KR100483674B1 - Loss-Encoding Method for Encoding Flexible Shape Information - Google Patents

Abstract

The present invention relates to a SHAPE INFORMATION CODING METHOD for separating and processing a moving image applied through an image input apparatus into a target image and a background image having arbitrary shape information. In order to enable lossy coding using SCAN INTERLEAVING during SCALABLE SHAPE CODING, the low resolution base layer image can be encoded into the high resolution enhancement layer image using the scan interleaving method. In this case, the coding efficiency is improved by changing and applying the probability table of the arithmetic coding method.

Description

Lossy Coding Method for Coding Flexible Shape Information

The present invention relates to a SHAPE INFORMATION CODING METHOD for separating and processing a moving image applied through an image input apparatus into a target image and a background image having arbitrary shape information. During the SCALABLE SHAPE CODING process, SCAN INTERLEAVING is used to modify the probability table used in arithmetic coding to enable lossy coding.

As is well known, the recent image processing technology breaks away from a method of compressing and encoding an image corresponding to one frame (FRAME) as a whole, and divides it into predetermined unit blocks having arbitrary shape information and compresses and encodes each of them. In the direction of transmission.

That is, the applied image is separated into each object image and the background image, and the compression efficiency and the encoding efficiency are intended by transmitting only the change of the object image, thereby preparing an international standard.

For example, WG11 under ISO / IEC, a world standardization organization, has arbitrary shape information unlike MPEG (MOVIGN PICTURE EXPERTS GROUP). MPEG-4, which is a standardization method for encoding an object, is in progress, and MPEG-4, which is being standardized, is a unit block having arbitrary shape information. The concept of VOP (VIDEO OBJECT PLANE) is called. Is based on.

Here, the VOP is defined as a quadrangle that divides the applied image into a background image and each object image, and includes the separated background image and the object image. In MPEG-4, a predetermined object or a predetermined object is included in the image. If there is a region of the object consisting of the region of, the video of the object is divided into respective VOPs, and the separated VOPs are encoded respectively.

The VOP has the advantage of freely synthesizing or decomposing a natural image or an artificial image as a unit of an object image, and is fundamental to processing images of an object in the field of computer graphics and multimedia.

Fig. 1 shows the configuration of an encoder (ENCODER) 100, which is primarily determined by the International Standards Organization (ISO / IEC JTC1 / SC29 / WG11 MPEG96 / N1172 JANUARY). The block diagram shown.

Here, when the VOP formation unit 100 receives a video sequence to be transmitted or stored, the VOP formation unit 100 divides it into a target image unit to form different VOPs.

2 shows an example in which one VOP is formed by setting an image of a cat as an object image.

Here, the horizontal size of the VOP is defined by the width of the VOP, the vertical size is defined by the height of the VOP, the formed VOP is M and N respectively in the X-axis and Y-axis with the upper left as the grid (GRID) starting point. It is partitioned into an MxN macroblock having two pixels. For example, it is divided into 16x16 macroblocks each having 16 pixels on the X-axis and the Y-axis.

In this case, when there are not M and N pixels of the X and Y axes of the macroblock formed on the right and the bottom of the VOP, the size of the VOP is extended to M and both of the X and Y axes pixels of each macro block are expanded. And N.

In addition, M and N are set to an even number in order to perform encoding in units of sub-blocks in a text encoding unit to be described later.

On the other hand, each VOP formed in the VOP forming unit 110 is input to the VOP encoding unit (120a, 120b, ..., 120n) is encoded for each VOP, multiplexed by the multiplexer 13 into a bit string (BIT STREAM) Is sent.

FIG. 4 is a block diagram showing the configuration of the VOP encoders 120a, 120b, ..., 120n of the VM encoder 100 determined primarily by the international standard subdivision.

First, when a VOP for each object image formed by the VOP forming unit 110 is input to a motion estimation unit 121, the motion estimation unit 121 moves in units of macro blocks from an applied VOP. Will be estimated.

