In this post, we will introduce two broad categories of image/video quality models, and discuss two algorithms that fall under a third, hybrid, category. We will heavily borrow from concepts of Natural Scene Statics (NSS) and Information theory discussed in my earlier post on VIF.
This article is a review of existing texture characterization, identification and segmentation methods. These methods typically involve statistical methods, i.e. inferring from histograms or co-occurrence matrices, and/or signal processing methods, either simple filtering and energy-based methods, or as a method of pre-processing before applying statistical analysis.
This article is a review of a popular Image Quality Model - Visual Information Fidelity (VIF) [1] which is all based on a statistical model of natural scenes. To do this, we will first review a powerful natural scene statistics (NSS) model, the Gaussian Scale Mixture.