In recent years, much research has been conducted on image super-resolution(SR). To the best of our knowledge, however, few SR methods were concerned withcompressed images. The SR of compressed images is a challenging task due to thecomplicated compression artifacts, while many images suffer from them inpractice. The intuitive solution for this difficult task is to decouple it intotwo sequential but independent subproblems, i.e., compression artifactsreduction (CAR) and SR. Nevertheless, some useful details may be removed in CARstage, which is contrary to the goal of SR and makes the SR stage morechallenging. In this paper, an end-to-end trainable deep convolutional neuralnetwork is designed to perform SR on compressed images (CISRDCNN), whichreduces compression artifacts and improves image resolution jointly.Experiments on compressed images produced by JPEG (we take the JPEG as anexample in this paper) demonstrate that the proposed CISRDCNN yieldsstate-of-the-art SR performance on commonly used test images and imagesets. Theresults of CISRDCNN on real low quality web images are also very impressive,with obvious quality enhancement. Further, we explore the application of theproposed SR method in low bit-rate image coding, leading to betterrate-distortion performance than JPEG.
展开▼
