The details of low contrast images are easy to be lost in the process of reconstruction after compressed sensing measurements. To solve this problem, the block compressed sensing reconstruction algorithm incorporating Wiener filtering into the projected Landweber iteration, together with sparse transform using multiscale geometric analysis, contributes to minimizing the loss of the details in the reconstructed images. The numerical experiments select different directional transforms in reconstruction and the results demonstrate that dual-tree complex wavelet transform offers the best performance, while contourlet transform shows superior performance only in high-resolution images. Therefore, appropriate directional transforms based on the contrast of images and time the reconstruction needs should be selected to reconstruct images.
