INTRODUCTION
1.5 Contribution made in this dissertation
Within H.264/AVC video coding, motion estimation and rate-distortion optimized mode decisions contributes to the largest gain in compression but both of these are also the most computationally intensive parts which limits to real time implementation. In motion estimation, similarities between different video frames are searched and identified; redundant data are then eliminated or minimized to reduce temporal redundancy within a video sequence.
Rate-distortion optimized intra/Inter mode decision is also one of the important coding tools of H.264/AVC encoder. Many motion estimation and mode selection scheme have been developed over the last decade; most of these methods come with a complex search flow, extensive computation complexity and with a limited speedup. To overcome the drawback of these in H.264/AVC, some novel and efficient mode selection scheme have been developed
1.5 Contribution made in this dissertation 7 which brought significant contribution in video coding and in this approach following contributions has been made.
1. This dissertation proposes a modified DC prediction (MDCP) mode scheme for Intra 4×4 based on the distance between the predicted and reference pixels In H.264/AVC intra coding, DC mode is used to predict regions with no unified direction and the predicted pixel values are same and thus smooth varying regions are not well de- correlated.
2. A new Z-scan scheme is proposed to improve intra coding efficiency for the AVS standard. In proposed method, traditional zigzag scan, horizontal z-scan, and vertical z-scan are used depending on the spatial prediction directions. It is easy to implement the proposed scheme into AVS codec without changing the syntax. The proposed scheme can remarkably reduce bit rates compared with AVS codec using zigzag scan, while the PSNR of video sequences are maintained. Within this scheme, according to the spatial prediction directions, two additional scan table called horizontal Z-scan and vertical Z-scan are used together with normal zigzag scan to organize coefficients.
3. A novel rate-based macroblock classification for fast mode decision in H.264/AVC video coding standard is proposed. The main idea is to classify each macroblock into simple motion or complex motion contents based on the Inter 16×16 mode’s residue block bit-rate and then according to the classification different mode searching orders with distinct early termination schemes are employed. This new algorithm is very simple for both hardware and software implementations without extra computational module. To speed up the intra mode decision, a new fast Intra 4×4 mode selection algorithm is also proposed by choosing most likely modes using the low complexity SATD cost as screening function. It is demonstrated by experimental results that the proposed rate-based fast algorithm can reduce significant encoding time with negligible degradation in the rate-distortion performance. A further speedup technique is also introduced in this scheme. As the encoding of the Inter 16×16 mode is part of the rate- distortion optimization mode selection process, thus no additional operation module is required for the proposed fast algorithm.
4. A novel relative sum of absolute difference (RSAD) based inter mode decision al- gorithm is proposed for H.264/AVC video standard. The main idea is to determine the best inter mode based on RSAD cost instead of the rate-distortion cost. This approach can avoid most of the computationally intensive processes in the H.264/AVC mode decision, such as integer transform, quantization, variable length coding and
pixel-reconstruction. Besides, it can solve the problem of SAD-based cost function for H.264/AVC mode decision in biasing to the smaller block size modes, as they are normally achieving higher prediction accuracy. The proposed algorithm can reduce sig- nificant encoding time with negligible degradation in the rate-distortion performance.
Complexity savings are also achieved by processing the macroblocks that are predicted as “skipped”. The proposed algorithm is also combined with SKIP mode early termi- nation algorithm for further time saving and computation reduction. The enhanced algorithm with SKIP mode early termination technique could save more encoding time with only a little rate-distortion degradation. The proposed scheme can avoid rate-distortion cost computation while maintaining high rate-distortion performance of the H.264/AVC codec.
5. A new probability based fast inter mode decision algorithm is proposed for H.264/
AVC video standard. The main idea is to classify the inter modes into large and small modes category and determine which one is suitable for a macroblock (MB).
Depending on this, the probability of each large mode corresponding to the initial HadamardSAD16×16 cost or the probability of each small mode corresponding to the initial HadamardSADs8×8 cost is determined. Then the best mode can be obtained from their probability characteristics. Thus, it is possible to determine the best inter mode only by calculating the Hadamard Transform (H-SAD) cost instead of rate distortion cost. Moreover, it avails the platform to determine the best mode without comparing the cost of each mode to other modes. As a result, this approach can avoid most of the complex computation processes like quantization, variable length coding, pixel reconstruction etc. The proposed based scheme can decide the best mode without calculating the rate distortion cost maintaining a high rate distortion performance of the H.264/ AVC codec.
6. Finally, a new approach of search point reduction, ranking and performance analysis of commonly used motion estimation (ME) algorithms is introduced for H.264/AVC encoder. Since number of searching point is not a direct measure of hardware perfor- mance, a cost function of accessing data in memory is defined and then, an adjustments is given on the existing ME algorithms to minimize the problem of memory access by reordering the sequence of searching points, where the number of searching points and MAE/PSNR remains the same as the original algorithms. In addition to this, the rank- ing of memory access cost for small and high motion video with lower computational complexity and similar quality is maintained is by the experimental evaluation.