the image decoding apparatus 1800 may identify an identifier indicating the maximum number of quad tree layers and the area size indicated by the lowest node of the quad tree layer from the header of the bitstream, the maximum number of tag tree layers, and the lowest node of the tag tree hierarchy. The identifier indicating the area size can be decrypted and restored.

In addition, the image decoding apparatus 1800 extracts and decodes the encoded difference vector from the bitstream, and decodes the difference vector for each region or motion vector by decoding according to the reconstructed region or the motion vector resolution for each motion vector. In addition, the prediction motion vector for each region or motion vector may be predicted, and the motion vector for each region may be reconstructed using the reconstructed difference vector and the prediction motion vector.

To this end, the image decoding apparatus 1800 may decode the difference vector using UVLC. In this case, the image decoding apparatus 1800 may decode using the K-th order exponential golem code, but according to the reconstructed region or the motion vector resolution for each motion vector. The order K of the exponential Gollum code can be changed. The image decoding apparatus 1800 may decode the difference vector by using context-based binary arithmetic coding. In this case, the image decoding apparatus 1800 may decode using the unary / K-order exponential golem combined code, but the motion vector for each reconstructed region or motion vector. The order (K) and the maximum value (T) of the unary / K-order exponential golem combined code can be changed and decoded according to the resolution. When decoding the difference vector using the CABAC, the image decoding apparatus 1800 may calculate and decode a cumulative probability differently according to the motion vector resolution for each reconstructed region or motion vector.

In addition, the image decoding apparatus 1800 predicts a predicted motion vector with respect to the motion vector of each region or the motion vector by using the motion vector of the surrounding region of each region, but the motion vector resolution of the surrounding region and the motion vector resolution of each region If is not equal, the motion vector resolution of the surrounding area may be predicted by converting the motion vector resolution of each area or the motion vector. The predicted motion vector may be obtained in the same manner in the image encoding apparatus and the decoding apparatus. Accordingly, various embodiments of deriving a predictive motion vector by the image encoding apparatus and embodiments of motion vector resolution conversion with reference to FIGS. 22 to 26 may be similarly implemented in the image decoding apparatus according to the present invention.

In addition, the image decoding apparatus 1800 may change the resolution identification flag decoding method according to the distribution of the motion vector resolution of each region or the motion vector resolution determined for each region or motion vector.

In addition, when the image decoding apparatus 1800 performs entropy decoding by arithmetic decoding, a method of generating a binarization bit string of a resolution identification flag based on a distribution of a motion vector resolution determined for each region or motion vector according to the distribution of the motion vector resolution of the peripheral region of each region. In addition, the arithmetic decoding and the probability update may be performed by differently applying the probability model according to the distribution of the motion vector resolution of the surrounding area and the probability of the motion vector resolution that has occurred up to now. In addition, arithmetic decoding and probability updating may be performed using different probability models according to bit positions.

When at least one of each region is a block and the block mode is a skip mode, the image decoding apparatus 1800 may convert and predict the motion vector resolution of the surrounding region to the highest resolution among the motion vector resolutions of the surrounding region.

51 is a flowchart illustrating a method of decoding an image using adaptive motion vector resolution according to a second embodiment of the present invention.

The video decoding method using the adaptive motion vector resolution according to the second embodiment of the present invention includes a resolution designation flag extraction step (S5102), a resolution decoding step (S5104), a differential vector decoding step (S5106), and an inter prediction decoding step ( S5108) and the resolution switching flag extraction step (S5110).

Here, the resolution designation flag extraction step S5102 is performed by the operation of the resolution designation flag extractor 4810, the resolution decoding step S5104 is performed by the resolution decoder 4820, and the differential vector decoding step S5106 is differential vector decoding. In operation 4830, the inter prediction decoding operation S5108 may correspond to the operation of the inter prediction decoder 4840, and the resolution switching flag extraction operation S5110 may correspond to an operation of the resolution conversion flag extractor 4850, respectively. Therefore, detailed description is omitted.

Meanwhile, the image encoding / decoding method according to an embodiment of the present invention may be implemented by combining the image encoding method according to the first or second embodiment of the present invention and the image decoding method according to the first or second embodiment. have.

An image encoding / decoding method according to an embodiment of the present invention determines a motion vector resolution for each region or motion vector, and inter-predictively encodes an image in units of regions using a motion vector according to the motion vector resolution determined for each region or motion vector. Image encoding step; And an image decoding step of extracting the resolution information from the bitstream, restoring the resolution, and inter-prediction-decoding each region by using a motion vector according to the reconstructed motion vector resolution of each region or motion vector.

