Video Compressed Sensing Algorithm Based On The Inter-frame Correlation

Compared with other signals, an obvious characteristic of the video signal is thatthere is a strong inter-frame correlation. For most of those existing video compressedsensing，an image compressed sensing method is ususlly utilized for the single frame.The measurement process is frame by frame. So is the reconstruction process. Thesealgorithms can’t obtain a satisfactory reconstruction quality because they underestimatethe truth of the inter-frame correlation. To address this issue, video compressed sensingalgorithms based on inter-frame correlation are studied. The main contents can besummarized as follows:Firstly, a novel adaptive framework with variable sampling rate is proposedaccording to the distribution characteristics of the correlations between neighboringframes. It divides the frame into some blocks. Then it classiifes blocks into different typedepending on their inter-frame correlation, and adjusts the sampling rate accordingly. Forthose less relative blocks carrying a sheer volume of information, high sampling rateshould be allocated. Conversely, low sampling rate should be issued to blocks with a highrelativity in that reference frames can provide adequate information enablingreconstruction. Our simulation results show that the measurement method with variablesampling rate can improve the utilization ratio of sampling rate in an effective way.Secondly，in order to improve the utilization ratio of inter-frame correlation andmake sure that the reconstruction algorithm is suitable for the measurement mode withvariable sampling rates, this paper makes some provements of the multihypothesispredictions reconstruction algorithm. Our simulation resultes show that the videocompressed sensing with variable sampling rates measurement and variable samplingrates multihypothesis predictions reconstruction algorithm reduces the number ofsampled measurements while still improving the quality of the reconstructed frames.Finally, the paper aims to use group sparsity concept in the video reconstructionalgorithm so that it can utilize the inter-frame correlation more adequately. A fast iterativeshrinkage-thresholding algorithm based on group is proposed. It searchs matching blocks according to the similarity and then reconstructs the matching blocks group usingreconstruction algorithm with group sparsity. Our simulation resultes show that thereconstruction algorithm with group sparsity can realize utilization of correlationbetween blocks，and acquire a high reconstruction qulity of video.