KR20100068108A - Adaptive coding mode decision method using the motion vector map in h.264/avc - Google Patents

Abstract

PURPOSE: A coding mode determination method using a movement vector map of an H.264 AVC is provided to implement high-speed encoding by reducing encoding complexity without bit increase and deterioration of picture quality. CONSTITUTION: A location having a minimum SAD(Sum of Absolute Difference) value is obtained in the sum total of two motion vectors(602). The obtained sum total of two motion vectors and motion estimating area size are compared. According to the comparison result, the MVM value toward the current macro block of the motion vector map is set to the designated value(603). The MVM value toward the current macro block of the motion vector map is used.

Description

Adaptive coding mode decision method using the motion vector map in H.264 / AVC}

The present invention relates to a method for determining an H.264 / AVC encoding mode, and more particularly, to define a motion in a previous frame as a motion vector map and to determine the current frame to be encoded according to the information of the motion vector map. The present invention relates to an adaptive encoding mode determination method using a motion vector map in H.264 / AVC that adaptively skips spatial prediction early (intra prediction mode adaptive early skipping).

Recently, ITU-T and MPEG have defined the H.264 / AVC video coding standard with high performance compression efficiency in response to the advancement of hardware and the high-definition needs of users. The H.264 / AVC has a compression structure similar to the existing MPEG-2, MPEG-4, H.263, etc., but has a higher coding rate (compression rate) and image quality, and is being applied to various fields (networks and systems).

1 is a diagram illustrating an embodiment of an H.264 / AVC encoder to which an algorithm proposed by the present invention is applied.

As shown in FIG. 1, the H.264 / AVC coder includes a DCT transform / quantization unit [DCT / Quant] (1), an entropy coding unit [Entropy Coding] (2), and a DCT inverse transform / inverse quantization unit [IDCT / IQuant. ] (3), reference frame memory [Ref. Frame Memory 4, Motion Estimation 5, Motion Compensation, Intra-Mode Prediction 6, and the like.

In H.264 / AVC, when the input current image is an intra frame, the DCT conversion / quantization unit 1 performs DCT conversion and quantization on the difference between the prediction pixel block determined as an optimal mode through the intra prediction process and the current image. The variable length encoding process is performed by the entropy coding unit 2 by obtaining a quantization coefficient through the signal, and transmitted to the decoder. The encoded image is stored as a reference frame image for inter-frame encoding in the reference frame memory 4 by adding an intra prediction pixel and an image obtained through inverse DCT transformation and inverse quantization by the DCT inverse transform / inverse quantization unit 3.

On the other hand, in the case of the inter frame, the difference value between the image inputted by the motion predictor 5 and the motion compensator 6 and the predicted image obtained by motion prediction and compensation is obtained, and the image predicted by the intra mode and The predicted image determined to be the optimal mode among the difference values of is different from the current input image and entropy encoded through DCT transformation and quantization. As in the case of the intra frame, the encoded image is stored in the reference frame memory 4 by combining an image obtained through inverse DCT transformation and inverse quantization and an image having motion compensation.

In general, the H.264 / AVC encoder encodes one image in units of a set of pixels (picture elements or pixels) of a predetermined size block instead of encoding the entire image unit. 16, 8x8, and 4x4. Herein, a 16 × 16 block is called a 'macroblock' and an 8 × 8 or 4 × 4 block is called a 'block'. Macroblocks constituting a single image are encoded in a raster scanning order (left to right, top to bottom).

FIG. 2 is a diagram illustrating an embodiment of macroblock unit division and macroblock encoding order of an image. FIG.

FIG. 2 shows an example of dividing an image into macroblocks. An image having a size of 176 × 144 is divided into 99 macroblocks, numbered in the order shown in FIG. 2, and sequentially encoded from a macroblock having a small number. do.

Meanwhile, a macroblock of H.264 / AVC is divided into variable blocks and determined in which mode to encode using a rate-distortion optimization (RDO). Herein, a mode having a variable block is divided into {16 × 16, 8 × 16, 16 × 8, SUB8 × 8, I4MB, I16MB}. Among these, the optimal mode is determined by the rate-distortion optimization technique that has the best conditions in terms of image quality and bit rate.

