Study On Texture And Noise Adaptive Video Coding Algorithms

With the advent of information society, the demand for multimedia communications is growing. Visual communications as the key to the video communication is one of the hotest researches in the field of communications. As the data of video signal is enormous, how to effectively improve the video compression efficiency and performance has become the main problem in video communication. There are high correlations between consecutive video frames and between neighborhood pixels in one frame. The correlations are redundant information for compression.In order to achieve the universal video communications, the International Standards Organization (ISO) and the International Telecommunication Union (ITU) have launched a series of video coding international standards: MPEG-1, MPEG-2, MPEG-4 and H.261, H. 263, H.264, and so on. Analyzing the video coding standards, we can find that the latest video coding standard greatly improves the compression performance by providing a variety of flexible encoding new options. High compression performance of video makes video communication possible even based on channels with very narrow bandwidth. Unfortunately, the flexible encoding options increase the encoding complexity dramatically. The complexity of video encoding algorithm becomes a main factor for real-time video communication. Therefore, in addition to compression performance, the complexity factor of the algorithm should be taken into account seriously when one is optimizing the algorithms of a video encoder. An efficient video coding algorithm can not only save much bandwidth or storage space, but also can save huge complexity.This thesis aims to propose adaptive video coding algorithms with high compression performance and lower complexity. The adaptability of the algorithms is based on the noise information and block edge pattern of the video sequences. The contributions of this thesis are listed as follows.First, two new video characteristic analysis methods are proposed. One is noise level estimation methods and another is block based image pattern calculation. The noise level is the key information for the noise robust fast motion estimation algorithm. The block based image pattern is the key factor for fast mode decision. Then, two fast mode decision algorithms are proposed for an H.264 video encoder. The first one is the fast intra prediction mode decision and the second one is the fast mode decision for inter Macro Blocks (MB). The fast intra prediction mode decision algorithm is based on the block edge direction of the block which is gotten by the image pattern calculation. The fast mode decision for an inter MB is based on the edge pattern of the residual block calculated by the motion compensation with a predicted motion vector (MV). These two mode decision algorithms reduce the encoding complexity significantly while keep almost the same compression performance. Motion estimation is the most complex procedure for a video encoder and it affects the compression performance significantly. Noise in video not only affects the accuracy of motion estimation, but also slows down the procedure of motion estimation. So, at last, this thesis proposes an anti-noise matching criterion for motion estimation and extents a fast full search motion estimation algorithm to noisy video sequences.Based on the analysis of noise level and image edge patterns of the image, the mode decision and motion estimation algorithms proposed have the adaptability to input video. Without decreasing compression performance significantly, the complexity of a video encoder with the two algorithms deceases dramatically. Both the video characteristics analysis methods and the adaptive mode decision and anti-noise motion estimation have broad practical applications in real-time video communications with limited computation resources. This thesis contributes to further popularization for video communications.