Multiresolution Approach to "Visual Pattern" Partitioning of 3D Images

This paper deals with the problem of low level representation of 3D image contents. The presented solution makes use of multiresolution techniques to recover the so-called visual patterns or integral features that form images. It consists of decomposing the image into a set of elementary image features, representing frequency channels, using a filter bank, and grouping them by means of clustering analysis. The method introduces a novel design of the bank of oriented scaled filters. In addition, a new measure of dissimilarity between pairs of features is applied to the hierarchical clustering technique.