Lossless Image Compression Using a Multi-scale Progressive Statistical Model

Lossless image compression is an important technique for storage and transmission when information loss not allowed. With the fast development of deep learning techniques, neural networks have been used in this field to achieve a higher rate. Methods based on pixel-wise autoregressive statistical models shown good performance. However, sequential processing way prevents these methods be practice. Recently, multi-scale proposed address limitation. Multi-scale approaches can use parallel computing systems efficiently build practical systems. Nevertheless, sacrifice performance exchange speed. In paper, we propose progressive model that takes advantage approach approach. We developed flexible mechanism where order pixels adjusted easily. Our method outperforms state-of-the-art lossless two large benchmark datasets by significant margin without degrading inference speed dramatically.