Derivation, experimental verification, and applications of a new color image model

Abstract

Image modeling is an important area of image processing. Good image models are useful, for example, in image restoration problems because they provide constraints that can be imposed on degraded images to retrieve better approximations of the original image. Many physical models of images are separable functions of position and wavelength, which means that images can be written as a color independent local average multiplied by a color dependent residual. We will present experimental results showing that this is the case in practice, and discuss the limitations of this model. We will also show that several commonly used observations in image processing follow from this model. Finally, we will demonstrate the results of imposing the model constraints in several image denoising problems and show that degraded images can be improved by imposing the model constraints.