In addition, motion information estimated by the motion estimation unit 121 is input to a motion compensation unit 122 to compensate for motion.

In addition, the VOP whose motion is compensated by the motion compensator 122 is input to the subtractor 123 together with the VOP formed by the VOP forming unit 110 to detect a difference value, and the difference detected by the subtractor 123. The value is input to the object internal encoding unit 124 to encode the internal information of the object in units of sub blocks of the macro block.

For example, after the X and Y axes of the macroblock are subdivided into 8x8 subblocks having 8 pixels each with M / 2 x N / 2, the object internal information is encoded.

Meanwhile, the VOP whose motion is compensated by the motion compensator 122 and the internal information of the object encoded by the object internal encoding unit 124 are input to the adder 125 and added, and the output of the adder 125 is added. The signal is input to a PREVIOU RECONSTRUCTED VOP 126 to detect a previous VOP which is a VOP of the image immediately before the current image.

In addition, the previous VOP detected by the previous VOP detector 126 is input to the motion estimation unit 121 and the motion compensation unit 122 and used for motion estimation and motion compensation.

The VOP formed by the VOP forming unit 110 is input to a SHAPE CODING BLOCK 127 to encode shape information.

Here, the output signal of the shape coding unit 127 is used or not depending on the field to which the VOP coders 120a, 120b, ..., 120n are applied. As shown by a dotted line, the shape coding unit ( The output signal of 127 may be input to the motion estimation unit 121, the motion compensator 122, and the object internal encoding unit 124, and used to encode motion estimation, motion compensation, and internal information of the object.

In addition, motion information estimated by the motion estimation unit 121, shape information encoded by the object encoding unit 124, and shape information encoded by the shape encoding unit 127 are applied to the multiplexer 128. After multiplexing, the signal is output to the multiplexer 130 of FIG. 1 through the buffer 9129 and transmitted in the bit string.

FIG. 3 is a block diagram showing the configuration of a VM decoder (DECODER) 200, which is determined primarily by an international standard subdivision.

The encoded signal of the VOP, which is encoded through the VM encoder 100 and transmitted in a bit string, is separated for each VOP in the demultiplexer 210 of the VM decoder 200.

In addition, the separated VOP coded signals are decoded by the VOP recorders 120a, 120b, ..., 120n, respectively, and the decoded signals output from the VOP decoders 120a, 120b, ..., 120n are synthesized by the synthesizer 230. ) Is synthesized and output as the original image.

In such MPEG-4, as a technique applied to the shape encoder 127 for encoding each VOP transmitted from the VOP forming unit 110, N × N blocks (N = 16, 8, 4) MMR SHAPE CODING TECHNIQUE, which encodes shape information based on the VERIFEX-BASED SHAPE CODING TECHNIQUE, which encodes shape information based on the vertices, and the baseline. BASELINE-BASED SHAPE CODING TECHNIQUE and CONTEXT-BASED ARITHMETIC CODING.

On the other hand, the stretched shape information encoding technology converts an image having a high amount of information (still image or a video) into a low resolution image and transmits it, and then converts it to an original image having a high resolution. It is a technology to reduce.

That is, when the encoder converts a high resolution image into a low resolution image having a small amount of information, encodes the image, and then transmits the encoded image to the decoder, the decoder decodes the low resolution image transmitted by the decoder to restore the high resolution image. The encoding process and the decoding process will be described as follows.

In the present description, for convenience of description, as shown in FIGS. 5 and 6, an image of a 4 × 4 enhancement layer is converted into an image of a base layer, encoded, transmitted, and then, again, is scanned by a scan interleaving method. The process of encoding the image of the health layer will be described in detail as follows.

In the encoder, the image of the 4x4 enhancement layer as shown in Fig. 5A is divided into four sections, and the pixels A5, A7, A13, and A15 at the lower right side are selected. Create an image of the base layer with a lossy encoding or lossless encoding and transmit the image to the decoder.

