DL-CSNet: Dictionary Learning based Compressed Sensing Neural Network

Abstract In this paper, we propose a novel neural network for Compressed Sensing (CS) application: the Dictionary Learning based Network (DL-CSNet). It is fairly simple but highly effective, which consists of only three layers: 1) DL layer latent sparse features extraction; 2) smoothing via Total Variation (TV) like constraint; and 3) CS acquisition training. particular, TV-like perfect complement to sparsity-oriented achieve smooth images. The trained DL-CSNet can learn optimal dictionary matrix so that images be reconstructed in high quality. At last, extensive experiments have been carried out on binary compared most classical algorithms, shows superior performance proposed DL-CSNet.