Color Image Noise Reduction using Perceptual Maximum Variation Modeling for Color Diffusion

Diffusion is an efficient localized image regularization method based on the analysis of image structures such as direction and magnitude. However, the localization at weak features which have small brightness variations is fundamentally difficult. This often results in removal of weak features. We address this problem with perceptual maximum variation modeling. In our method, diffusion flow of color images is performed by evaluating the perceptual maximum variations which combine the small differences in both brightness and chromaticity, using a least squares optimization with principal component analysis (PCA). A consistency constraint is employed to avoid influence from global color distributions and to enhance homogeneous color regions. We apply our approach for de-noising of color images and obtain excellent improvements over existing methods.