Color reproduction for noisy CFA data using directional cycle-spinning

This paper addresses color filter array (CFA) color reproduction problem where the aim is to utilize an image captured by the CFA to produce an image with full color information. First, conventional subband synthesis based color reproduction techniques do not consider the noise during image acquisition and assume that the CFA data are noiseless. To tackle the noisy CFA data, a novel approach is proposed by inserting a subband denoising scheme into the conventional subband synthesis framework. Second, conventional subband synthesis based techniques exploit the decimated wavelet transform that is not shift-invariant and could result in ringing artifacts in the result. To alleviate these artifacts, the directional cycle-spinning (DCS) technique is exploited. Furthermore, a new cycle-spinning pattern is proposed according to the sampling pattern of the Bayer CFA data. Extensive experiments are conducted to demonstrate that the proposed approach outperforms several approaches.