Modified Smoothed Projected Landweber Algorithm for Adaptive Block Compressed Sensing Image Reconstruction

Abstract

In this article, according to the block compressed sensing (BCS) framework, a novel adaptive sampling method is proposed. First, the statistical information of image block is calculated. Then, based on the statistical information of image block and DCT coefficients, a weighted allocation factor is constructed and each block is categorized into the complex block or the simple block. And according to the different types of blocks, the sampling rate can be allocated adaptively. For validating the effectiveness of proposed sampling method and further improving the efficiency of BCS reconstruction, the modified smoothed projected Landweber (MSPL) algorithm is proposed to speed up comparatively slow rate of convergence. The results of the experiment demonstrate that the proposed method (ABCS-MSPL) significantly improves the peak-signal-to-noise ratio (PSNR) and running efficiency compared to other advanced methods for this purpose.