Image demosaicking based on chrominance regularization with region-adaptive weights
Osama A. Omer, Toshihisa Tanaka · 2007
A novel method to reconstruct full color images from single color filter array (CFA) data is proposed. This method interpolates the missing color values by minimizing a novel regularization term with region-adaptive weights motivated by the hypothesis that the chrominance is slowly varying within the same object region. The region-adaptive weight, which is deduced from a so-called edge indicator, successfully prevents a demosaicked image from being over-smoothed across an edge. The optimization problem is solved by a combination of steepest decent method with convex projections. The so-generated demosaicked images are compared with ones obtained by four state of- the-art demosaicking techniques in terms of subjective and objective image quality. It is shown that the proposed method outperforms all of them in the case of most of twenty test images. In addition, the proposed method requires less computational cost than the recently proposed alternative projection technique.