Lossy Compression of Noisy Medical Images

This paper concerns lossy compression of medical images contaminated by noise. It has been shown earlier that, for such a situation, optimal operation point (OOP) might exist and it is worth compressing images in OOP neighbourhood. Meanwhile, OOP existence has been demonstrated for JPEG, JPEG2000, and some compression techniques based on discrete cosine transform (DCT). Relatively novel coders such as HEIF and AVIF have not been studied. Recently, they have been shown efficient for compressing remote sensing noisy images. Here, we consider an opportunity of HEIF/AVIF application to compressing noisy medical images. Additive white Gaussian noise model is investigated as the first step. Possible existence of OOP is demonstrated. Comparison of performance characteristics in OOP is carried out for five coders. It is shown that modern DCT-based coders outperform JPEG significantly and demonstrate similar results.