FIG. 3 is a diagram illustrating an embodiment of a block size division partition in an inter mode, and FIG. 4 is a diagram illustrating an example of a prediction process for each block size in an intra prediction mode.

Hereinafter, a method of determining an optimal mode of a macroblock according to the prior art will be described with reference to FIGS. 3 and 4.

First, as shown in FIG. 3, macroblocks, which are basic blocks of an inter mode, are divided into {16 × 16, 16 × 8, 8 × 16, and SUB8 × 8} partitions. Here, SUB8 × 8 may be further subdivided into {8 × 8, 8 × 4, 4 × 8, 4 × 4} modes independently for each 8 × 8 block. For each partition block thus partitioned, Equation 1 below determines a motion vector having a minimum value and a reference picture.

은 라그랑지안 계수로서 양자화 계수에 의존하면서 왜곡값과의 차수를 맞춰준다.In [Equation 1] and [Equation 2], J _motion is a function for determining the optimal motion vector and reference image, SAD _mode is decoded with the current input image block s for each partition block frame memory It is obtained using the difference of the past image blocks r stored in. H and V are the height and width of each partition block, and m _x and m _v are the motion vectors. R (MV, REF) is the number of bits needed to encode the motion vector MV and the reference picture REF.

Is a Lagrangian coefficient that matches the order of distortion, depending on the quantization coefficient.

Next, prediction for the intra prediction mode is performed. As shown in FIG. 4, intra prediction is divided into an I4MB mode having nine prediction directions in a 4 × 4 block and an I16MB mode having four directionalities in a 16 × 16 block. The mode for I4MB and I16MB determines a prediction mode in which the difference between the prediction value according to the current macroblock and the direction is minimized. The difference value is calculated using Equation 2 above.

As described above, the prediction modes determined by the inter mode and the intra mode determine and encode a mode in which the J _mode value is the minimum as the optimal mode for macroblock coding using Equation 3 below.

는

의 제곱 값이다. H와 V는 모든 블록에 대한 블록 사이즈이며, M은 부호화 가능한 모든 예측 모드이다. R(s, r_M, M)은 M에 해당하는 모드로 부호화했을 때 실제로 발생하는 비트수이며, SSD_mode는 현재 입력 영상과 복호된 영상 r_M과의 차에 제곱으로 구한다.In [Equation 3] and [Equation 4],

Is

Is the squared value of. H and V are block sizes for all blocks, and M is all predictable prediction modes. R (s, r _M , M) is the number of bits actually generated when coding in the mode corresponding to M, and SSD _mode is obtained by square the difference between the current input video and the decoded video r _M.

That is, in the prior art, the H.264 / AVC encoding process, for example, the method of determining the optimal mode of the macroblock, has an optimal rate-distortion value (RDC) through motion prediction of various block sizes and spatial prediction in nine directions. Distortion Cost) is used to perform the encoding in the mode having the minimum rate-distortion value.

However, in the prior art, since the rate-distortion value has to be calculated through prediction of all modes that can be encoded for each macroblock, the coding complexity increases by at least 10 times compared to the conventional MPEG-2, MPEG-4, H.263, etc. There is this. As the coding complexity increases, there is a problem in that real-time encoding of an image of HD or higher is almost impossible.

For example, considering the characteristics of the image, since the temporal correlation is higher than the spatial correlation, the probability of selecting the intra prediction mode in the inter frame is very low. However, if the probability of being encoded in the intra mode is low, the coding efficiency gain may be reduced when the intra mode is forcibly omitted.

Accordingly, the H.264 / AVC encoding fastening technique that adaptively omits the spatial prediction of the current frame to be encoded early [intra prediction mode adaptive early omission] while preventing the encoding efficiency gain from being reduced without deterioration of image quality and increase of bits. There is an urgent need.

