Video denoising through sparse representations using trained dictionaries

Abstract

We discourse video denoising problem, in which, it is required to eradicate AWGN from a certain video. For this, we aim to focus on a particular approach which is very promising and extremely efficient, that is, sparse representations using trained dictionaries. To train the dictionary we use KSVD, so that it can efficiently defines the frame/image contents. But, KSVD can only handle small-sized images. So, we describe a global prior of image which introduces sparsity in each patch of the image, and thus, extends the KSVD deployment to images of random sizes. Now we have a simple and proficient denoising algorithm, as a result of this Bayesian treatment. The denoising performance of the proposed methodology is as good as newly published denoising techniques, and at times surpasses them.