Multi-Length Directional Interpolation Filter based on Color Correlation for Bayer Pattern Demosaicking

Abstract

In this paper, we proposed a multi-length directional interpolation filter (MLDIF) for color filter array interpolation, where the green channel is interpolated using directional filters which have different filter lengths. In the MLDIF, the filter length is determined using the green pixel correlation between the center and outside regions, and the whole process is non-recursive. To increase the performance of the green channel interpolation, we used a calibration term to compensate for the misguided green pixel values caused by the mismatch between the color channels. Furthermore, we avoided recursive processes, and every interpolation process exploits only one set of weights. By doing this, the complexity burden can be reduced. Experimental results demonstrate that the proposed algorithm shows great improvement in the objective and subjective image quality for 18 images of the IMAX dataset; the proposed algorithm, in particular, successfully suppresses the demosaicking artifacts.