Joint Image Denoising and Demosaicking by Low Rank Approximation and Color Difference Model

Xia Zhai¹, Weiwei Guo², Yongqin Zhang³, Jinsheng Xiao⁴, and Xiaoguang Hu⁵

¹Arts Department, Henan University of Economics and Law, China

²China Unicom in Chongqing Branch, China

³School of Information Science and Technology, Northwest University, China

⁴School of Electronic Information, Wuhan University, China

⁵School of Criminal Science and Technology, Peopleʼs Public Security University of China, China

 Abstract: Digital cameras generally use a single image sensor which surface is covered by a color filter array. The Color Filter Array (CFA) limits each sensor pixel by sampling one of the three primary color values (red, green or blue), whereas the other two missing color values would be acquired by the post-processing procedure called demosaicking. From the noisy CFA data, the full color images are reconstructed through an imaging pipeline of demosaicking and denoising. However, image denoising in the RGB space has expensive computation cost. In this paper, to increase the efficiency and the color fidelity, we propose a novel joint denoising and demosaicking strategy to reconstruct the noiseless full color image from the input noisy CFA data. The low-rank approximation technique is first used to remove the noise from CFA data. Then, image demosaicking using both color difference space and signal correlation are applied to the denoised CFA data to obtain the noise-less full color image. The experimental results show that the proposed algorithm not only improves the quality of full color image but also outperforms the existing state-of-the-art methods both subjectively and objectively.

Keywords: denoising, image demosaicking, CFA, low rank approximation, color difference model.