Author: Wang Zhibing; Wang Wensheng; Lu Yang; Cui Huijuan; Tang Kun
Further improve video compression efficiency, predict compensation and H. 264 Variable Size block motion prediction in the video coding frame, the weighted coefficient of adaptive adjustment of the overlapping block motion compensation is adapted according to the encoding mode of different blocks. The experimental results show that the proposed adaptive overlapping block motion compensation for more complex sequences, the maximum coding gain can reach 0.21 dB, significantly increase the encoding efficiency of H.264, with significant increased coding efficiency of H.264 at higher target code rates. .
With the development of networks and multimedia technology, more and more digital video services continue to emerge. H. 264 Standard is in order to adapt to various business growth for higher requirements for moving image compression ratios. H. In 264, a variable size block motion prediction, multi-frame prediction, 1/4 pixel interpolation and other techniques are introduced to further improve the accuracy of motion prediction and improve video compression efficiency.
At high compression efficiency, block motion prediction and compensation, BMC's video compression algorithm will generate block effects, which seriously affects the subjective quality of rebuilding video. H. In 264, the removal block effect filter is introduced into the motion prediction / compensation loop to reduce the effects of block effects. On the other hand, the text proves that overlapped block movement compensation (OBMC) is also a effective means to reduce block effects, improve prediction accuracy, and studied the coefficient design method of the optimal overlapping block matching plus window. However, the above algorithm has not yet utilized in the overlapping block movement compensation. 264 A variety of prediction modes in the existing encoding frame, in H. The 264 framework introduces overlapping block motion compensation, and its coding efficiency can be further improved.
This paper proposes an adaptive overlap block motion compensation algorithm based on encoding mode. The algorithm is adaptively selected weighting factor to increase the motion prediction accuracy of the current coded block according to the encoding mode of the encoded block. The experimental results show that this algorithm is at high yard ratio, the maximum coding gain can reach 0.21 dB, improve H. The coding efficiency of 264.
1 Adaptive overlap block motion compensation based on encoding mode
Text proposes of overlapping block motion compensation to improve prediction accuracy and coding efficiency, and balance the distortion distribution. In OBMC, for the current coding block, multiple alternate motion vectors, including their own motion vector and motion vector of the surrounding blocks. By introducing a set of weighted coefficients, the final motion prediction is obtained for each candidate.
Considering H. The most basic coding unit in 264 is 44. For the convenience, the block size of the OBMC in this algorithm is 44. In actual implementation, in order to improve the computational speed, the corresponding OBMC block size can be taken for different encoding modes. For each block in the current frame, as shown in FIG. 1, its alternative motion vector set consists of the motion vector of its own motion vector and the surrounding block. Let BC represent the current coding block, BI (i = 1, 2, 3, 4) indicates its adjacent block, the corresponding motion vector is Vc, Vi.
As shown in FIG. 2, it is assumed that P represents the coordinates of the pixels in the current encoded block, indicating that the pixel is an OBMC's motion prediction value, there is
Wherein, Fn-1 (p) represents the pixel value at the position P in the reference frame, and ωc (p), ωi (p), 1 ≤ I ≤ 4 represent the weighted coefficient of OBMC, respectively, and should satisfy the following constraints
When encoding the BC block, the motion vector of the B2 and B4 may not be encoded yet (when B2, B4 block is not in the same macroblock). At this time, in order to ensure the synchronization of the compilation, the decoding synchronization, the motion vector of the convention B2 and B4 is set to VC.
Since the weighting coefficients of different positions P in the block BI are uneven, their value constitutes a weighted matrix WI. The setting of Wi has largely affected the encoding efficiency of OBMC. The text gives a method of selecting a number of Wi, however, it is for the case where the motion prediction block is fixed.
Due to H. A variable size block motion prediction is taken, from qualitatively, the motion vector corresponding to the divided motion prediction block is generally more accurate, and it is more accurate and more capable of reflecting the movement of adjacent blocks. Based on this hypothesis, the motion vector belonging to the smaller divided block is given to greater weight in OBMC to further improve the motion prediction accuracy of OBMC, and improve the coding efficiency.
According to this inference, this algorithm uses an adaptive OBMC algorithm based on encoding mode. For each neighbor block BI of the current coded block BC, a weighted matrix set Si is defined, and the appropriate weighting matrix Wi∈Si is selected according to the encoding mode of BC and BI. That is, the selection policy of the weighted matrix depends on the respective coding modes of the two blocks. For example, it is assumed that the motion prediction block size corresponding to the adjacent block B is 88, and when the prediction block size of the BC is 1616, corresponding to the Weighted coefficient matrix of V1 is W161, when the prediction block size of the BC is 44, corresponding to V1 weight The coefficient matrix is W41. The selection policy of the weighted matrix should ensure that the coefficients of each position (m, n) in the weighted matrix are available. In practical implementations, a lookup table can be used to store the selection policy. At the same time, the weighted matrix collection and corresponding selection strategies can be obtained by training a number of test sequences.
After using OBMC, the measurement of the corresponding motion search should be corrected. In this article, in order to reduce the complexity of achieving complexity, the search for motion vectors does not take iteration. At this time, when performing a motion search, the motion vector selected by the current block not only determines its own motion predictive value, but also affects the motion prediction value of the right and lower blocks (the motion vector of the left and upper blocks has been determined. Request again). In order to reflect this effect, the criterion of the sports search is redefined as
Among them: The first item is an absolute error and the prediction error of the current block and the second, 3 reflect the effect of predictive error on the right and lower blocks; the BV is required for the coding of the motion vector in paragraph 4.
2 experimental results
This experiment adopts H. 264 Reference Code JML0.1 As the simulation platform, the test results of various CIF (COMMON Intermediate Format) and QCIF (Quarter Common Intermediate Format) test sequences are given. In this experiment, the encoder is set to: the rate distortion is opened, the reference frame is 1, the optional encoding mode is 7, and the content adaptive variable long entropy encoded, the motion search range is 1616, each frame group includes 1 I frame and 49 P frames, all frames use the same quantization parameters.
Table 1 shows that the QP settings of each test sequence are 20, 24, 28, 32, and the peak signal-to-noise ratio, PSNR, and the comparison rate. Table 2 shows the H. standard H. 264 Comparison of compression efficiency. It can be seen from these results: encoding gains using OBMC increases as the coding rate increases. At the high yard ratio, the highest coding efficiency can be increased by 0.21 dB. When the coding sequence is more complicated, such as the COASTGUARD and TEMPETE sequences, the encoding gain is also more obvious. This phenomenon also conforms to the above analysis of OBMC superiority.
Table 3 shows the comparison of the number of specifications of the information of each part of the P frame when the Foreman sequence is encoded. It can be seen that the use OBMC algorithm significantly reduces the rate of encoding motion vector information and luminance residual information. For the sake of simplicity, there is no motion compensation of the chromaticity signal by overlapping block mode, and therefore, the code rate indicating the chromaticity residual signal is comparable to the original algorithm.
3 knot
In order to combine the OBMC with the variable block size motion in the standard, an algorithm of selecting an OBMC weighted matrix according to the neighboring block encoding mode is proposed. Simulation tests show that OBMC adopts an average encoding gain of more than 0.1 DB or more. For high-complexity sequences at high rates, the highest gain can reach 0.2 dB.
Next, adaptation of the weighted coefficient matrix will be studied to further improve the coding efficiency. In addition, it is not that each macroblock can bring the encoding gain. It can be studied to introduce a flag at the macroblock level, indicating whether the macroblock is OBMC. This flexibility will further increase coding efficiency.
Editor in charge: GT, read full text
Our other product:
