The Image Processing Algorithm Research Based On Multiscale Block Compressed Sensing

Compressed Sensing(CS) Theory stands out in the image compression technologies because it breaks the limit of Nyquist sampling rate. In CS process, signals can be sampled and compressed simultaneously, in which way, the sampling and transport costs are greatly reduced. CS has important theoretical and practical application value in many fields.According to multi-resolution analysis characteristic, multiscale block compressed sensing proposes a multiscale-weighting sampling adjustment strategy based on the work of block compressed sensing and achieves good results. However, it ignored low-frequency coefficients containing a large amount of prior information and gives the same sampling rate among each level. To further enhance the quality of the reconstructed image, taking the importance differences of the wavelet coefficients of each decomposition level into account, a novel adaptive sampling method combining the texture information and directionality is proposed.The proposed algorithm extracts low-frequency coefficients and compute an inverse wavelet transform. Break the obtained image into blocks, then calculate the image entropy to get the initial adaptive subrate and estimate the directionality. Deploy the initial adaptive subrate according to the dominating direction of each block of each subband, achieving adaptive sampling.Experimental results with six different test images reveal that, at each total subrate, the proposed algorithm significantly improves both the reconstruction quality and the visual effect with the increased PSNR and SSIM and the maximum PSNR gain up to 1.38 dB.This paper also does some image reconstruction algorithm research and analysis, including the SPL algorithm adopted in this paper, minimum 1l norm algorithm of the convex relaxation algorithms and the compressed sampling matching algorithm of the greedy algorithm, also gives the simulation results.