Image denoising method by endorsement of neighborhood pixels

Abstract

Noise removal from an image is yet very hot area in image processing. It is a vital preprocessing step in many applications. The objective of image denoising is to estimate a clean image from a noisy observation. In this context noise is defined to be a disturbance in the observed signal, leading to an inaccurate measurement of the observed quantity and thus to a loss of information. In this paper, a denoising algorithm is proposed which works blindly i.e. without any prior information about the noise variance. Input image is divided into 3×3 sized patches and similar patches are searched in the neighborhood. Original value of a pixel is estimated by endorsing neighborhood pixels. Endorsement is decided according to the degree of similarity between the pixel under consideration and pixels around it. Significance of the proposed technique is verified by comparing it with other state of the art techniques, qualitatively and quantitatively.