Analysis of noisy image lossy compression by BPG using visual quality metrics
Богдан Коваленко, Владимир Васильевич Лукин, Виктория Владимировна Науменко, Sergey Krivenko · 2021
Image lossy compression is widely used nowadays. A common assumption is that images subject to compression are noise-free. If intensive (visually noticeable) noise is present, lossy compression has several peculiarities that have to be taken into account. These peculiarities have been earlier studied for several coders, but they have not yet been analyzed for Better Portable Graphics (BPG) coder proposed recently and controlled by a quality parameter Q. In this paper, performance of this coder is analyzed for the case of additive white Gaussian noise using two known visual quality metrics, PSNR-HVS-M and MS-SSIM. It is demonstrated that for both metrics it is possible that optimal operation point exists. This usually happens for quite simple structure images and/or if noise intensity is high enough. Based on simulation results obtained for a set of grayscale images initial recommendations concerning Q setting are given for practice.