In addition, when the image of the base layer is extracted, the image of the base layer (reduced image) is compared with the image of the enhancement layer (original image), and the transitional sample data (TRANSITIONAL SAMPLE DATA: hereinafter referred to as TSD). And EXCEPTIONAL SAMPLE DATA (hereinafter referred to as ESD). In this case, when a TSD or ESD is detected, the context-based arithmetic encoding (CONTEXT-BASED ARITHMETIC ENCODING: hereinafter referred to as CAE) is transmitted to the decoder along with the image of the base layer.

Here, the reason for detecting and encoding the TSD and ESD is as follows.

7 shows a horizontal search (HORIZONTALSCANNING) in the process of comparing the image of the base layer and the ingestion image. In the figure, a reference scan line (hereinafter referred to as RSL) is a representation of a base layer image. A pixel column including the horizontal pixels, and a code scan line (hereinafter referred to as CSL) represents a horizontal pixel column of an enhancement layer image that is excluded from the base layer image extraction process. In addition, in order to encode the CSL, an RSL existing under and above the pixel to be encoded is used. As shown in FIG. 6, it can be seen that there are certain rules.

That is, when two neighboring pixels on the RSL existing under and above the pixel to be encoded on the CSL have the same value, the pixel values of the current position to be encoded may have the same value. If the pixel value at the current position is also the same as the two neighboring pixel values, no encoding is performed.

However, when two neighboring pixel values are different, the pixel values of the current position are likely to be different. Therefore, the pixel values of the current position to be encoded should be encoded. In this case, the transitional sample (TRANSITIONAL SAMPLE) ( Ellipses in solid lines).

In addition, when two neighboring pixel values are the same but the pixel values of the current position are different, encoding should be performed. This case is called an EXCEPTIONAL SAMPLE (ellipse indicated by a dotted line in FIG. 6).

Accordingly, in order to encode the enhancement layer, two types of data, that is, TSD and ESD, may exist. When the TSD and ESD exist, information about the TSD and ESD must be transmitted to the decoder to display the image of the enhancement layer. It can be restored accurately. CAE is a technique used to reduce the number of bits required when encoding the information on the TSD and ESD.

On the other hand, the decoder decodes the coded signal for the base layer image transmitted from the encoder and the coded signal for TSD or ESD, thereby reconstructing the image of the enhancement layer, which is a horizontal search and a vertical search (VERTICAL). SCANNING) divided in detail as follows.

FIG. 6A is a view for explaining a horizontal direction search method. When the pixel at the current X (A10) point is encoded, the pixels adjacent to the X point are A5, A6, A7, as shown in the figure. A13, A14, A15 and BO (A9). In this case, the values of A5, A6, A7, A13, A14, A15, and BO pixels are obtained by the scan interleaving method of the previous step, and the pixel of X point to be obtained is A5, A6, A7, A13, A14. The values of, A15 and B0 are obtained by performing CAE. That is, CAE is encoded by using a CONTEXT of seven pixels in the horizontal and vertical directions.

Similarly, in the vertical search method, as shown in Fig. 6B, the pixel value of Y point to be obtained is obtained from the pixel values of C0, C1, C2, A5, A7, A13, and A15.

Here, the pixels BO, CO, and C2 represent values obtained by the horizontal direction search method, the C1 pixels represent values obtained by the vertical direction search method, and the hatched pixels represent pixels to be next obtained.

Therefore, when the horizontal direction search is performed, the pixels in the vertical direction are doubled, and when the vertical search is performed, the pixels in the horizontal direction are doubled, resulting in a 4 × 4 enhancement layer image from the 2 × 2 base layer image. Will be restored.

In addition, in the case of a video, MPEG-4 supports elastic shape information encoding. In the case of MPEG-4, as shown in FIG. 8, the I-VOP (INTRA VIDEO) of the base layer video is displayed due to its video transmission characteristics. Reconstruction of the enhancement layer image using OBJECT PLANE: (hereinafter referred to as I-VOP), and P-VOP (PREDICTED VIDEO OBJECT PLANE: hereafter referred to as P-VOP). There is a method of restoring an image and a method of restoring an image of an enhancement layer using B-VOP (BY-DIRECTION VIDEO OBJECT PLANE).

