Lossless Image Compression Using List Update Algorithms

Arezoo Abdollahi, Neil D. B. Bruce, Shahin Kamali, Rezaul Karim

Abstract: We consider lossless image compression using a technique similar to bZip2 for sequential data. Given an image represented with a matrix of pixel values, we consider different approaches for linearising the image into a sequence and then encoding the sequence using the Move-To-Front list update algorithm. In both linearisation and encoding stages, we exploit the locality present in the images to achieve encodings that are as compressed as possible. We consider a few approaches, and in particular Hilbert space-filling curves, for linearising the image. Using a natural model of locality for images introduced by Albers et al. [J. Comput. Syst. Sci. 2015], we establish the advantage of Hilbert space-filling curves over other linearisation techniques such as row-major or column-major curves for preserving the locality during the linearisation. We also use a result by Angelopoulos and Schweitzer [J. ACM 2013] to select Move-To-Front as the best list update algorithm for encoding the linearised sequence. In summary, our theoretical results show that a combination of Hilbert space-filling curves and Move-To-Front encoding has advantage over other approaches. We verify this with experiments on a dataset consisting of different categories of images.