As described above, according to an exemplary embodiment of the present invention, the motion vector resolution may be adjusted in units of an area or a motion vector of any size according to the characteristics of the image (for example, the complexity of the image or the degree of motion of the image). Since inter prediction encoding may be performed using a motion vector having a determined and adaptive motion vector resolution, the bit rate due to encoding may be reduced while improving the quality of an image, thereby improving compression efficiency. For example, in some pictures of an image, if some areas have a high degree of complexity but a small amount of motion, and other areas have a low degree of complexity but a large degree of motion, for some areas, the motion vector resolution may be increased for some areas. Predictive coding improves the accuracy of inter prediction, which reduces the residual signal and the amount of encoded bits. In addition, even if the motion degree is small and the resolution of the motion vector is increased, the amount of bits encoding the motion vector is not significantly increased, thereby improving image quality. In addition, the amount of encoded bits can be reduced. In addition, for other regions, the motion vector resolution is lowered so that even if the inter prediction encoding is performed, the motion vector resolution is small without significantly reducing the image quality, thereby reducing the amount of bits encoding the motion vector, thereby reducing the overall image quality without significantly reducing the image quality. The amount of coding bits can be reduced, so that the compression efficiency can be improved.

In the above description, it is described that all the components constituting the embodiments of the present invention are combined or operated in one, but the present invention is not necessarily limited to these embodiments. In other words, within the scope of the present invention, all of the components may be selectively operated in combination with one or more. In addition, although all of the components may be implemented as one independent hardware, each or some of the components of the program modules are selectively combined to perform some or all of the functions combined in one or a plurality of hardware It may be implemented as a computer program having a. Codes and code segments constituting the computer program may be easily inferred by those skilled in the art. Such a computer program may be stored in a computer readable storage medium and read and executed by a computer, thereby implementing embodiments of the present invention. The storage medium of the computer program may include a magnetic recording medium, an optical recording medium, a carrier wave medium, and the like.

In addition, the terms "comprise", "comprise" or "having" described above mean that the corresponding component may be inherent unless specifically stated otherwise, and thus excludes other components. It should be construed that it may further include other components instead. All terms, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art unless otherwise defined. Terms commonly used, such as terms defined in a dictionary, should be interpreted to coincide with the contextual meaning of the related art, and shall not be interpreted in an ideal or excessively formal sense unless explicitly defined in the present invention.

The above description is merely illustrative of the technical idea of the present invention, and those skilled in the art to which the present invention pertains may make various modifications and changes without departing from the essential characteristics of the present invention. Therefore, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention but to describe the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of the present invention.

As described above, the present invention is applied to an image processing field for compressing a video, and the image can be efficiently encoded by inter-prediction coding by adaptively changing a motion vector resolution in a predetermined area unit. It is a very useful invention that occurs.

CROSS-REFERENCE TO RELATED APPLICATION

This patent application is related to US patent law No. 119 (a) for patent application No. 10-2009-0077452 filed in Korea on August 21, 2009 and patent application No. 10-2010-0081096 filed on August 20, 2010. If priority is claimed under Article 35 (35 USC § 119 (a)), all of this is incorporated by reference in this patent application. In addition, if this patent application claims priority for the same reason as above for a country other than the United States, all the contents thereof are incorporated into this patent application by reference.

Claims

In the video encoding / decoding device,

An image encoder for determining a motion vector resolution for each region or motion vector and inter prediction encoding using a motion vector according to the motion vector resolution determined for each region or motion vector; And

The image decoder extracts the resolution information from the bitstream, restores the resolution, and inter-predicts and decodes the motion vector according to the motion vector resolution of each reconstructed region or motion vector.

