Research On The Inter Prediction Techniques And The Fast Algorithms For AVS2

With high-definition, ultra-high definition video more widely, H.264 / AVS standards have been difficult to meet the requirements of the application, a higher compression efficiency of video coding standard is required. For this reason, MPEG / VCEG developed HEVC video coding standard, and AVS Working Group began to develop a new generation of video coding standard AVS2 at the end of 2011.In order to improve the coding efficiency, AVS2 using a variety of techniques, including chunks quad-tree coding structure, more flexible prediction block partition mode, more the intra prediction mode, more inter prediction mode, more flexible reference image configuration technique and sample offset compensation filtering technology. New technologies improve the coding efficiency, but also significantly increase the coding complexity. For this reason, research AVS2 fast encoding technology, especially the most time-consuming inter prediction technique is very urgent.The main research of this paper contains the inter prediction mode selection technique and the reference frame technique, and the reference image technology consists two parts: a reference picture management and reference picture selection optimization.As the AVS2 use the more inter modes, the coding complexity also increase vastly, this paper propose a fast inter mode decision method based on texture features of coding unit. This paper analyzes the probability distribution of each inter mode is selected as the optimal inter mode,and found that the inter mode distribution differ as the texture complexity of the video sequences. According to this phenomenon, before encoding an coding unit we analyze its texture, determining prediction unit is divided horizontally or vertically symmetric symmetrical division, and then further analyze two asymmetric direction by dividing the dividing direction, determine whether one of the non-symmetrical division. Experimental results showed that, by this analysis texture feature, selecting the most likely mode of division manner, an average reduction of 37% encoding time, effectively reducing the coding complexity.For the issue of using a reference image of a B-picture, image reordering and decoded picture buffer management more complex, this paper proposed a decoded image buffer output delay management.For multi-reference image motion estimation technology to bring a substantial increase in the complexity of the problem, this paper proposed a reference image coding unit level based on the correlation between the predicted and the adjacent unit reference image correlation algorithm to quickly select a reference image, by analyzing the lower layer encoding unit within the best reference picture prediction unit correlation, the optimum use of the reference image set lower layer after trying the four sub-coding unit obtained by encoding the reference picture and the best prediction unit adjacent to the current coding unit, as each reference image candidate prediction unit, thereby reducing the number of candidate reference image, reducing coding complexity. Experimental results show that this method of coding efficiency in ensuring the premise, the encoding time can be shortened by about 39 percent on average.