| Video compression is not only necessary but also possible. Intra-and inter-frame redundancy of video sequence makes video compression possible.The latest video coding standard H.264 partitions each video image into 16x16-pixel blocks, allowing video images be processed in pixel-block unit. H.264 video processing uses a variety of techniques: time-domain correlation, spatial residual redundancy, transform, quantization, scanning output and entropy coding, and many different image types.Because of the use of these advanced coding techniques, we get a great improvement on compression and decoding performance. Some studies of the experimental data show that, H.264 is much better than the other existing video coding standard in the SNR, compression efficiency and visual effects.H.264 employs search algorithms based on the block-matching method. Full search algorithm searches each point covering the full search window, and too much computation leads to very slow search speed which places great impact on its application. In order to reduce the complexity of H.264 encoder, people developed a lot of fast search algorithms, such as three-step search method (TSS), two-dimensional logarithm (TDL), the cross-search (CS), nearest neighbor search method (NNS), conjugate direction search (CDS), and block-based gradient descent search method (BBGDS).In this paper, several fast search algorithms used commonly are improved. The early termination method of partial residual computation is added, and the candidate block which has been searched previously is marked. The points of searching template are selected to search in optimal order. Therefore, the improved algorithm method maintains smaller SNR loss and further increases search speed. In addition, the paper also proposes two complex search algorithms -- simplex minimization search algorithm(SM) and improved EPZS algorithm (IEPZS).Many fast algorithms assume residual surface is a single peak monotonous one, but actually there are many cases of complex multi-peak residual surface, and therefore the search tends to fall into the trap of local optimum. According to the idea of simplex optimal search in mathematics, SM algorithm employs reflection, expansion and contraction, and other means in the search window of the multi-peak residual surface, so as to avoid the local optimum and solve the general problem of general fast algorithms as well. After adopting predictive vector sets of EPZS algorithm, the algorithm greatly increases the possibility of initialized search points' covering the global optimal point and further avoids the problem of falling into the global optimization.EPZS is an adaptive search algorithm based on predictive motion vector field. There are three main ways to improve prediction performance: selection of initial predictive vector, adaptive-stop strategy, and different search templates. EPZS achieves good search performance, maintaining the SNR almost the same of full search and increasing search speed about 200 times. On the other hand, because EPZS provides rich predictive vector sets. There are a lot of unnecessary searches for many blocks which have very little motion in temporal and spatial domain and it leads to a waste of searching time. In order to judge the characteristics of block's motion, we make use of the characteristics of predictive vectors in time and spatial domain and the relationship between them. Particularly, for the blocks which have small motion, we set rapid Jump search to avoid unnecessary predictive vector sets and wipe out a lot of redundant search time. |
PDF Full Text Request