On the size and shape of multi-level context templates for compression of map images

We present a method for estimating optimal context templates that are used for conditioning the pixel probabilities in context-based image compression. The algorithm optimizes the location of the context pixels within a limited neighborhood area, and produces an ordered template as a result. The ordering can be used to determine the shape of the context template for a given template size. The optimal template size depends on the size of the image, when the template shape depends on the image type. We apply the method to the compression of multi-component map images consisting of several semantic layers represented as binary images. We estimate the shape of the context-template for each layer separately, and compress the layers as generic regions using the Joint Bi-level Image Group standard JBIG2 compression technique