Using dendronal signatures for feature extraction and retrieval

A dendrone is a hierarchical thresholding structure that can be automatical ly generated from a complex image. The dendrone structure captures the conn ectedness of objects and subobjects during successive brightness thresholdi ng, Based on connectedness and changes in intensity contours. dendronic rep resentations of objects in images capture the coarse-to-fine unfolding of f iner and finer detail, creating a unique signature for target objects that is invariant to lighting, scale, and placement of the object within the ima ge. Subdendrones within the hierarchy are recognizable as objects within th e picture. Complex composite images san be autonomously analyzed to determi ne if they contain the unique dendronic signatures of particular target obj ects of interest. In this paper, we describe the initial design of the dend ronic image characterization environment (DICE) for the generation of dendr onic signatures from complex multiband remote imagery. By comparing subdend rones within an image to dendronic signatures of target objects of interest , DICE can be used to match/retrieve target features from a library of comp osite images. The DICE framework can organize and support a number of alter native object recognition and comparison techniques depending on the applic ation domain. (C) 2001 John Wiley & Sons, Inc.