The present invention has been proposed to solve the above problems and to meet the above demands. The technical problem to be achieved is to define a motion vector map in a previous frame as a motion vector map. The present invention provides an adaptive encoding mode determination method using a motion vector map in H.264 / AVC, which adaptively omits [intra prediction mode adaptive early omitting] the spatial prediction of the current frame to be encoded according to the information of H.264 / AVC. .

The objects of the present invention are not limited to the above-mentioned objects, and other objects which are not mentioned will be clearly understood by those skilled in the art from the following description.

In order to achieve the above object, the method according to an embodiment of the present invention, in the adaptive coding mode determination method using a motion vector map in H.264 / AVC, the motion information is moved for each macroblock in the previous frame Setting a vector map (MVM) and obtaining a sum of two motion vectors of positions having a minimum SAD value; Setting an MVM value for a current macroblock on the motion vector map to a predetermined value according to a comparison result of the sum of the two motion vectors and the size of a motion search region; And determining an encoding mode by using an MVM value for a current macroblock on the motion vector map from the next frame to the previous frame.

As described above, the present invention has the effect of adaptively omitting (intra prediction mode adaptive early skipping) the spatial prediction of the current frame to be encoded by defining a motion vector map.

In addition, the present invention has the effect of eliminating the possibility that a coding error may occur in a portion where a new image region appears due to the omission of spatial prediction.

In addition, the present invention can reduce the coding complexity to an average of 30% or more without degrading the image quality and increasing the bit without conforming to the H.264 / AVC coding standard, thereby realizing the encoding speed. There is.

Specific details of other embodiments are included in the detailed description and the drawings.

Advantages and features of the present invention and methods for achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be embodied in various different forms, and the present embodiments merely make the disclosure of the present invention complete, and are common in the art to which the present invention pertains. It is provided to inform those skilled in the art to the fullest extent of the invention, the invention being defined only by the scope of the claims. Like reference numerals refer to like elements throughout.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

In the present invention, when performing H.264 / AVC encoding on an image in a macroblock unit (preferably, determining an optimal mode of a macroblock), the motion in the previous frame is defined as a motion vector map. An algorithm for adaptively skipping early (intra prediction mode adaptive early skipping) the spatial prediction of the current frame to be encoded according to the information of the motion vector map for each frame is proposed.

In the motion vector map (MVM) proposed in the present invention, the MVM value is set to "0" when the sum of the absolute magnitudes of x and y defined by the 16x16 refinery search is smaller than the motion search size, and vice versa. Is set to "1". That is, in the present invention, when the MVM value of the same position as the macroblock to be encoded is "0", spatial prediction is omitted early to reduce coding complexity.

In addition, in consideration of the fact that the encoding result may not be good in the region where the new image region appears due to the early omission of the spatial prediction, in the present invention, the threshold value is set using the SUB8x8 and I4MB determined as the optimal mode in the previous frame. The encoding speed is implemented while minimizing the error that may occur when using a map.

5 is a diagram illustrating an embodiment of a motion vector map according to the present invention.

In the present invention, attention was paid to the size of the motion information in order to adaptively omit the intra prediction mode early. For example, when an object in the image moves as shown in (a) and (c) of FIG. 5, new information is likely to appear in the place where the object is located, and thus it is likely to be encoded in the intra mode. In addition, if the size of the moved motion vector is large, the RD value may be large, which is also likely to be encoded in the intra mode.

Accordingly, the present invention defines a motion vector map (MVM) to be used to determine whether to skip the intra prediction mode, and sets the motion information to "0" or "1" for each macroblock in the motion vector map (MVM).

The motion vector map (MVM) update according to the present invention first _obtains the motion vectors MV _16x and MV _16y at the position having the minimum sum of absolute difference (SAD) value through the motion search in the 16x16 mode, and then the following [math] A process of determining whether to update the motion vector map by using the sum MV _val of two motion vectors MV _16x and MV _16y as shown in Equation 5]. In the present invention, it is determined whether the intra prediction mode is omitted using the updated motion vector map (MVM).

Then, the process of updating the motion vector map (MVM) will be described in detail with reference to Equation 5 and FIG. 5.

