Skip to Main Content
We propose a method for filtering raster map images by context tree modeling. This is a two-pass method. At the first pass, the filter utilizes statistical information about the image spatial structure and stores the statistics in a tree structure. At the second pass, these statistics are used in actual filtering to calculate the probability of the current pixel in its neighborhood. We test this method on a set of map images. We use different noise models to evaluate the performance of the proposed filter. Finally we compare the performance results for our method with the vector median filter. The proposed method does not destroy the object borders and outperforms the vector median filter for a moderate level of noise.