TWI478587B - A Fast Method for Predicting Modular Multimedia Video Coding - Google Patents

Description

Mode fast prediction method for adjustable multimedia video coding

The invention relates to the technology of image processing, in particular to a mode fast prediction method for adjustable multimedia video coding.

In the adjustable multimedia video coding, the prediction of the coding mode is composed of three prediction modes, one is texture prediction within the picture, the other is motion prediction between pictures, and the last is cross-resolution/cross-layer prediction.

The coding mode of the macroblock in the picture is divided into 16x16, 16x8, 8x16, 8x8, 8x4, 4x8, and 4x4, and in the search mode of each macroblock, if all 16x16, 16x8, 8x16, The designation of the 8x8, 8x4, 4x8, and 4x4 coding modes is defined as a full mode in this case.

The texture prediction in the picture is formed by the pixels around the current macroblock, and is divided by two kinds of macroblocks to make textures and predict the texture most similar to the current macroblock. The motion prediction between pictures is to predict the motion vector (Motion Vector, MV) of the current macro block in the adjacent macro block, and predict the Predict Motion Vector (PMV) to the Reference Frame. The search range (SR) is used to predict the most similar macroblock of the current macroblock in the search range. The vector between the two is called the motion vector, and the mobile prediction is through eight kinds. The macroblock partition is divided into dynamic vectors of up to forty different block partitions. The cross-resolution/cross-layer prediction is to use a smaller resolution image to predict the picture information of the smaller resolution completed mode, and give the larger resolution picture as the mode to predict, including three cross-resolutions. / Cross-layer prediction mode: cross-resolution / cross-layer intra-texture prediction (Interlayer Intra), cross-resolution / cross-layer difference (Interlayer Residual) and cross-resolution / cross-layer motion measurement (Interlayer Motion), and cross The resolution/cross-layer difference prediction includes an inter-picture motion prediction mode that removes smaller resolution differences. There are eight macroblock partitions, and forty-one dynamic vectors are also calculated. The full-mode prediction is based on texture prediction in the above picture, motion prediction between pictures, and cross-resolution/cross-layer prediction.

In order to obtain the best prediction mode, it is necessary to complete the texture prediction in the above picture, two kinds of macro block partitioning, eight modes of motion prediction between pictures, and ten modes of cross-resolution/cross-layer prediction. Only one of them will become the most Good prediction mode, while the other 19 modes are discarded, resulting in a very large amount of redundant computational complexity. In small resolution pictures such as QCIF and CIF, due to the small number of macro blocks, the computational complexity of cumulative redundancy seems to be less obvious, but when the resolution reaches D1, HD720, Full HD1080 or even larger, The number of macroblocks is quite large, and the complexity of this redundancy becomes the biggest cause of slowing down overall performance.

Since the data stream of the adjustable multimedia video coding includes a large resolution and a small resolution, it is necessary to generate redundant calculations during compression. Therefore, we use a variety of innovative algorithms in the design of the algorithm. The method reduces the computational complexity and balances the picture quality.

The main object of the present invention is to provide a mode fast prediction method for adjustable multimedia video coding, which can reduce computational complexity and balance picture quality.

