While using compressed sensing to measure and recover images

the approach to divide an image into blocks may improve the speed of reconstruction but will bring strong blocking effects. In order to solve this problem

we propose an adaptive block compressed sensing based on edge detection at the encoder

and a smoothed projected Landweber(SPL) reconstruction algorithm based on principal components analysis(PCA) at the decoder. The reconstruction algorithm uses PCA to train a dictionary adapting to image structure using hard thresholding

thus the image blocking effects are eliminated effectively and the reconstructed image quality is improved. In order to improve the efficiency of the hard thresholding and to reduce the training complexity

three hard thresholding schemes using patch based PCA are proposed: global PCA

local PCA

and hierarchical PCA. As a comparison between our experimental results and the traditional block compressed sen- sing shows

the peak signal-to-noise ratio(PSNR) of the recovered images using the SPL reconstruction algorithms combined with the adaptive block compressed sensing is improved by 13 dB; Regardless of which method is being used at the encoder

the SPL reconstruction algorithms using PCA based hard thresholding are better than SPL algorithms using directional wavelet based thresholding on PSNR.