Non-separable Haar wavelets and the clustering of magnetic resonance imagery

Non-separable Haar wavelets and the resulting multiresolution decomposition of the toroidal integer lattice provide a mechanism to cluster, i.e., obtain an unsupervised classification, an image. The statistical nature of imagery obtained by the measurement of magnetic resonance makes context free classification less than ideal. The technique proposed uses a definition of homogeneity defined by the tilings imposed by the non-separable Haar wavelets to obtain an estimate of the distribution over irregular structures in the image. Subject to the choice of the tile, a clustering is obtained which represents regions in the image which share similar distributions up to a specified significance level.