Color Image Super Resolution by Using Cross-Channel Correlation.

Single image super resolution (SR) aims to reconstruct a high resolution (HR) image from a low resolution (LR) image. However, many SR methods are only designed for the grayscale images. As a result, when dealing with the color images, the cross-channel information is often ignored by those approaches. In this paper, we propose a color image SR method by taking the cross-channel correlation of color images into consideration. In our method, the gradients of the differences between color channels are required to be sparse. In addition, to make our SR framework more robust, the average signal of the three color channels is also enforced to be sparse in the gradient domain. Finally, to solve the optimization problem which includes the cross-channel-correlation-based regularization terms, an efficient algorithm is presented. Experimental results demonstrate the effectiveness of the proposed method quantitatively and visually.