On the use of context tree for binary image compression

We consider the use of a static context tree for binary image compression. The contexts are stored in the leaves of a variable-depth binary tree. The tree structure itself is fully static and optimized off-line for a training image. The structure of the tree is similar for different images of the same type. The benefit from optimizing the tree for each input image separately is usually overweighed by the overhead required from storing the tree structure. The static approach is therefore applicable in most situations as the compression can be performed much faster and during a single pass over the image.