Abstract
The compressed sensing (CS) has been successfully applied to imagecompression in the past few years as most image signals are sparse in a certaindomain. Several CS reconstruction models have been proposed and obtainedsuperior performance. However, these methods suffer from blocking artifacts orringing effects at low sampling ratios in most cases. To address this problem,we propose a deep convolutional Laplacian Pyramid Compressed Sensing Network(LapCSNet) for CS, which consists of a sampling sub-network and areconstruction sub-network. In the sampling sub-network, we utilize aconvolutional layer to mimic the sampling operator. In contrast to the fixedsampling matrices used in traditional CS methods, the filters used in ourconvolutional layer are jointly optimized with the reconstruction sub-network.In the reconstruction sub-network, two branches are designed to reconstructmulti-scale residual images and muti-scale target images progressively using aLaplacian pyramid architecture. The proposed LapCSNet not only integratesmulti-scale information to achieve better performance but also reducescomputational cost dramatically. Experimental results on benchmark datasetsdemonstrate that the proposed method is capable of reconstructing more detailsand sharper edges against the state-of-the-arts methods.