Multiway Image Denoising Using Block Diagonal Representation

During acquisition,transmission and storage,images are inevitably contaminated by noise.To reduce the influence of noise on other image processing tasks,image denoising plays an important role.Filtering images of more than one channel or spectral band such as color image,video,multispectral image and magnetic resonance image is relatively challenging in terms of both effectiveness and efficiency.By grouping similar patches to utilize the self-similarity and sparse linear approximation of natural images,recent nonlocal and transform-domain methods have been widely used in multi-channel or-spectral denoising.Many related works focus on the modeling of group level correlation,which often resorts to a recursive strategy with a large number of similar patches.The importance of the patch level representation is understated.This paper proposes to train a global patch basis with block diagonal matrix,along with a local principal component analysis transform in the grouping dimension.A simple transform-thresholdinverse method is applied to enhance sparsity and denoising effects.Fast implementation is also developed to reduce computational complexity.To better evaluate compared methods,we propose a large real-world dataset for color image and video denoising.Extensive experiments on both simulated and real-world datasets demonstrate the proposed method's robustness,effectiveness and efficiency.