INTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING
Abstract
The most critical issue in the restoration of color pictures from noise is the preservation of the useful information embedded in the image data. The Color Peak Signal-to-Blur Ratio (CPSBR) is a new full-reference method that measures the color/detail preservation yielded by a color image denoising filter. The approach is based on a simple and effective algorithm for the estimation of the filtering blur that operates in the RGB color space. An extensive study of the accuracy of the CPSBR is provided in this paper focusing on two key paradigms for image denoising: the family of order-statistics smoothers and the class of nonlinear weighted average filters. In this framework, the exact values of color distortion and detail blur produced by weighted vector medians, scalar and vector bilateral filters are theoretically evaluated and used for a comparison in order to validate the method. Results of many computer simulations dealing with color images corrupted by different amounts of Gaussian and impulse noise show that the novel CPSBR is a very accurate measure of color/detail preservation.