Research On Redundant Dictionary Based Distributed Video Compressed Sensing

In the recent years, with the development of technology, the demand of videofrom people is growing. However, the traditional method of video processing is firstsampling and then to compressed. Meanwhile, as the Nyquist sampling theoremrequired that the sampling frequency should exceed twice as much as the bandwidthof base band signal so as to get the information without distortion. Additionally, thereis a high coding complexity of the construction based on motion estimation in thetraditional method. Thus, the low complexity and effective video coding methodshould be came up with an urgent requirement.In this situation, distributed video coding technology has been brought. Ittransferred compensation part of traditional motion estimation decoder terminal whichcan reduce the complexity of encoder with the cost of coding performance. The theoryof compressed sensing has brought a revolution in the field of signal processing, notonly revise the basic method of Nyquist sampling theorem, but also come up withnew ideas for the research of video coding algorithm.On the basis of existing method of compressed sensing and distributed videoencoding algorithms, started from the correlation between successive frames, thedistributed compressed sensing and video encoding algorithm based on redundantdictionary has been put forward so that launched a new inter joint sparsedecomposition algorithm. In this paper, the main points discussed as follow:(1) Sparse decomposition and reconstruction algorithm for single-frame videoimages based on redundant has been studied. For the captured video from fixedcamera, started from the correlation between successive frames of video sequence, thealgorithm extracts and decomposes the single frame so that reconstruct it based on thebuilt redundant dictionary.(2) Distributed compressed coding algorithm based on background dictionaryhas been proposed. Firstly, the background image from video has been extracted andsparse decomposed. Secondly, the image has been reconstructed based on theredundant dictionary constructed by the background image. Finally, the effectivesparse decomposition of successive video based on background information has beenlaunched so as to lay the foundation for the efficient reconstruction of video sequence.(3) Started from the correlation of successive frames in the video sequence, ajoint sparse decomposing algorithm between frames based on redundant andbackground dictionary has been put forward. Firstly, the foreground and backgroundmodel of video sequence and redundant dictionary have been established, the contentwill be updated through updating formula. Secondly, constructing the redundant dictionary based on foreground and background, detecting and determining whetherthe frame include the foreground target which relied on the difference between currentimage and background image. Furthermore, the classification of foreground target andbackground image has been made based on the mean and variance of sub-block pixel.Thirdly, different method of image reconstruction will be applied according to theclassification result. For the background frame, the image will be updated and sparsedecomposed and then to reconstructed based on the redundant dictionary constructedby background information, the updated dictionary will be numbered and transmittedby compressed sampling algorithm. Similarly, for the video frame includedforeground target, the target will be reconstructed through redundant dictionary builtby background image and the background image will be updated by the redundantdictionary created by foreground image. The dictionary based on foreground andbackground built by sparse decomposition will be joint compressed and sampled. Inthe decoding terminal, the whole video sequence will be got through the imagerebuilding algorithm.