In order to achieve the foregoing objective, an adjustable multimedia video coding mode fast prediction method according to the present invention has three different resolution sizes, which are respectively a base layer of minimum resolution and a medium resolution. a first enhancement layer and a second enhancement layer of maximum resolution, the method comprising the following steps: a) starting from the minimum resolution picture, sequentially performing correlation coding mode prediction on each macro block, After all the macroblocks of the minimum resolution picture complete the coding mode prediction, the coding mode prediction is performed on the macroblocks of the medium resolution picture sequentially, and finally the macro block of the maximum resolution picture is sequentially sequenced. Performing coding mode prediction; b) defining a macroblock currently being processed as a current macroblock, determining whether the current macroblock is the minimum resolution, and if so, directly performing the operation in the full mode, and Determining that the full mode is the best coding mode of the current macroblock, and then jumping to step f); if not, proceeding to step c); The optimal coding mode refers to the mode when the signal-to-noise distortion ratio is minimum; c) refers to the optimal coding mode of the sub-small resolution macroblock corresponding to the current macroblock, and utilizes the reference neighbor block mode technique Referring to the optimal coding mode of the neighboring macroblocks of the same resolution corresponding to the current macroblock, the reference coding modes are counted, and the number of occurrences of the same coding mode is judged, and then the number of occurrences is The coding mode is sorted; d) performing the action of predicting the optimal coding mode early, which performs each of the adjacent macroblocks corresponding to the sorting order of step c) one by one, and for a neighboring macroblock, The ratio of the signal-to-noise distortion ratio generated by the signal-to-noise distortion ratio generated by the sub-small-resolution macroblock corresponding to the adjacent macroblock is calculated to obtain the ratio of the two, and then the current macro is obtained. The signal-to-noise distortion ratio of the block is multiplied by the ratio, and compared with the signal-to-noise distortion ratio of the sub-small-resolution macroblock corresponding to the current macroblock, if the comparison result is smaller than the latter, the current judgment is ratio The comparison mode is the optimal coding mode, and stops the comparison of the signal-to-noise distortion ratio corresponding to the remaining coding modes. If the former is greater than or equal to the latter, the information generated by the next adjacent macroblock is continuously compared. The noise-to-distortion ratio, if no comparison occurs to the last coding mode that is less than the noise-to-noise ratio of the small-resolution macroblock, the last coding mode in the sorting order is the optimal coding mode; Determining the optimal coding mode determined in step d) as the best coding mode of the current macroblock; f) determining whether the current macroblock is the last macroblock, and if so, performing steps g) If no, the next macroblock is designated as the current macroblock and jumps to step b); and g) ends. Thereby, the computational complexity can be reduced and the picture quality can be balanced.

In order to explain the technical features of the present invention in detail, the following preferred embodiments are described below with reference to the accompanying drawings, wherein, as shown in FIG. 1 to FIG. 4, a preferred embodiment of the present invention provides a preferred embodiment of the present invention. Adjustable multimedia video coding mode fast prediction method, wherein the adjustable multimedia video coding has three different resolution sizes, respectively, a base layer of minimum resolution, a first enhancement layer of medium resolution, and maximum resolution A second enhancement layer of the degree, the method mainly has the following steps:

a) starting from the minimum resolution picture, performing correlation coding mode prediction on each macroblock in sequence, and performing coding mode prediction on all macroblocks of the minimum resolution picture, and then sequentially on the medium resolution picture The macroblock block performs coding mode prediction, and finally performs coding mode prediction on the macroblock block of the maximum resolution picture.

b) defining a macroblock currently being processed as a current macroblock, determining whether the current macroblock is the minimum resolution (ie, is the base layer?), and if so, directly in the full mode Operate, and decide that the full mode is the best coding mode of the current macroblock, and then jump to step f); if not, then proceed to step c). Among them, the optimal coding mode refers to the mode when the signal-to-noise distortion ratio is the smallest.

c) referring to the best coding mode of the sub-small resolution macroblock corresponding to the current macroblock, and using the reference neighboring block mode technique to refer to the proximity of the same resolution corresponding to the current macroblock The optimal coding mode of the macroblock block, the statistics of the reference coding modes are counted, the number of occurrences of the same coding mode is judged, and each coding mode is sorted by the number of occurrences. Wherein, the foregoing sub-small resolution, for example, if the current macroblock is medium resolution, the sub-small resolution is the minimum resolution, if the current macroblock is the maximum resolution, then The small resolution is medium resolution. The foregoing operation of referring to the neighboring macroblock is as illustrated in the second figure. Taking the current macroblock MB (E1A) of the first enhancement layer as an example, the reference sub-small resolution macroblock is referred to. That is, the corresponding macro block MB (BA) of the base layer, and the neighboring macro block of the same resolution referenced is the macro block MB (E1B), MB (E1C), MB of the same enhancement layer. (E1D); similarly, taking the current macroblock MB (E2A) of the second enhancement layer as an example, refer to the sub-small resolution macro block MB (E1A), and refer to the same resolution macro block. MB (E2B), MB (E2C), MB (E2D).

