An approach to knowledge-driven segmentation

摘要

Use of knowledge in a multiresolution multipredicate representation of image data is considered for the purpose of object segmentation and classification. Techniques for producing multiresolution data in the same computational process, and for producing segmentations based on grey level, colour, texture and optical flow predicates, are illustrated. In particular the use of knowledge encapsulated in 2D models is considered for compact and distributed objects.