Optimization Of AVS Video Encoder's Algorithms And Its Implementation On DSP

AVS is the second generation of source coding standard with China's independent intellectual property rights. The proposal of this standard not only changes the lagging situation of the first generation source coding technology, also it avoids the heavy burden of patent fees of MPEG-4 and H.264/AVC. Compared with the current popular international standard MPEG-4 and H.264/AVC, AVS has the advantage of high-performance, low computational complexity, and low patent costs. However, AVS codec chip and software have not been ripe yet, there is still a long way to go for the realization of its industrialization. Therefore, research on the AVS codec is of vital significance.Based on the research of the development and basic techniques of video coding, the thesis focuses on the sub-pixel motion estimation and inter-mode selection algorithms, and proposes the fast algorithms for the two modules. Then AVS encoder and the proposed fast algorithms are transplanted to the DM642. The main achievements of the thesis include:1. Research on the sub-pixel motion estimation and propose two fast algorithmsIn order to obtain better image quality and higher coding efficiency,1/4 pixel motion estimation is used in AVS. The computation overhead required by fractional pixel motion estimation has become relatively significant, thus reducing the sub-pixel search time is particularly important. People have found that video sequences have the center biased property and strong spatial-temporal correlation, so we can make use of the two aspects to reduce the time cost of sub-pixel search. On one hand, we use the center-biased property, the sub-optimal integer match points are utilized to predict the sub-pixel motion vector direction, and thereby the number of search points can be reduced from 16 to 4-8. The algorithm is simple and has better coding effection. It is better for application in hardware and real-time system. On the other hand, we use the spatial and temporal correlation characteristic, the spatial and temporal neighboring blocks'motion vectors are employed to predict the direction of the current block motion vectors, which can reduce the search time. This algorithm can reduce the total encoding time by 21%～36%. The coding effection of this algorithm is a little better than the first algorithm, but it's harder to program. You can choose an appropriate algorithm according to your needs.2. Based on statistics, propose an inter-mode selection fast algorithm utilizing prediction and thresholdIn order to achieve the best compression performances, AVS video standard adopts the full-search mode selection method based on rate distortion optimization (RDO) technique. It calculates the rate distortion cost (RDCost) of all possible six modes to choose the best one with the minimum RDCost. Obviously, this mode selection algorithm requires high computational complexity and consumes a large amount of coding time. Our investigation revealed that the probability that the best mode of current macroblock is the same as the best mode of its co-located or surrounding macroblocks varies from 60% to 90% for different video sequences with different motion intensity. So for the current macroblock, we can use the best modes of temporal and spatial adjacent blocks as the candidate mode list, whose number is less than that of the total modes. When one of them satisfies the setted condition, we can skip the other modes. Thereby the encoding time can be reduced substantially.3. Transplant the AVS encoder and fast algorithms to the embedded DM642 platform and the implementation of two video access methods. One method is network transmission. PC transmits the video sequences to be encoded to DM642, with UDP/IP protocol. On DM642, we directly operate the netcard to receive data and return the flag that indicates received data are right or wrong. The other method is to use the video camera to capture videos, namely encoded information from the monitoring system. Encoded data will be transmitted over the network to the host computer which stores them. All of these will do good to promote the industrialization process of the AVS.