ANALYSIS OF DISTORTIONS DUE TO BPG-BASED LOSSY COMPRESSION OF NOISE-FREE AND NOISY IMAGES

Abstract

Compression algorithms have become a popular way to deal with the problem of rapidly increasing data volume produced by images from different sources. These algorithms can be divided into two groups called lossless and lossy compression. In the case of providing a high compression ratio, lossy compression is needed. The paper utilizes Better Portable Graphics (BPG) as one of the promising lossy compression coders. For compression procedures, noise presence plays an important role in the sense that noise can affect compression performance and resulting image quality. This article deals with an analysis of the statistical characteristics of distortions introduced due to image lossy compression by the BPG coder.

The subject of this paper is the statistical characteristics of distortions introduced due to lossy compression of noise-free and noisy grayscale images. The goal is to analyze the characteristics of distortions introduced by the BPG coder and provide information about residual noise that can appear due to the compression of noisy images. The task is to calculate the statistical characteristics of distortions for several cases (noise-free vs. compressed; noise-free vs. noisy-compressed; noisy vs. noisy-compressed), create histograms, and analyze their behavior.

The main results are the following. The properties of the introduced distortions have to be studied for the reasonable setting of the coder parameters and control of compressed image quality. We show that the distortions might have distribution close to a Gaussian but it can be also heavy-tailed. This depends on several factors including the noise presence in an image to be compressed, image complexity, and the coder parameter Q. The noise presence leads to specific properties of introduced distortions and residual noise that can be useful for further post-filtering of compressed noisy images.