The Research Of Image Reconstruct Algorithm Based On Compressed Sensing

Compressed sensing theory breaks through the traditional sampling theorem limitations of sampling frequency.It is a new signal sampling theory which make full use of the signal sparsity.It can finish sampling and compressing at the sametime. If it applies to image compression technology, it can reduce the requirement of storage and data processing ability and make up for the deficiency of the traditional image compression technology. Based on compressed sensing of image reconstruction algorithm,this paper conduct the following research aspects.First, based on the two-dimensional DCT image coefficients energy distribution characteristics, we propose a variable-step compressed sensing image reconstruction algorithm.The algorithm dynamically adjust the atomic step, which conduct search in high amplitude coefficient corresponding atomic subset intensive with small step focus area.Whereas,it search with a big step in the other atoms. The algorithm obtain better image reconstruction accuracy than the original algorithm and reduce the image reconstruction time.Then, the observation and reconstruction of image compression base on the column that it will cut off the correlation between column and column. This paper propose a improved reconstruction algorithm for compressed sensing which is based on the correlation ranks. The algorithm down-sample three high frequency subband coefficients of wavelet transform.and after downsampling the three high frequency subband according to the row and column are observed and reconstructed. Without increasing the quantities of data, the algorithm obtains improved reconstruction quality.Finally, because of the block compressed sensing having the advantage of computing speed and occupying small memory, we make research on it.But it have an obvious fault which have blocking appearance in its reconstructed image.The compression rate adaptive reconstruction algorithm is put forward based on the texture information compression by adaptive learning. Based on the measure of texture information,the algorithm can adjust the size of the sample rate and achieve adaptive sampling rate and sparsity combined with the use of reconstruction algorithm while achieving adaptive sparsity. The experimental results show that the algorithm has obvious advantages in the objective evaluation and subjective visual effect of reconstructed image.