Post-processing of Interpolated Color Filter Array Images using Modified Mean- removed Classified Vector Quantization

To reduce the cost of digital still cameras, the CFA (color filter array) is usually coasted upon an image sensor. Each pixel of image sensor can sense only one of the R, G, B colors under color filter array. The missing two colors of a pixel have to be estimated from its neighboring pixels. Many color filter array interpolation methods are proposed. However, there are always annoying artifacts presented in the edge and texture areas and these artifacts are hard to be avoided by existing interpolation algorithms. To reduce the artifacts and improve the fidelity of interpolated images, an algorithm using modified mean-removed classified vector quantization (MMRCVQ) to improve the missing colors is proposed. The algorithm extends and modifies vector quantization (VQ) to discover the relationships between differential images of G channels and their CFA versions. The discovered relationships are stored in a codebook and are used to refine the G channels of CFA images. The experimental results show that the proposed approach can reduce the artifacts of interpolated images effectively. In terms of PSNR, the average improvements for R, G, and B channels are 0.71 dB, 0.54 dB, and 0.61 dB, respectively.