An Image Super-Resolution Scheme Based on Compressive Sensing with PCA Sparse Representation

Abstract

Image super-resolution (SR) reconstruction has been an important research fields due to its wide applications. Although many SR methods have been proposed, there are still some problems remain to be solved, and the quality of the reconstructed high-resolution (HR) image needs to be improved. To solve these problems, in this paper we propose an image super-resolution scheme based on compressive sensing theory with PCA sparse representation. We focus on the measurement matrix design of the CS process and the implementation of the sparse representation function for the PCA transformation. The measurement matrix design is based on the relation between the low-resolution (LR) image and the reconstructed high-resolution (HR) image. While the implementation of the PCA sparse representation function is based on the PCA transformation process. According to whether the covariance matrix of the HR image is known or not, two kinds of SR models are given. Finally the experiments comparing the proposed scheme with the traditional interpolation methods and CS scheme with DCT sparse representation are conducted. The experiment results both on the smooth image and the image with complex textures show that the proposed scheme in this paper is effective.