Lossless Image Compression using Reversible Integer Wavelet Transforms and Convolutional Neural Networks

Abstract

Image compression is an area of data compression which looks to exploit various redundancies that exist within images to reduce storage and transmission requirements. In information critical applications such as professional photography, medical diagnostics, and remote sensing, lossless image compression may be used to ensure the original data can be restored at a later time. In this work, a lossless compression framework is proposed which incorporates Convolutional Neural Networks (CNNs) to predict wavelet detail coefficients from coefficients within neighboring subbands. The main premise of the proposed framework is that information which can be recovered at the decoder via CNN prediction can be excluded from the compressed codestream, resulting in reduced file sizes. An end-to-end encoder and decoder is implemented to test the validity of the proposed, model and compression performance is compared with current state of the art methods.