Objective Quality Assessment in Color Image Denoising: New Tools and Validation Procedures

摘要

Color image denoising is a very difficult task because color distortion and detail blur should be avoided during noise removal. Current metrics address only a subset of filtering features that should be considered to really assess the quality of a filtered picture. In order to address this issue, a new method for objective quality assessment of filtered images is presented in this paper. The proposed approach operates in the YC_bC_r color coordinate system and is based on the decomposition of the mean squared error (MSE) into six different components. Each component considers a different class of filtering errors affecting the luminance or chroma channels of the denoised picture. Results of computer simulations dealing with color images corrupted by Gaussian and impulse noise show that the proposed approach is effective and can estimate the exact distributions of residual noise, color distortion and detail blur, whereas other metrics cannot.