A Fuzzy Filter for the Removal of Gaussian Noise in Colour Image Sequences

In this paper a new fuzzy filter for the removal of additive Gaussian noise in colour image sequences is presented. The proposed method, which is a colour extension of our previous work on greyscale images, consists of two subfilters. The first subfilter averages the noise in each of the colour bands separately. In this averaging, the weights of the neighbouring pixels are determined by fuzzy rules. Even though the filtering is performed in each colour band separately, information from the other colour bands is also used in the fuzzy rules. In this way, we can be more confident whether a neighbouring pixel is similar to the pixel being filtered. In order to further improve the results, a second subfilter is applied too. This colour restorating filter is based on the simplified assumption that for similar pixels the pixel value difference should be approximately the same in all colour bands. A pixel is estimated from its neighbours by estimating the differences in each colour band equal to the average over all colour bands. Experimental results show that, in terms of average PSNR and average NCD, the proposed colour extension of our previous work outperforms the usual colour extension of a greyscale filter, in which the Y component in the YUV transform is filtered with the original greyscale method and the chrominance bands U and V are averaged.