Image encoding / decoding apparatus comprising a.
In the apparatus for encoding a video,

A resolution determiner for determining a motion vector resolution for each region or motion vector; And

An inter prediction encoder for inter prediction encoding using a motion vector according to a motion vector resolution determined for each region or motion vector

An image encoding apparatus comprising a.
The video encoding apparatus of claim 2,

And a resolution change flag generator for generating a resolution change flag indicating whether the motion vector resolution is determined for each region or motion vector.
The video encoding apparatus of claim 2,

And a resolution encoder for encoding a resolution identification flag indicating a motion vector resolution determined for each region or motion vector.
The video encoding apparatus of claim 2,

And further comprising a difference vector encoder for encoding a difference vector that is a difference between a motion vector and a predicted motion vector according to the motion vector resolution determined for each region or motion vector, wherein the difference vector encoder determines the difference vector according to the motion vector resolution. An image encoding device, characterized in that for encoding differently.
The method of claim 4, wherein

The resolution identification flag,

And at least one of a resolution of the motion vector and a resolution of the differential motion vector.
The method of claim 5,

The difference vector encoder is

When encoding a differential motion vector, a reference resolution is determined, a motion vector having a resolution other than the reference resolution is converted to a motion vector having a reference resolution, and the differential motion vector is obtained by using a reference prediction motion vector obtained from the reference motion vector of the neighboring block. And a video encoding apparatus.
The method of claim 5,

The difference vector encoder is

And a reference resolution flag is further encoded when the resolution of the reference resolution and the motion vector are different.
The method of claim 2,

The video encoding apparatus,

And a resolution specifying flag generator for differently specifying a resolution set for each region or motion vector of the image.
The method of claim 2,

The resolution determiner,

And when the resolution of the current block is known using the reference picture index, information about the resolution is encoded by encoding the reference picture index without generating a resolution identification flag.
The method of claim 9,

The resolution designation flag generator,

And a resolution set is set differently for each reference picture according to a predetermined reference without encoding a resolution designation flag when a plurality of reference pictures are used.
The method of claim 11,

The preset criteria,

And a distance value between the reference picture and the current picture, an error value between resolutions measured using error measuring means, and a table in which a resolution set is set for each reference picture.
The method of claim 11,

The resolution designation flag generator,

And a resolution set differently for a picture to be used as a reference picture and a picture not to be used as a reference picture.
The method of claim 2,

The video encoding apparatus,

And a resolution change flag generator for generating a resolution change flag indicating a change in the peripheral resolution of the region to be encoded with the current resolution or the resolution of the previous block.
In the apparatus for decoding an image,

A resolution change flag extractor for extracting a resolution change flag from the bitstream;

A resolution decoder which extracts and decodes an encoded resolution identification flag from the bitstream according to the extracted resolution change flag to restore a motion vector resolution for each region or motion vector; And

An inter prediction decoder for performing inter prediction decoding on the reconstructed motion vector resolution of each region.

Video decoding apparatus comprising a.
The video decoding apparatus of claim 15,

And extracting and decoding the encoded difference vector from the bitstream, and decoding the encoded difference vector according to the reconstructed region or the motion vector resolution for each motion vector to reconstruct the difference vector for each region. And a predictive decoder predicts a predicted motion vector for each region and reconstructs the motion vector for each region by using the reconstructed difference vector and the predicted motion vector.
In the apparatus for decoding an image,

A resolution decoder for reconstructing resolution for each region or motion vector using information received from the bitstream; And

An inter prediction decoder that performs inter prediction decoding using motion vectors according to the resolutions of the reconstructed regions

Video decoding apparatus comprising a.
The image decoding apparatus of claim 17,

Further comprising a resolution specifying flag extractor for extracting a resolution specifying flag from the bitstream,

And the resolution decoder restores the resolution according to the extracted resolution designation flag.
The method of claim 18,

The resolution designation flag is,

And a flag indicating that the image is fixed at a single resolution or a resolution set having one or more resolutions.
The method of claim 18,

The resolution designation flag extractor,

When the plurality of reference pictures are used, the image decoding apparatus characterized in that different resolution sets are set for each reference picture according to a predetermined reference.
The method of claim 20,

The preset criteria,

And a distance value between the reference picture and the current picture, an error value between resolutions measured using error measuring means, and a table in which a resolution set is set for each reference picture.
The method of claim 18,

The resolution decoder,

And a resolution set for a reference picture index.
The method of claim 17,

The video decoding apparatus,

And a resolution change flag extractor for extracting a resolution change flag from the bitstream and extracting a resolution identification flag according to a value of the resolution change flag.
The method of claim 23, wherein

The resolution decoder is

Extracting a resolution identification flag according to the resolution type of the resolution conversion flag extracted and decoded from the header of the bitstream, and extracting a value of a differential motion vector corresponding to the corresponding resolution by referring to a code number extracted from a code number table using the resolution identification flag And a video decoding apparatus.
A method of encoding / decoding an image, the method comprising: an image encoding step of determining a motion vector resolution for each region or motion vector and inter prediction encoding using a motion vector according to a motion vector resolution determined for each region or motion vector; And

