A novel segmentation based video-denoising method with noise level estimation

Abstract

Most state-of-the-art video-denoising algorithms assume an additive noise model, but such a model does not often reflect true conditions experienced in practice. In this paper, two main issues are addressed, namely, segmentation-based block matching and estimation of noise level. Unlike previously reported block-matching methods, the present method uses an efficient algorithm to perform block matching in spatially consistent segmentations of each image frame. To estimate the noise level function (NLF), which describes the noise level as a function of image brightness, a fast bilateral-median-filter-based method is proposed herein. Under the assumption of short-term coherence, this method of estimation is extended from a single frame to multiple frames. Coupling these two techniques together creates a segmentation-based, customised BM3D method that can be used to remove coloured multiplicative noise from videos. Experimental results obtained for benchmark data sets and real videos show that this method significantly outperforms other methods in removing coloured multiplicative noise.