Object-oriented Image Analysis by Evaluating the Causal Object Hierarchy of a Partitioned Reconstructive Scale-space

Mathematical morphology interprets images as unions of individually shaped and superpositioned coherent structures. Within this framework, reconstruction filters preserve shapes of image structures that correspond to given base structures, while other structures are removed. This shape-selective transform enables consistent processing of visually perceivable image structures, e.g. of those image entities that have a semantic interpretation for a specific observer in a certain context. Such content-based processing can be regarded as being object-oriented. Also, it is commonly accepted that high-level image processing needs to consider appropriate ranges of scales. For these reasons, a morphological multiscale approach to object-oriented image analysis is established by a reconstructive scale-space preserving the inflection contours of an image, each of which represents a distinct object. Towards coarser scales the objects merge in a causal way and finally form a completely partitioned scale-space with an inclusion relation between fine and coarse objects. This object hierarchy is transformed into a relational database representing properties and inter-scale relations of independent objects by invariant attributes and inclusions, respectively. It serves as interface to a rule-based expert system that evaluates observer-specified queries automatically by selecting optimal sets of objects regarding the scale-behavior of their descriptive attributes. The object-oriented im∗correspondence: [email protected], Phone: +49 451 3909 559 (Fax: -555) H. Talbot, R. Beare (Eds): Proceedings of ISMM2002 Redistribution rights reserved CSIRO Publishing. ISBN 0 643 06804 X 265 age analysis already yielded promising results for various clinical tasks concerning macroscopic and microscopic medical images.