Image decoding step of extracting the resolution information from the bitstream to restore the resolution and inter-prediction decoding using the motion vector according to the motion vector resolution of each reconstructed region or motion vector

Image encoding / decoding method comprising a.
In the method of encoding an image,

Determining a motion vector resolution for each region or motion vector; And

Inter-prediction encoding using a motion vector according to the motion vector resolution determined for each region;

Image encoding method comprising a.
The method of claim 26, wherein the motion vector resolution,

The video encoding method of claim 1, wherein the x component and the y component of the motion vector are different.
The method of claim 26, wherein the video encoding method,

And generating a resolution change flag indicating whether a motion vector resolution is determined for each region or motion vector.
The method of claim 26, wherein the video encoding method,

And encoding a motion vector resolution determined for each of the regions or the motion vectors.
The method of claim 29, wherein the encoding of the motion vector resolution comprises:

And encoding a motion vector resolution determined for each region or motion vector hierarchically and quadratically encoding regions having the same motion vector resolution.
The method of claim 30, wherein the encoding of the motion vector resolution comprises:

And encoding an identifier indicating a maximum number of the quad tree layers and an area size indicated by the lowest node of the quad tree layer.
The method of claim 29, wherein the encoding of the motion vector resolution comprises:

And encoding the motion vector resolution determined for each region or motion vector using the motion vector resolution predicted using the motion vector resolution of the surrounding area of each region.
The method of claim 29, wherein the encoding of the motion vector resolution comprises:

And a motion vector resolution determined for each region or motion vector using run and length.
The method of claim 29, wherein the encoding of the motion vector resolution comprises:

And encoding the motion vector resolution determined for each region or motion vector hierarchically using a tag tree.
The method of claim 29, wherein the encoding of the motion vector resolution comprises:

And encoding the number of bits allocated to the motion vector resolution according to the frequency of the motion vector resolution determined for each region or motion vector.
The method of claim 29, wherein the encoding of the motion vector resolution comprises:

It is determined whether or not the motion vector resolution determined for each region or motion vector can be estimated according to a pre-scheduled estimation scheme by the image decoding apparatus, and an identifier indicating that the motion vector resolution can be estimated is encoded. And encoding a motion vector resolution and an identifier indicating that the motion vector resolution cannot be estimated in the region where the motion vector resolution cannot be estimated.
The method of claim 29, wherein the encoding of the motion vector resolution comprises:

And encoding a different number of bits according to the distribution of the motion vector resolution of the surrounding area of each region.
The method of claim 29, wherein the encoding of the motion vector resolution comprises:

When encoding using arithmetic coding, the motion vector resolution determined for each region or motion vector is encoded using a method of generating a binarized bit string according to the distribution of the motion vector resolution of the peripheral region of each region. Encoding method.
The method of claim 29, wherein the encoding of the motion vector resolution comprises:

When encoding using arithmetic coding, the motion vector resolution determined for each region or motion vector is arithmetic encoded using a different probability model according to the distribution of the motion vector resolution of each region and the occurrence probability of the motion vector resolution. The video encoding method, characterized by updating the probability model.
The method of claim 29, wherein the encoding of the motion vector resolution comprises:

And a motion vector resolution determined for each region or motion only when each component of the difference vector is not '0'.
27. The method of claim 26, wherein determining the motion vector resolution comprises:

If the size of the predicted motion vector or the difference vector of the motion vector according to the motion vector resolution determined for each region or motion vector is larger than a threshold, determining the motion vector resolution determined for each region or motion vector as a predetermined value. An image encoding method.
The method of claim 26, wherein the video encoding method,

And encoding a difference vector that is a difference between a motion vector and a predicted motion vector according to the motion vector resolution determined for each region or motion vector, wherein the difference vector is encoded differently according to the motion vector resolution. The video encoding method.
43. The method of claim 42, wherein encoding the difference vector comprises:

In the case of encoding the difference vector using universal variable length coding, the difference vector is encoded using a K-order exponential golem code, and the order K of the exponential golem code is changed according to the motion vector resolution determined for each region. An image encoding method.
43. The method of claim 42, wherein encoding the difference vector comprises:

