A novel infrared image super-resolution method based on sparse representation

This paper presents a novel method for the reconstruction of super-resolution infrared images based on sparse representation. Assume we can get a pair of dictionaries which makes the low-resolution image patches share the same sparse representation with high-resolution image patches, then the high-resolution image patches can be reconstructed through the sparse representation of low-resolution image patches. Firstly, considering the instability of the sparse decomposition and the self-similarity of the image patches, the stable multi-dictionary pairs can be obtained by training the classified samples twice. Then, in order to get more accurate sparse coefficients, detail patches are selected as the object of our method. Finally, high-resolution infrared images can be reconstructed. In addition, some non-uniform images with fixed pattern are added into training samples, allowing our algorithm robust to infrared imaging system. The proposed method can greatly reduce the reconstruction artifacts caused by the instability of the sparse representation and have a high reconstruction precision. (C) 2015 Elsevier B.V. All rights reserved.