d) performing an action of predicting the optimal coding mode early, which is performed by successively selecting each adjacent macroblock corresponding to the sorting sequence of step c), and for a neighboring macroblock, the system generates The noise-to-noise distortion ratio is calculated by the ratio of the noise-to-noise distortion ratio generated by the sub-small-resolution macroblock corresponding to the adjacent macroblock to obtain the ratio of the two, and then the noise of the current macroblock is obtained. After the distortion ratio is multiplied by the ratio, the signal-to-noise distortion ratio of the sub-small resolution macroblock corresponding to the current macroblock is compared. If the comparison result is smaller than the latter, the currently compared coding mode is determined. Is the best coding mode, and stops the comparison of the signal-to-noise distortion ratio corresponding to the remaining coding modes. If the former is greater than or equal to the latter, then continue to compare the signal-to-noise distortion ratio generated by the next adjacent macroblock sequentially. When the comparison to the last coding mode does not occur less than the noise-to-noise ratio of the small-resolution macroblock, the last coding mode in the sorting order is the optimal coding mode. The foregoing comparison operation for predicting the optimal coding mode in advance is illustrated by taking the third picture as an example, wherein the current macro block MB (EA_ET) is located in the first enhancement layer, and the macro block MB on the left side (EB_ET) ) is a neighboring macroblock, therefore, the signal-to-noise distortion ratio of the left macroblock MB (EB_ET) and the corresponding sub-small resolution (ie, the base layer) macroblock MB (BB_ET) The noise distortion ratio is obtained by the ratio (ie, the ratio of the signal-to-noise ratio, RDcost ratio=RDcost(EB_ET)/RDcost(BB_ET)), and then the noise of the current macro block MB (EA_ET) is distorted. After multiplying by the ratio (RDcost(EA_ET) * RDcost ratio), the sub-small resolution (ie, the base layer) macro block MB (BA_ET) corresponding to the current macro block MB (EA_ET) The noise to noise ratio is compared.

Due to the early prediction of the optimal coding mode in step d), an error condition in which the optimal coding mode is not selected may occur, and if the optimum coding mode of the current macroblock in the medium resolution has generated an error, It will affect the correctness of the best coding mode of the current macroblock of the subsequent maximum resolution, and expand the aforementioned error condition.

Therefore, in this embodiment, the step d1) is further included, which is an action of dynamically adding the complete mode, mainly when performing the macroblock block mode prediction to the medium resolution picture, and the current macro area is The signal-to-noise distortion ratio of the neighboring macroblocks of the same resolution corresponding to the block is averaged to obtain an average value, and the average value is divided by the signal-to-noise distortion ratio of the current macroblock, and the medium is obtained. The signal-to-noise distortion ratio trend of the current macroblock of the resolution is stored and stored. When the macroblock block mode prediction to the maximum resolution picture is performed, the signal-to-noise distortion ratio of the neighboring macroblocks of the same resolution corresponding to the current macroblock is averaged to obtain an average value, and then an average value is obtained. The average value is divided by the current noise distortion ratio of the current macroblock, and the maximum resolution of the current macroblock block is obtained. Comparing the signal-to-noise distortion ratio trend of the current macroblock of the medium resolution to the signal-to-noise distortion ratio trend of the current macroblock of the maximum resolution, if the comparison result is not the same, it represents the step d) The action of predicting the best coding mode early produces an error, so the optimal coding mode of the next macroblock of the current macroblock that determines the maximum resolution is the full mode; if the comparison result is the same, the subsequent steps are continued. .