In the case of encoding the difference vector using context-based binary arithmetic coding, encoding is performed using a unary / K-order exponential golem combined code, but the unary / K-order exponential golem combined according to the motion vector resolution determined for each region. And encoding and changing the order (K) and the maximum value (T) of the code.
43. The method of claim 42, wherein encoding the difference vector comprises:

When encoding the difference vector using context-based binary arithmetic coding, the image encoding method according to the present invention is characterized by calculating and encoding a cumulative probability differently according to a motion vector resolution determined for each region.
43. The method of claim 42, wherein encoding the difference vector comprises:

The predicted motion vector for the motion vector of each area is predicted using the motion vector of the surrounding area of each area, but when the motion vector resolution of the surrounding area and the motion vector resolution of each area are not the same, the surrounding area And converting the motion vector resolution of the motion vector resolution into the motion vector resolution of each region to predict the motion vector resolution.
47. The method of claim 46, wherein encoding the difference vector comprises:

When at least one of each region is a block and the block mode of the block is a skip mode, the motion vector resolution of the region to be predicted is predicted by the highest resolution among the motion vector resolutions of the surrounding region. Image coding method.
The method of claim 26, wherein the inter prediction encoding is performed.

Determining a filter coefficient that minimizes an error between a current picture and a reference picture for each of the motion vector resolutions, using interpolation filters having the determined filter coefficients, and inter-prediction-encoding using different reference pictures for each motion vector resolution; And encoding the determined filter coefficients.
The method of claim 42,

In the encoding of the difference vector,

When encoding a differential motion vector, a reference resolution is determined, a motion vector having a resolution other than the reference resolution is converted to a motion vector having a reference resolution, and the differential motion vector is obtained by using a reference prediction motion vector obtained from the reference motion vector of the neighboring block. The video encoding method, characterized in that for calculating.
The method of claim 42,

In the encoding of the difference vector,

And a reference resolution flag is further encoded when the resolution of the reference resolution and the motion vector are different.
The method of claim 26,

The video encoding method,

And a resolution designation flag generation step for differently designating a resolution set for each region or motion vector of the image.
The method of claim 26,

In the determining of the resolution,

And when the resolution of the current block is known using the reference picture index, information about the resolution is encoded by encoding the reference picture index without generating a resolution identification flag.
The method of claim 51,

In the resolution designation flag generation step,

And a resolution set is set differently for each reference picture according to a predetermined standard without encoding resolution designation flags when a plurality of reference pictures are used.
The method of claim 53,

The preset criteria,

And a distance value between the reference picture and the current picture, an error value between resolutions measured using error measuring means, and a table in which a resolution set is set for each reference picture.
The method of claim 53,

In the resolution designation flag generation step,

And a resolution set differently for a picture to be used as a reference picture and a picture not to be used as a reference picture.
The method of claim 26,

The video encoding method,

And a resolution conversion flag generation step of generating a resolution conversion flag informing of a change in the peripheral resolution of the region to be encoded with the current resolution or the resolution of the previous block.
In the method of decoding an image,

Extracting a resolution change flag from the bitstream;

Extracting and decoding a resolution identification flag encoded from the bitstream according to the extracted resolution change flag to restore motion vector resolution for each region or motion vector; And

Inter-prediction decoding using a motion vector according to the motion vector resolution of each reconstructed region or motion vector

Image decoding method comprising a.
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

And decoding a resolution identification flag hierarchically coded in a quad tree structure by tying regions or motion vectors having the same motion vector resolution and restoring the motion vector resolution for each region or motion vector.
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

And decoding an identifier indicating a maximum number of the quad tree layers and an area size indicated by the lowest node of the quad tree layer.
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

An image comprising: decoding the encoded resolution identification flag by using a motion vector resolution predicted by using a motion vector resolution of each region or a surrounding region of the motion vector to restore a motion vector resolution for each region or motion vector Decryption method.
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

And reconstructing the motion vector resolution for each region by decoding a resolution identification flag encoded with a run and a length of the motion vector resolution for each region or motion vector.
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

And reconstructing the motion vector resolution of each region or motion vector by decoding a resolution identification flag hierarchically encoded using a tag tree.
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

And decoding the number of bits allocated to the resolution identification flag according to the frequency of the motion vector resolution for each region or motion vector.
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

