Deep-Learned Regularization and Proximal Operator for Image Compressive Sensing

Deep learning has recently been intensively studied in the context of image compressive sensing (CS) to discover and represent complicated structures. These approaches, however, either suffer from nonflexibility for an arbitrary sampling ratio or lack explicit deep-learned regularization term. This paper aims solve CS reconstruction problem by combining term proximal operator. We first introduce a using carefully designed residual-regressive net, which can measure distance between corrupted clean set accurately identify subspace belongs. then address operator with tailored dilated residual channel attention enables learned map distorted into set. adopt adaptive selection strategy embed network loop algorithm. Moreover, self-ensemble is presented improve recovery performance. further utilize state evolution analyze effectiveness networks. Extensive experiments also demonstrate that our method yield superior accurate (PSNR gain over 1 dB) compared other competing approaches while achieving current state-of-the-art The test code available at https://github.com/zjut-gwl/CSDRCANet.