Sub-band Filtering for MR Texture Segmentation

We consider the problem of classifying textured regions. First, several artificial and natural textures will be used to describe the frequency and orientation multiresolution sub-band filtering. Next, Magnetic Resonance images will be used to discriminate anatomical structures. A high resolution MR data of a knee is filtered with the proposed second orientation pyramid of several levels.The filtered results were used as feature vectors as input to a supervised k-means. Results of the segmented pixel regions that represent bone, tissue, muscle and background are finally presented.