A learned pixel-by-pixel lossless image compression method with 59K parameters and parallel decoding

This paper considers lossless image compression and presents a learned system that can achieve state-of-the-art performance but uses only 59K parameters, which is one or two order of magnitudes less than other systems proposed recently in the literature. The explored based on pixel-by-pixel method, where each pixel’s probability distribution parameters are obtained by processing causal neighborhood (i.e. previously encoded/decoded pixels) with simple neural network comprising parameters. causality causes decoder to operate sequentially, i.e. has be evaluated for pixel increases decoding time significantly common GPU software hardware. To reduce time, parallel algorithms implemented. compared traditional literature terms performance, encoding-decoding times computational complexity.