Blind Evaluation of Noise Characteristics in Multichannel Images

In this chapter, the ways to improve the accuracy of blind estimation of noise characteristics in multichannel images are considered. The main idea is to process several channel images jointly in order to suppress the influence of image content. The focus is a method based on the analysis of parameters of DCT coefficients’ distributions obtained for different spatial frequencies in 8 × 8 pixel blocks. Several modifications of this method for color and hyperspectral images are described. Based on the results of numerical simulation obtained for a large database of test images, as well as according to the results of testing on real remote sensing images, it is shown that joint processing of images with high levels of inter-channel correlation allows decreasing the estimation bias by up to several times.