HIERARCHICAL IMAGE SEGMENTATION BY MULTIDIMENSIONAL CLUSTERING AND ORIENTATION-ADAPTIVE BOUNDARY REFINEMENT

In this paper we present a new multi-dimensional segmentation algorith m. We propose an orientation-adaptive boundary estimation process, emb edded in a multiresolution pyramidal structure, that allows the use of different clustering procedures without spatial connectivity constrai nts. The presence of noise in the feature space, mainly produced by mo deling errors, causes a class-overlap which can be reduced in a multir esolution pyramid. At the coarsest resolution level, the separation be tween the different classes is increased and the within-class variance reduced. Thus, at this level, the classes can be obtained with differ ent multi-dimensional clustering algorithms without connectivity const raints. Small and scattered classes as well as isolated class labels a re reassigned to their neighborhood by a process which guarantees the spatial connectivity. The resolution is then increased by projecting d own the class labels. At each level, the borders are improved by reass igning the boundary pixels to their spatially closest class. However, the class-uncertainty astride the borders has first to be reduced, and we propose to do this by means of orientation-adaptive butterfly-shap ed filters. This refinement process further eliminates spatially miscl assified pixels produced by the unconstrained clustering. Experimental results show that similarly accurate boundaries are obtained with dif ferent clustering algorithms for various test images.