In the foregoing step d1), the fourth figure is taken as an example to illustrate how to obtain the signal-to-noise distortion ratio trend. When the current macro block MB (E1A_R) of the medium resolution picture is performed, the signal-to-noise distortion ratio of the neighboring (left and upper) macro blocks MB (E1B_R) and MB (E1C_R) is obtained. After averaging, the signal-to-noise distortion ratio of the current macroblock MB (E1A_R) is divided, and the signal-to-noise distortion ratio trend (RDcost average/RDcost(E1A_R)) is obtained and stored. Then, when the current macro block MB (E2A_R) of the maximum resolution picture is performed, the signal-to-noise distortion ratio of the neighboring (left and upper) macro blocks MB (E2B_R) and MB (E2C_R) is obtained. After averaging, the signal-to-noise distortion ratio of the current macroblock MB (E2A_R) is divided, and the signal-to-noise distortion ratio trend (RDcost average/RDcost(E2A_R)) is obtained.

In this embodiment, the signal-to-noise distortion ratio trend in the foregoing step d1) has a certain range, and the size of the range can be adjusted by the user, and the current resolution of the current macro block is moderately resolved. When the noise distortion ratio is compared with the trend of the signal-to-noise distortion ratio of the current macroblock of the maximum resolution, the smaller the range of the signal-to-noise distortion ratio trend is, the easier it is to be the same. It can be seen that the larger the range is set, the less likely it is to achieve the same situation. Therefore, the more the coding mode is determined to be the full mode, the better the coding quality will be;

e) determining the optimal coding mode determined in step d) as the best coding mode for the current macroblock.

f) determining whether the current macroblock is the last macroblock, if yes, proceeding to step g); if not, designating the next macroblock as the current macroblock and jumping to step b ).

g) End.

The adjacent macroblocks in the steps c) and d1) refer to macroblocks of at least two of the left, upper left, upper, and upper right directions of the current macroblock. In this embodiment, the neighboring macroblock refers to the macroblock located on the left side, the upper side, and the upper right side of the current macroblock in step c), and is located in the current macro in step d1). The macro block on the left and above the block. The more neighboring macroblocks, the more complicated the operation but the better the encoding quality. The less the operation, the simpler the operation but the lower the encoding quality.

It should be added that the above step d1) is not a necessary step, and in the case where the step d1) is not added, the slight error caused by the action of the optimal coding mode is predicted early, even if the error of the screen after the resolution is exceeded. The extension will only make the coding quality worse, which can still meet the needs of lower picture requirements. That is, the present invention is not necessary to add the step d1).

It can be seen from the above steps that the present invention performs coding mode prediction on each macroblock sequentially, and mainly refers to the optimal coding mode of the neighboring macroblocks, and sorts them to perform early prediction and optimal coding. The action of the mode can predict the optimal coding mode of the current macroblock early, thereby omitting the operations required for subsequent unnecessary coding mode prediction, and reducing the computational complexity. Furthermore, since the optimal coding mode of the current macroblock can be predicted early, the quality of the picture can be balanced without being deteriorated.

The first figure is a flow chart of a preferred embodiment of the present invention.

The second figure is a schematic diagram of the operation of a preferred embodiment of the present invention, showing the action of referring to the adjacent macroblock.

The third figure is a schematic diagram of the action of a preferred embodiment of the present invention, showing a comparison action for predicting the optimal coding mode early.

The fourth figure is a schematic diagram of the action of a preferred embodiment of the present invention, showing an action of obtaining a signal-to-noise distortion ratio trend.

Claims (5)

