Region adaptive color demosaicing using cross-channel correlation

In this paper, we consider the problem of recovering full-color images from color-sampled observation. Each pixel location is classified into flat, edge and pattern region and edge indicator function is used to avoid artifacts in high frequency regions and improve the performance. The horizontal and vertical direction of edge is decided using local statistics based on the concepts of spectral-spatial correlation. As a postprocessing, the modified filtering on color difference domain is adopted to improve the quality of the image. Experimental results illustrate the benefits of the proposed method. When compared to the conventional methods, the proposed method outperforms them on quantitative and qualitative criteria.