Adaptive CFA Demosaicking using Bilateral Filters for Colour Edge Preservation

Colour Filter Array (CFA) demosaicking is a process to interpolate missing colour values in order to produce a full colour image when a single image sensor is used. For smooth regions, a higher order of interpolation will usually achieve higher accuracy. However when there is a colour edge, a lower order of interpolation is desirable as it will avoid interpolation across an edge without blurring it. In this paper, a bilateral filter, which has been known to preserve sharp edges, is used to adaptively modify the weights for interpolation. When there is a colour edge, the weights will bias towards a lower order of interpolation using closer pixel values only. Otherwise, the weights will bias towards a higher interpolation for smooth regions. In order to avoid interpolation across a possible edge adjacent to the missing pixel location, four estimates using the adaptive bilateral filter are first determined for each cardinal direction. A classifier comprising a weighted median filter together with a bilateral filter is then used to produce an output of the missing colour pixel value from the four estimates. It has been shown that our proposed method has improved performance in preserving sharp colour edges with minimal colour artifacts, and it outperforms other existing demosaicking methods for most images.