Color Demosaicking using Direction Categorization

We propose a novel color demosaicking algorithm using direction categorization. Each pixel is classified as vertical, horizontal or smooth before interpolation. The categorization is based on gradient change within same channel, color differences and neighbors categories, so it explores relationship between intra-and inter-color channels. Color artifacts in reconstructed images are significantly reduced because directions of interpolation across edges are greatly avoided. Experimental results show that our proposed algorithm has high PSNR and the visual quality of reconstructed images is also obviously improved.