Non-Local Image De-noising and Post Processing Using KL Transform

Basically there are two types of image de-noising methods such as Local means algorithms and Non-Local means algorithms. In first case, to restore the intensity of particular pixel only the local neighborhood of pixel being processed is used whereas in second case the entire image is taken into account to restore the intensity of particular pixel. Former case makes assumption about the frequency content of the image whereas later case assumes that image contains an extensive amount of redundancy, which means to say that the image has large number of similar patterns. The assumption made by non-local method works very well for images which are having no much grey level transitions or the edges. For pixels that fall near the edges of the image, there are less numbers of pixels with the similar neighborhoods. Because of this reason we see blurred de-noised image near the edges. To avoid this blurring effect we are introducing a local post processing filer which is based on KL transform. This post processing method is more effective for images that are corrupted by higher noise variances, for example image corrupted by random noise with noise variance, . In this paper a non-local means algorithm is implemented for standard database images such as Lena image, Barbara image, Baboon image, Couple image etc. Fig 1: Block diagram of Image De-noising and Post Processing. This paper is organized as follows: Section I gives brief introduction, section II deals with basic Non-Local means algorithm and noise model, which generate Additive White Gaussian Noise (AWGN) (uncorrelated), section III deals with implementation of post processing filter, its pseudo code, section IV is about results and discussions while conclusion is given in section V.