Video Compressive Sensing Reconstruction Method Based On Frame Block Constraint And Evolutionary Computation

In recent years, it appeared a new data theory of compressive sensing(CS)in the signal processing field, the data acquisition and data compression is achieved at the same time in the theory, it breaks through the limitation of traditional acquisition Nyqurnst sampling theorem and brings a revolutionary change on data acquisition technology, making the theories in compressed imaging system, military cryptography, analog / biological information, wireless sensor and other fields has a wide application prospect. In many fields, high speed camera is playing a more and more important role, but it is a challenge to measure high speed video to the camera’s design. The current main compressed sensing model about video compression perception is that, the video is divided into key frames and non key frames, and Yang Jianbo proposed the Gauss mixture model, the model uses01 observations of video frames, then sum the sensed data of the eight frames before and after, then using mixture gaussian model to reconstruct the observed data, the reconstruction effect in video frame changes before and after which use the method is fuzzy, and this method need to training a class of mean value and variance as the gaussian parameters before it works. Based on this the paper puts forward the video compression perception reconstruction based on evolutionary algorithm, this paper’s main work is as follows:1. A new sampling scheme is proposed. Firstly, the video data is divided into many data sets which is made of 8 frames. According to the two norm of the difference of same location image blocks, we divide them into non changed blocks and changed blocks. For the non changed blocks, we just sense the image block of the first frame in each group, the non changed blocks in other frames are not sensed. For the changed blocks in each group, We sense the all of them. The sensing rate of changed blocks and non changed blocks can be different.2. The video compressive sensing reconstruction method based on the Ridgelet redundant dictionary and genetic evolutionary algorithm is proposed. On the reconstruction of video data, we use the existing image block structure identification algorithm, firstly judge the image blocks. For the each group, we capture the structure characteristics of the block, and then cluster for each category of image blocks, reconstruct video data of each type in each group. In genetic evolutionary algorithm, we put forward the mutation operator based on the statistics of direction and new population initialization scheme. In the process of mutation, selection of atomic minimum directional sub dictionary atoms are replaced, to maintain the population diversity. About the population initialization scheme,for the single direction blocks, the population is initialized in 5 directions. For the image blocks in multi-direction, we mark each image block using its associated optimal 3directions, then cluster them. In each cluster, add up 3 marked directions, we select the 3most directions in the statistics as the optimal relevant direction of the cluster. Compared with the Gauss Mixture model of Jianbo Yang, this paper has a better reconstruction effect,and especially in the changed area of the foreground, its reconstruction effect is better.