New perspectives on image compression using a cartoon - texture decomposition model

A new method for multilayered image representation and its application in image compression are presented. The basic idea is to separate an image into two different layers having fundamentally different structures. Nonlinear diffusion is used to smooth out image texture in order to obtain a large scale piecewise smooth image representation called "cartoon". The residual part is considered as "pure texture". Since the best approximations of these two structurally different image layers can be realised by different transforms, higher compression ratios are achieved using suitably chosen transforms prior to individual encoding of each layer. Transform basis tailored to each layer are derived from a comprehensive empirical study. Several experiments were conducted to evaluate the performance of the proposed approach. The results show that in most cases the proposed technique achieves objectively slightly better and subjectively superior results than well-established image compression algorithms.