In the case of a flag indicating that the resolution identification flag decoded for each region or motion vector can be estimated, the motion vector resolution is estimated according to a pre-determined estimation method, and the estimated motion vector resolution is restored as the motion vector resolution of the corresponding region. Or if the resolution identification flag to be decoded for each motion vector is a flag indicating that it is impossible to estimate, reconstructing the motion vector resolution indicated by the decoded resolution identification flag as a motion vector of a corresponding region.
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

And reconstructing the motion vector resolution for each region or motion vector only when each component of the difference vector is not '0'.
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

And decoding the resolution identification flag by allocating a different number of bits according to the distribution of the motion vector resolution of the surrounding area of each area.
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

When the resolution identification flag is decoded using the arithmetic coding, the resolution identification flag is decoded by using a method of generating a binary bit stream according to a distribution of motion vector resolutions of surrounding regions of each region. .
59. The method of claim 57, wherein restoring the motion vector resolution comprises:

When the resolution identification flag is decoded using the arithmetic coding, the resolution identification flag is decoded using a different probability model according to the distribution of motion vector resolution of the surrounding area of each region and the occurrence probability of the motion vector resolution. An image decoding method.
The method of claim 57, wherein the video decoding method,

Extracting and decoding coded filter coefficients for each motion vector resolution from the bitstream, and reconstructing filter coefficients for each motion vector resolution, wherein the inter-prediction decoding comprises performing the filter for each reconstructed motion vector resolution. A reference picture is interpolated using coefficients, and inter prediction decoding is performed using a different reference picture for each motion vector resolution.
The method of claim 57, wherein the video decoding method,

Extracting and decoding an encoded difference vector from the bitstream, and decoding the encoded difference vector according to the reconstructed region or the motion vector resolution for each motion vector to restore the difference vector for each region, wherein the inter prediction decoding is performed. And predicting the predicted motion vector for each region and reconstructing the motion vector using the reconstructed difference vector and the predicted motion vector.
71. The method of claim 70, wherein restoring the difference vector comprises:

In the case of decoding the difference vector using universal variable length coding, the difference vector is decoded using a K-order exponential golem code, and the order of the K-th exponential gollum code according to the motion vector resolution for each reconstructed region or motion vector ( K) is changed.
72. The method of claim 71 wherein the reconstructing the difference vector comprises:

When decoding the difference vector using context-based binary arithmetic coding, decoding is performed using a unary / K-order exponential golem combined code, but the unary / K-difference according to the motion vector resolution for each reconstructed region or motion vector. And decoding (K) and maximum value (T) of an exponential golem combined code.
71. The method of claim 70, wherein restoring the difference vector comprises:

When decoding the difference vector using context-based binary arithmetic coding, cumulative probability is calculated and decoded differently according to the motion vector resolution for each reconstructed region or motion vector.
73. The method of claim 70, wherein predicting the predictive motion vector comprises:

When the predicted motion vector of the motion vector of each area is predicted by using the motion vector of each area or the surrounding area of the motion vector, but the motion vector resolution of the surrounding area and the motion vector resolution of each area are not the same. And predicting the motion vector resolution of the surrounding area by converting the motion vector resolution into the motion vector resolution of each area or motion vector.
75. The method of claim 74, wherein predicting the predictive motion vector comprises:

If at least one of the regions is a block and the block mode of the block is the skip mode, the image is predicted by converting the motion vector resolution of the surrounding region to the highest resolution among the motion vector resolutions of the surrounding region. Decryption method.
In the method of decoding an image,

A resolution decoding step of reconstructing the resolution for each region or motion vector using information received from a bitstream; And

An inter prediction decoding step of performing inter prediction decoding using a motion vector according to the resolution of each reconstructed region

Image decoding method comprising a.
The method of claim 76, wherein the image decoding method,

A resolution specifying flag extraction step of extracting a resolution specifying flag from the bitstream;

In the resolution decoding step, the resolution is reconstructed according to the extracted resolution specification flag.
78. The method of claim 77 wherein

The resolution specifying flag extraction step,

And using a plurality of reference pictures, different resolution sets for different reference pictures according to preset criteria.
77. The method of claim 76,

The video decoding method,

And a resolution conversion flag extraction step of extracting a resolution conversion flag from the bitstream and extracting a resolution identification flag according to a value of the resolution conversion flag.
The method of claim 79,

The resolution decoder is

Extracting a resolution identification flag according to the resolution type of the resolution conversion flag extracted and decoded from the header of the bitstream, and extracting a value of a differential motion vector corresponding to the corresponding resolution by referring to a code number extracted from a code number table Image decoding method characterized in that.