Single-Image Super-Resolution Using Compressive Sensing

This paper proposes a novel algorithm that unifies the fields of compressed sensing and sparse representations to generate a super-resolution image from a single, low-resolution input along with the use of a training data set. Super-resolution image reconstruction is currently an active area of research, as it offers the promise of overcoming some of the inherent resolution limitations of the imaging systems. In this paper, super-resolution has been achieved by exploiting the fact that the image data is highly sparse over some redundant transforms. Inspired by this observation, we seek a sparse representation for each patch of the low-resolution image, and then use the coefficients of this representation to generate the high-resolution image. The sparsifying dictionary is learned with the use of a training data set that has been obtained from a collection of high resolution images, to generate a global dictionary. The results clearly demonstrate the efficacy of the compressive sensing in image reconstruction. When compared with two of the very popular interpolation based techniques, the proposed method shows much better results, both visually and quantitatively.