Prediction of optimal operating point for BPG-based lossy compression of images corrupted by Poisson noise
Богдан Коваленко, Владимир Васильевич Лукин · 2023
Lossy image compression becomes a standard way to solve the problem of the rapidly increasing image size that is faced due to better quality of sensors and spatial resolution. Compression of noise-free images is mostly considered. However, images are often corrupted by noise and this property needs to be taken into account in solving the problem of efficient lossy compression. One of the peculiarities of lossy compression of noisy images is the filtering effect that often leads to existence of the so-called optimal operation point (OOP) for which a compressed image in sense of quality is closer to noise free compared to the noisy image. OOP existence can be predicted for some simple noise models such as Gaussian noise for greyscale and color images. Here our study deals with OOP prediction for images corrupted by Poisson noise compressed by Better Portable Graphics (BPG) coder. We demonstrate the possibility of OOP prediction for such metrics as PSNR and PSNR-HVS-M and verify the proposed approach performance using images that have not been used in the preliminary “training”. In addition, the formula for the OOP position determination is presented.