Since all macroblocks of the first frame are encoded in the intra mode, the motion vector map (MVM) is initialized to mode "0" as shown in (b) of FIG.

In encoding the second frame, that is, the first inter frame, motion prediction (motion search) is performed in units of macroblocks.At this time, motion vectors MV _16x and MV _16y having a minimum SAD value are obtained through motion prediction. Using Equation 5], the sum MV _val of two motion vectors MV _16x and MV _16y is obtained.

Then, if the sum MV _val of the two motion vectors MV _16x and MV _16y is greater than the motion search range size, the MVM value for the current macroblock on the motion vector map MVM is “1”. Set it. For example, if the sum MV _val of the two motion vectors MV _16x and MV _16y is larger than the size of the motion search region, the macroblock is a region having a lot of motion, and thus it is highly likely to be encoded in the intra mode.

On the other hand, if the sum (MV _val ) of the two motion vectors (MV _16x and MV _16y ) is smaller than the size of the motion search region, the current macroblock is regarded as a region with less motion and the MVM for the current macroblock on the motion vector map (MVM). Set the value to "0".

Starting from the next frames, encoding is performed by referring to an update state of an MVM value for a current macroblock on a motion vector map (MVM). That is, if the referenced MVM value of the same region as the macroblock to be encoded is set to "0", the current encoded region is regarded as a region with little movement, and the intra prediction mode is omitted, that is, the intra prediction mode is not performed. In this case, if the referenced MVM value of the same region as the current macroblock to be encoded is set to "1", the intra prediction mode is performed by considering the current encoding region as a high-motion region.

On the other hand, since the motion vector map (MVM) described above contains motion information of an image, an error may occur with respect to an image changed into a new area.

In order to reduce the error as described above, in the present invention, the threshold value using the rate-distortion value of SUB8 × 8 and I4MB, which has a high probability of being determined as an optimal mode in a lot of motion and complex areas, is set as shown in Equation 6 below. We present an algorithm that determines if a new image occurs.

In Equation 6, J _{min {inter}} means a rate-distortion value having a minimum value among inter modes of a macroblock to be currently encoded, and Avg (J _{I4MB} , J _{{SUB8 × 8}} ) The optimal mode among the previously encoded macroblocks means an average of rate-distortion values determined as SUB8 × 8 or I4MB.

In other words, the present invention determines whether the intra prediction mode is omitted using the motion vector map (MVM), and if the equation (6) is not satisfied, the intra prediction mode is performed by considering the new region as a part where a new region is generated.

In summary, the algorithm proposed in the present invention determines whether the intra prediction mode is omitted in two steps.

That is, it is determined whether the intra prediction mode is omitted using the MVM value on the motion vector map (MVM) updated by the previous frame.

In order to compensate for macroblocks of newly emerging objects or complex screen structures, if the minimum rate-distortion value for the inter mode is obtained and Equation 6 is not satisfied (without skipping the intra prediction mode), the intra prediction mode Perform

Finally, the MVM value on the motion vector map (MVM) to be used for next frame encoding is updated.

An adaptive encoding mode determination method using a motion vector map in H.264 / AVC according to the present invention as described above will be described with reference to FIG.

6 is a flowchart illustrating an adaptive coding mode determination method using a motion vector map according to the present invention.

First, an MVM value on a motion vector map (MVM) is initialized to "0" in an intra frame (601).

Then, in the case of inter frame, the motion vectors MV _16x and MV _16y at the position having the minimum SAD value are obtained through 16 × 16 motion prediction, and the sum of these two motion vectors (MV _16x and MV _16y ) (MV _val ) is obtained. Obtain (602).

Then, in case of an inter frame, the motion vector map (MVM) of the current frame is updated (603). If the sum of the two motion vectors (MV _16x and MV _16y ) (MV _val ) is larger than the size of the motion search region [M _Vval > Search range], the MVM value for the current macroblock on the motion vector map (MVM) is set to "1". If the sum of the two motion vectors (MV _16x and MV _16y ) (MV _val ) is less than the size of the motion search region [M _Vval ≤ Search range], the MVM value for the current macroblock on the motion vector map (MVM) is " Set to 0 ".

