Compression of map images by multi-level context tree modeling

We propose a method for compressing color map images by context tree modeling and arithmetic coding. We consider multi-component map images with semantic layer separation, and images that are divided into binary layers by color separation. The key issue in the compression method is the utilization of inter-layer correlations, and to solve the optimal ordering of the layers. The inter-layer dependencies are acquired by optimizing the context tree for every pair of image layers. The resulting cost matrix of the inter-layer dependencies is considered as a directed spanning tree problem, and solved by algorithm based on the Edmond’s algorithm for optimum branching, and by the optimal selection and removal of the background color. The proposed method gives 50:1 compression on a set of map images, which is about 50% better than JBIG, and 16% better than the best comparative method.