On the other hand, one of the biggest features of MPEG-4 is to process based on the object (OBJECT). In other words, one image can be divided into several objects and each object can be individually encoded and processed. So you need to know the shape information to make an object. The shape information referred to here is commonly referred to as a mask. In the image, the object part is represented by '1' and the outside part of the object (background part) is represented by '0'. This shape information can be used to obtain an object from the image. Since the decoder decodes the object part using the shape information, the shape information should be encoded and transmitted to the decoder side.

CAE is currently used to encode shape information in MPEG-4. CAE uses a context template as shown in FIG. 9 in each pixel in 16 × 16 BAB (BINARY ALPHA BLOCK) units to obtain a context number as shown in Equation [1]. When it occurs, the probability of occurrence of '0' and the probability of occurrence of '1' are obtained, and then the encoding is performed using the arithmetic coding method using the obtained probability.

C = SIGMA CK × 2K -------- [1] In order to perform lossy coding, the size of BAB is reduced and reduced by CR (CONVERSION RATIO: CR) before arithmetic coding. When the BAB is restored to its original size, if the absolute difference between the original BAB and the restored BAB is less than or equal to the threshold, the CAB is encoded using the reduced size and if it is larger than the threshold, the original BAB is encoded by the CAE. CR is a variable that changes the size of BAB. MxM BAB is reduced to (MxCR) x (MxCR). Therefore, when CR = 1/2, the number of pixels to be encoded is reduced to 1/4. Therefore, as shown in FIG. 2, the CRB is used to reduce the size of the BAB, and when the absolute value of the difference between the restored BAB and the original BAB is obtained, the CAE is performed using the reduced BAB when the error is less than the threshold.

However, in the conventional CAE, the context number is used to obtain a context number, and when the context is generated using the obtained context number, the probability of occurrence of '0' and the probability of occurrence of '1' are calculated. There is a problem that the coding efficiency is lowered due to arithmetic coding using the obtained probability.

An object of the present invention is to solve the above-mentioned conventional problems. In particular, since the probability table used for arithmetic coding is modified, loss in coding of elastic shape information enabling loss coding using scan interleaving is possible. The present invention provides a coding method.

In order to achieve the above object, according to the present invention, the lossy coding method of the flexible shape information encoding includes separating a moving picture applied through an image input device into respective object images and background images having arbitrary shape information, and then resolving the resolution. In the flexible shape information encoding method which is transmitted and decoded to a plurality of different layers, a probability table of the arithmetic encoding method when a low resolution base layer image is encoded into a high resolution enhancement layer image by using a scan interleaving method It is characterized by the methodological configuration that is applied to change.

Hereinafter, the configuration and operation of the present invention will be described in detail with reference to the accompanying drawings, in accordance with an embodiment of the present invention.

First, in order to lossy encode shape information encoding, the present embodiment uses a method of applying a probability table of CAE.

First, if the inside of the object is 1 and the outside of the object is 0 in the standard probability table used in CAE, the probability table is newly created so that the larger value is 1 and the smaller value is 0. In the case of using the modified probability table, all the probabilities are 1 or 0, so if the context number is known without additional coding, the pixel value of the position to be encoded can be immediately known using the probability table.

That is, for example, if the probability table is

In the conventional case, the maximum value of the probability table is 65535 (= 2 ^16-1 ), which is a power of 2, which represents the probability that the TSD and ESD to be encoded are inside the object. If the value of the referenced index is greater than 1/2 of the maximum value of the probability table, it is determined to be inside the object, and if smaller than 1/2, it is determined to be outside of the object. However, in the present invention, as described above, it is not necessary to calculate whether the value corresponding to the reference index is larger or smaller than 1/2, but if it is larger than 1/2 by modifying the above-described probability table, it will naturally be inside the object. If '1' is smaller than 1/2, the probability table can be modified by replacing the corresponding value with '0' to directly refer to whether the object is inside or outside the object.