A mode fast prediction method for adjustable multimedia video coding, the adjustable multimedia video coding has three different resolution sizes, a basic layer of minimum resolution, a first enhancement layer of medium resolution, and maximum resolution a second enhancement layer, the method comprises the following steps: a) starting from the minimum resolution picture, sequentially performing correlation coding mode prediction on each macro block, in all macro areas of the minimum resolution picture After the block completes the coding mode prediction, the coding mode prediction is performed on the macroblock of the medium resolution picture, and finally the coding mode prediction is performed on the macro block of the maximum resolution picture; b) the current processing is being processed. A macroblock is defined as a current macroblock, and it is determined whether the current macroblock is the minimum resolution, and if so, the operation is directly performed in the complete mode, and the complete mode is determined as the current macroblock. The best coding mode, then jump to step f); if not, then proceed to step c); wherein the optimal coding mode means that the signal-to-noise distortion ratio is minimum a mode; c) referring to an optimal coding mode of the sub-small resolution macroblock corresponding to the current macroblock, and using the reference neighboring block mode technique to refer to the same resolution corresponding to the current macroblock The optimal coding mode of the neighboring macroblocks is to count the coding modes of these reference codes, determine the number of occurrences of the same coding mode, and then sort the coding modes according to the order of the number of occurrences from large to small; d) Perform the action of predicting the best coding mode early, which will be step c) Each neighboring macroblock corresponding to the sorting order is performed one by one, and for a neighboring macroblock, the signal-to-noise distortion ratio generated by the neighboring macroblock is compared with the second smallest parsing corresponding to the neighboring macroblock. The noise-to-noise distortion generated by the macroblock is calculated to obtain the ratio of the two, and then the signal-to-noise distortion ratio of the current macroblock is multiplied by the ratio to correspond to the current macroblock. Comparing the signal-to-noise distortion ratio of the sub-resolution macroblocks, if the comparison result is smaller than the latter, it is judged that the currently compared coding mode is the optimal coding mode, and the signal-to-noise distortion ratio corresponding to the remaining coding modes is stopped. The comparison action, if the former is greater than or equal to the latter, continues to compare the signal-to-noise distortion ratio generated by the next adjacent macroblock sequentially, and if the comparison to the last coding mode does not occur less than the small resolution macroblock The condition of the signal-to-noise distortion ratio of the block is determined by the last coding mode in the sorting order as the optimal coding mode; e) determining the optimal coding mode determined in step d) as the most current block of the current macroblock Good coding mode (f) judging whether the current macroblock is the last macroblock, and if so, proceeding to step g), if not, designating the next macroblock as the current macroblock and jumping to Step b); and g) end. According to the fast prediction method of the adjustable multimedia video coding mode described in claim 1, the method further includes: step d1) performing the action of dynamically adding the complete mode, which is performed on the macro region of the medium resolution picture. In the block coding mode prediction, the signal-to-noise distortion ratio of the neighboring macroblocks of the same resolution corresponding to the current macroblock is averaged An average value, and then dividing the average value with the signal-to-noise distortion ratio of the current macroblock block, and obtaining a medium-resolution trend of the signal-to-noise distortion ratio of the current macroblock block, and storing it; When performing the macroblock block mode prediction to the maximum resolution picture, the signal-to-noise distortion ratio of the adjacent macroblocks of the same resolution corresponding to the current macro block is averaged to obtain an average value, and then The average value is divided by the signal-to-noise distortion ratio of the current macroblock, and the maximum resolution of the current macroblock block is obtained. The medium resolution block of the current macroblock is obtained. The signal-to-noise distortion ratio trend is compared with the trend-to-maximum resolution of the current macroblock block. If not, the maximum resolution of the next macroblock of the current macroblock is determined. The best encoding mode is the full mode; if they are the same, the next step is continued. According to the fast prediction method of the adjustable multimedia video coding mode described in claim 2, wherein: the noise-to-noise distortion ratio trend has a certain range, and the noise of the current macro block of the medium resolution is distorted When comparing the trend of the signal-to-noise distortion ratio of the current macroblock with the trend and the maximum resolution, the smaller the range of the signal-to-noise distortion ratio trend is, the easier it is to be the same. The smaller the smaller, the easier it is. The mode fast prediction method of the adjustable multimedia video coding method according to claim 2, wherein: the adjacent macroblock in step c) and step d1), and the left side of the current macro block, A macroblock of at least two of the four upper left, upper, and upper right directions. According to the fast prediction method of the adjustable multimedia video coding mode described in claim 1, wherein: the adjacent macro zone in step c) A block refers to a macroblock of at least two of the four directions of the left, upper left, upper, and upper right of the current macroblock.