Real-world image denoising via weighted low rank approximation

Most of existing denoising algorithms are based on the assumption of additive white Gaussian noise. As the realistic noise in color images captured by CCD or CMOS cameras is much more complex than additive white Gaussian noise, most methods will be not effective. In this paper, we present a weighted low rank approximation for real color image denoising, which effectively models the statistical property of the noise and intrinsic characteristic of the image. Specifically, we employ two weighted matrices to model the realistic noise property along channels and in the spatial dimension in consideration of their different statistics. The intrinsic characteristic of the image is explored via low rank regularization. Then, we formulate the denoising problem into a variational optimization model, which can be solved via the alternating direction method of multipliers (ADMM). Experiments on synthetic and real-world noisy color images show that our proposed method outperforms state-of-the-art denoising methods.