Research Of Vector Quantization And Video Segment Based On Conditional Random Field

With further development of information technology, images and videos become anincreasingly significant part of information source. As a result of large data of original source,it calls for much more bands. To store and to transport original data directly is not only awaste of resources which will limit their application, but also not to show agreement withstereoscopic display. Therefore, we need to compress them. What’s more, data redundancy invideo frames (images) and inter-frames make it possible to compress them. Regarding of allthese above, this paper takes research on image coding and video segmentation.Firstly, we analyze the research status of image coding and video segmentation, pointingout the deficiencies of existing algorithms. And then we study them, eventually proposedcodebook design algorithm of Vector Quantization (VQ) based on improved Particle SwarmOptimization (PSO) and Video Segmentation algorithm based on the Partitioning ProcessingConditional Random Fields.In image coding, this paper focuses on codebook design of VQ, with regard of PSO.Innovatively, this paper presents a pair of parallel particle swarm optimization algorithmwhich can be used to get a good codebook in the vector quantization of video coding. Withusing the two populations evolve separately and monitoring each other, when one or twospecies fall into local optimum, it can jump out of local optimum through interaction. Bysorting the training vector, we select the initial codebook to enhance the diversity of search.And then, we use Equal-Average Nearest Neighbor Search (ENNS) to realize coding of inputvectors. Finally, we use vector quantization to realize image segmentation, and thesegmentation results can be used as pixel-level segmentation of video segmentation.While in video segmentation, to reduce the time complexity of the traditional videosegmentation based on conditional random fields (CRF), a partitioning processing methodwas proposed. Firstly, taking full advantage of the correlation between the pixels among ablock, the algorithm partitions the original frames to form new frames. Then we segmentthem by CRF. Finally, the video sequences of the original size are recovered by the initialsegmentation, the result of previous frame and the result of current frame after CRFsegmentation. Experimental result shows that the algorithm can reduce the time complexity effectively without increasing the false rate of the video segmentation obviously.Finally, we design computer simulation experiments according to framework depictedabove. Regard of experimental results, we compare various aspects of image coding andvideo segmentation and obtain performance indicators accordingly. Result shows: VQ basedon IPSO can gain PSNR with different degree, while video segmentation algorithm canreduce the time complexity effectively without increasing the false rate of the videosegmentation obviously.