Single image super-resolution based on compressive sensing and TV minimization sparse recovery for remote sensing images

Abstract

In this paper we address the problem of super resolution in remote sensing images from a single low resolution image without using an external database. This method uses the techniques of Compressive Sensing (CS), Structural Self Similarity and Total Variation (TV) Minimization. The approach is based on sparse and redundant representations over trained dictionaries. The method involves identifying a dictionary that represents high resolution patches in a sparse manner. Extra information from similar structures that exist in remote sensing images can be introduced to dictionary in the CS framework. K-SVD method is used for finding the dictionary and TV Minimization method is used for finding the sparse representation coefficients. Instead of using the HR patches from an external database, the proposed method uses the patches from the interpolated version of the LR image for training the dictionary. The method is compared with other single image super resolution algorithms that use sparse recovery methods such as Orthogonal Matching Pursuit algorithm. The proposed method is tested with satellite images from USC_SIPI database. The method gives better results than other methods both visually and quantitatively. Performance of the method is evaluated using the metrics: PSNR, MSSIM, FSIM and Blur Metric.