Color Filter Array Interpolation Using Cellular Neural Networks Considering Self-Congruence

Abstract:
In this paper, we propose a novel color filter array (CFA) interpolation technique based on color difference correlation model. The difference model exploits the interchannel correlation, where the interpolation is achieved by using the color differences R-G and B-G. At first the green channel is handled, and the other color channels are estimated based on the result of green channel. Missing color difference values are estimated by using discrete-time cellular neural network (DTCNN) predictor. First, the DT-CNN transforms color difference values into the optimal coefficients which make possible to establish the optimal prediction using the quincunx A-template. Then, the optimal missing color differences are obtained by using the convolution of the B-template which is derived by rotating the A-template. Moreover, we utilize self-congruence property in order to improve the prediction performance around edges. Experimental evaluation shows that the proposed method has a better performance compared with the conventional method.