Trio Constrained Adaptive Noise Removal (TCANR) Mechanism for Salt and Pepper Noise in Image Classification

Abstract

Noise removal (also called as denoising) of an image is a vital task in multi-class image classification. Three major shortcomings in Weighted Nuclear Norm Minimization (WNNM) are identified. Firstly, WNNM's patch matching based on the noisy data will considerably augment the risk of patch mismatching. This shortcoming is overcome by performing the grouping task based on noise contentment. Secondly, the fixed feedback percentage which keeps on feeds back ten percent of the residual image to the next iteration despite the consequences of noise levels. This shortcoming is ruled out by incorporating relative feedback mechanism. Finally, the unchanged / constant number of iterations for different noise not considering the distinctions in image content that which will certainly fails to deem the degree of detail in the image. For this variable termination criterion is used. The proposed work is named as Trio Constrained Adaptive Noise Removal (TCANR). Performance metric peak signal to noise ratio (PSNR) is chosen. Four existing methods are taken into account for comparing the proposed TCANR. Extensive simulations are conducted using MATLAB and the results prove that the proposed TCANR performs better in terms of PSNR when compared with the existing methods.