On Modeling Interchannel Dependency for Color Image Denoising

摘要

In this article, we study the modeling of interchannel dependency and its application into removing additive noise from corrupted color images. We start from an ad hoc spatially invariant linear interpolative model for characterizing color-difference and demonstrate how it can be jointly used with a variational intrachannel dependency model to suppress noise by alternating projections. Then we present two nonlinear diffusion-based models in the color-difference domain and suggest their spatial adaptation property better matches the class of images with strong chromatic edges. The diversity of interchannel dependency models also motivates us to fuse multiple denoised images from different models to obtain better denoised results. Experimental results are reported to justify the importance of matching the hypothesized dependency model with observation data as well as the benefit of multihypothesis fusion.