Therefore, if the error between the original BAB information to be encoded by the restored BAB using the corrected probability table is less than or equal to the threshold, only the information indicating that the modified probability table is used can be loss encoded to improve efficiency. have.

The threshold used in this method is to divide one BAB into 4 × 4 subblocks (SUB-BLOCK) and set a threshold of 16 * alphaTH for each subblock. If all subblocks meet the threshold, the modified probability table is used.

This method is equally applicable to encoding P and B-VOP as well as I-VOP.

In other words, if I-VOP is used, a new probability table for 210 contexts is created, and for P and B-VOP, a new probability table for 29 contexts is created.

On the other hand, apart from the above embodiment, the method proposed in this embodiment can be combined with the existing lossy coding method using the existing CR.

That is, as shown in FIG. 11, when the CR error is smaller than the threshold value when applying the CR, the CAB is performed after reducing the BAB size according to the CR. In this example, the modified probability table is first applied before performing the CAE. When the BAB is newly reconstructed, if the error is smaller than the threshold, only the information of using the CR and the modified probability table is transmitted. If the error is larger than the threshold, the CR and CAE information is transmitted. This is shown in FIG. In this case, the threshold of whether or not to use the modified probability table uses 16 * alphaTH * CR2.

In addition, if the error when using the modified probability value is smaller than the threshold value before converting the image size using the CR, the information indicating that the modified probability value is used is transmitted. Encoding can be done.

In addition, if the error when the corrected probability value is smaller than the threshold value before switching the size of the image using the CR is sent, the information indicating that the corrected probability value is used, and when the error is large, the above method is used. Loss coding can be performed.

In addition, in order to send additional information on whether the modified probability table is used, the following three ((8), (9), ( 10)) is added.

(1) MVDs == 0 No Update (2) MVDs! = 0 No Update (3) all_0 (4) all_255 (5) intraCAE (6) MVDs == 0 interCAE (7) MVDs! = 0 interCAE (8) intraCAE with modified probability table (9) MVDs == 0 interCAE with modified probability table (10) MVDs! = 0 interCAE with modified probability table (I), (4), (5), (8) ) Only.

The shape information encoding mode may be encoded using a variable length code.

In addition, instead of setting the shape information encoding mode differently, there is a method of adding additional information on whether the modified probability table is used in the CAE process for each BAB.

In this case, the shape information mode is encoded with a variable length code considering the method of expressing additional information about the BAB ((5), (6), (7)) performing CAE among seven types and the added coding mode. There is this.

In addition, the CAE algorithm used for encoding the TSD when encoding the enhancement layer image in the flexible encoding process uses the modified probability table, and if the error is less than or equal to the threshold, the modified probability without performing arithmetic coding By sending only the additional information that the table is used, the image of the enhancement layer can be encoded.

As described above, the lossy coding method of the flexible shape information encoding according to the present invention uses the modified probability table that changes all the probabilities to 1 or 0, especially when performing shape information loss coding using CAE. If is not exceeded the threshold is effective to improve the coding efficiency by transmitting only the additional information that the modified probability table is applied without encoding.

1 is a block diagram showing the configuration of a VM encoder primarily determined by the International Standards Organization.

2 is a diagram illustrating a VOP having shape information divided into macro blocks;

3 is a block diagram showing the configuration of a VM decoder primarily determined by an international standard subdivision;

4 is a block diagram showing a configuration of a VOP encoder determined primarily by an international standard computing instrument;

5 to 7 are diagrams for explaining a scan interleaving method in flexible encoding;

8 is a diagram for explaining spatial stretching coding;

9 (a) shows an intra context template,

9 (b) shows an inter-context template,

10 is a diagram showing a size switching relationship;

11 is a signal flow diagram illustrating a CR determination process;

12 is a signal flow diagram illustrating a shape information encoding process when a modified probability value of the present invention is applied.