Then, the encoding mode is determined using the MVM value of the current macroblock on the motion vector map (MVM) of the previous frame from the next inter frame (604).

That is, when the MVM value corresponding to the current macroblock position is "1", all the codeable modes are predicted (605). When the MVM value corresponding to the current macroblock position is "0", it is checked whether Equation 6 is satisfied. (606).

If Equation 6 is satisfied, that is, the optimal mode among the macroblocks in which the rate-distortion value [J _{min {inter}} ] having the minimum value among the inter modes of the macroblock to be currently encoded is SUB8 ×. If the average [Avg (J _{I4MB} , J _{{ SUB8x8 }} ] of the rate-distortion value determined as 8 or I4MB is smaller than the intra prediction mode, the intra prediction mode is omitted (607).

Next, if the currently encoded macroblock is an intra mode, the current MVM value is updated to "1" and the process proceeds to step '603'.

Although the embodiments of the present invention have been described above with reference to the accompanying drawings, those skilled in the art to which the present invention belongs may be embodied in other specific forms without changing the technical spirit or essential features of the present invention. You will understand that. Therefore, the embodiments described above are to be understood in all respects as illustrative and not restrictive.

1 is a diagram illustrating an embodiment of an H.264 / AVC encoder to which an algorithm proposed by the present invention is applied.

FIG. 2 is a diagram illustrating an embodiment of macroblock unit division and macroblock encoding order of an image. FIG.

3 is a diagram illustrating an embodiment of a block size partitioned partition in an inter mode.

4 is a diagram for explaining an example of a prediction process for each block size in an intra prediction mode.

5 is a diagram illustrating an embodiment of a motion vector map according to the present invention.

6 is a flowchart illustrating an adaptive coding mode determination method using a motion vector map according to the present invention.

Claims (6)

In the method of adaptive coding mode determination using a motion vector map in H.264 / AVC, Setting motion information in a motion vector map (MVM) for each macroblock in a previous frame, and obtaining a sum of two motion vectors of positions having a minimum SAD value; Setting an MVM value for a current macroblock on the motion vector map to a predetermined value according to a comparison result of the sum of the two motion vectors and a motion search region size; And A method of determining an adaptive encoding mode using a motion vector map in H.264 / AVC, comprising determining an encoding mode using an MVM value of a current macroblock on the motion vector map from a next frame to a previous frame. The method of claim 1, Obtaining the sum of the two motion vectors, An adaptive coding mode determination method using a motion vector map in H.264 / AVC, wherein two motion vectors of positions having a minimum SAD value are obtained through 16 × 16 motion prediction, and the sum of the two motion vectors is obtained. . The method of claim 1, The setting of the MVM value for the current macroblock on the motion vector map to a predetermined value may include: If the sum of the obtained two motion vectors is larger than the size of the motion search region, the MVM value for the current macroblock on the motion vector map is set to "1". If the sum of the two motion vectors is smaller than the size of the motion search region, An adaptive encoding mode determination method using a motion vector map in H.264 / AVC, wherein the MVM value for the current macroblock on the motion vector map is set to "0". The method of claim 3, wherein The determining of the encoding mode may include: When the MVM value corresponding to the current macroblock position on the motion vector map is "0", the rate-distortion value having the minimum value among the inter modes of the macroblock to be currently encoded is optimal among the previously encoded macroblocks. The method of adaptive coding mode using the motion vector map in H.264 / AVC, wherein the intra prediction mode is omitted if the mode is smaller than the average of the rate-distortion value determined as SUB8 × 8 or I4MB. The method of claim 3, wherein The determining of the encoding mode may include: The method for adaptive coding mode determination using the motion vector map in H.264 / AVC, characterized in that all the codeable modes are predicted when the MVM value corresponding to the current macroblock position on the motion vector map is "1". The method according to claim 4 or 5, And updating the current MVM value on the motion vector map to “1” if the currently encoded macroblock is an intra mode.

Images