Demosaicking filter using modified color conversion matrix

A large number of demosaicking methods have been proposed. However, the algorithms get more complicated. In this paper, we proposed an efficient demosaicking filter using modified conversion matrix and the CFA (color filter arrays) pixels. The CFA image is first directly expanded into 3 full layers of R, G and B, where the missing values for each color are set as unknown numbers. The missing values are then produced by the proposed 3×3 filter to obtain temporary full RGB layers. The temporary full RGB layers are then used to produce Cb and Cr layers. To improve the image quality, the Cb and Cr layers are further filtered by the bilateral filter. To obtain the final full RGB layers, based on the standard YCbCr-to-RGB matrix conversion, we proposed to use a novel CfCbCr-to-RGB matrix conversion. For each location in an image, the Cf is the available CFA pixel value, which can be R, G or B, therefore there are 3 corresponding conversion matrix. Compared with the state-of-the-art methods using the Kodak dataset, the average improvement CPSNR gains are 2.22dB, 2.03dB and 1.54 dB.