Analysis of Statistical and Spatial Spectral Characteristics of Distortions in Lossy Image Compression

Image lossy compression is currently widely employed in different fields. Compared to lossless compression, it allows providing a considerably larger compression ratio but distortions are introduced inevitably. Properties of these distortions depend on a used coder, an image subject to compression, and compression parameters. Distortions affect visual perception of compressed images and efficiency of their further processing. Thus, statistical and spatial spectral properties of introduced distortions have to be understood well to explain effects observed in compressed image processing and, possibly, enabling their simulation. In this paper, we describe the tools that can be used in analysis of distortion properties and give an example of such analysis for grayscale images compressed by AGU coder. It is shown that distortions, in general, have spatially variant statistics where they are more intensive in locally active areas. Besides, distortions are spatially uncorrelated for all considered quantization steps and, respectively, compression ratios. Experiments are carried out using four typical remote sensing images.