Explanation of symbols for main parts of the drawings

10: VM encoder 11: VOP forming unit

12A, 12B,... , 12N: VOP coder 13, 38: multiplexer

20: VM decoder 21: Demultiplexer

22A, 22B,... 22N: VOP Recorder 23: Synthesis

31: Motion tracking 32: Motion compensation

33: Subtractor 34: Object internal coding unit

35: Adder 36: Old VOP detector

37: Shape coding department

Claims (16)

A stretchable shape information encoding method of separating a moving image applied through an image input apparatus into each object image having a certain shape information and a background image, and transmitting the same to a plurality of layers having different resolutions. When encoding a low resolution base layer image to a high resolution enhancement layer image by using a scan interleaving method, a lossy encoding method for elastic shape information encoding is applied by changing a probability table of an arithmetic encoding method. . The method of claim 1, wherein the encoding method comprises a greater probability of generating '0' or '1' when the inside of an object is '1' and the outside of the object is '0' in a reference probability table used in CAE. Flexible shape characterized by loss coding the shape information by applying a new probability table that has a value of 1 and a small value of 0 and a modified probability table that changes all the probabilities when shape information is encoded using CAE. Lossy coding method in information coding. The method of claim 1, wherein the encoding method performs shape information loss coding by applying a modified probability table in which all probabilities are changed to '1' or '0' for I-VOP. Lossy coding method. The elastic shape information of claim 1, wherein the encoding method performs shape information loss coding by applying a modified probability table in which all probabilities are changed to '1' or '0' for P and B-VOP. Lossy coding method in coding. The method of claim 1, wherein the encoding method is a method of modifying a probability table in an apparatus for performing shape information loss coding by applying a modified probability table in which all probabilities are changed to '1' or '0' for an I-VOP. The lossy coding method of the case of elastic shape information encoding, characterized in that the probability of each context in the reference INTRA probability table is set to 1 with a higher probability and 0 with a smaller probability. The method of claim 1, wherein the encoding method comprises applying a modified probability table in which all the probabilities are changed to '1' or '0' for P and B-VOPs to modify the probability table in an apparatus for performing shape information loss coding. The lossy coding method according to the present invention, wherein the probability of each context in the existing INTRA probability table is set to 1 with a higher probability and 0 with a smaller probability. The method of claim 1, wherein the encoding method performs shape information loss coding by applying a modified probability table in which all probabilities are changed to '1' or '0' for I-VOP. Lossy coding method. The elastic shape information of claim 1, wherein the encoding method performs shape information loss coding by applying a modified probability table in which all probabilities are changed to '1' or '0' for P and B-VOP. Lossy coding method in coding. The method of claim 1, wherein the encoding method converts the size of a BAB using CR, and when performing the CAE based on the converted BAB when the converted BAB satisfies a threshold, all probabilities are '1' or '0'. The lossy coding method of the case of elastic shape information encoding, characterized in that the encoding is performed using the modified probability table. 10. The method of claim 9, wherein the encoding method comprises setting a threshold for whether to use a modified probability table. 2. The method of claim 1, wherein the encoding method adds a shape information encoding mode to transmit information about whether the modified probability table is used. 2. The method of claim 1, wherein the encoding method adds one bit of additional information to the BAB executing CAE to transmit information on whether the modified probability table is used. Lossy coding method. The stretchable shape of claim 1, wherein the encoding method encodes the encoding mode using a variable length code in consideration of the mode added in claim 10 to transmit information on whether the modified probability table is used. Lossy coding method in information coding. 2. The method of claim 1, wherein the encoding method encodes using a probability value modified in a CAE used when encoding an image of a base layer in a flexible encoding process. 3. 2. The method of claim 1, wherein the encoding method encodes using a probability value modified in a CAE used when encoding an image of an enhancement layer in a flexible encoding process. . 16. The method of claim 15, wherein the encoding method encodes using the probability shape modified in the CAE used when encoding the TSD.

Images