Compressed Sensing Reconstruction Methods On The Stereo Video Residuals

The video coding technology has experienced more than half a century ofdevelopment, people have been exploring simple, fast and efficient encoding method.From a single viewpoint, binocular stereo to multi-view video coding, changes inmarket demand, making it no longer satisfied with a single video image, but tend tobe vivid three-dimensional video, however, each additional point of view inevitablyincrease the cost of acquisition, transmission bandwidth and storage capacity. Themain purpose of the compressed sensing is to capture terminal target image and thelinear measurement combination of greatly reducing the sampling rate, the decodercan get a small amount of measured values is the most optimal solution to restore thetarget image. Compressed sensing makes the three-dimensional video capture,transmission, storage cost reduction is possible.In this paper, for the compressed sensing measurement matrix most difficult tohardware implementation proposed a diagonal block column vector of random sortingof measurement matrix. The measurement matrix only0,1two elements, sparse highly.When the number of its sub-matrix of two or more, each sub-matrix of the columnvector of independent random permutation column vector between unrelated, andprove that the mutual coherence between the measurement matrix and the natureimage sparse to meet the measurement matrix selection criteria. The experimentalresults show that the construction method of this article in the performance close to arandom Gaussian matrix, more than random Gaussian matrix sparse and easyhardware implementation.The search path of Gradient Pursuit is jagged, its fast convergence, but not optimalfeatures, the introduction of the concept of threshold. Each approximation signal andthe original signal as a match to find the matching degree, when the match is less thanthreshold shows the signal with the original signal a lot of difference, which is toselect the larger of the BB step, that is approximated by the step to ensure theconvergence speed. On the contrary, has approximately close to the goal, to improveaccuracy, the steepest descent step shortened to improve the accuracy of the algorithm.It can be seen from the experimental data, the improved algorithm on thetime-consuming close to the Gradient Pursuit method, but the quality of thereconstructed signal equivalent to the Orthogonal Matching Pursuit.Static image restoration and reconstruction, confirmed the feasibility of themeasurement matrix and the improved algorithm. This article will be compressedsensing proposed, which combines the three-dimensional video encoding stereoscopicvideo coding scheme based on compressed sensing and proposed a video codingscheme based on the residual sparse block of compressed sensing.We need dividedthe video sequence into groups, the first frame of the direct use of block compressed sensing reconstruction as a reference frame. Second frame could not help directly inthe measured value reconstruction, but with the current frame to find the bestmatching block in the reference frame obtained motion compensation, and thenpredict the approximate current frame. Then the current frame and the predictionframe residuals, residuals and motion compensation block compressed sensingreconstruction. The first frame plus residuals and compensation, find the secondframe. Then the second frame as a reference frame, the reconstruction of the third one.The previous one plus residuals and compensation for the following frame. Disparityestimation is similar to the characteristics of the motion estimation, this method isused in three-dimensional video procession which can reconstruct the right point ofthe view video.