A Graph Theoretic Approach for Object Shape Representation in Compositional Hierarchies Using a Hybrid Generative-Descriptive Model

摘要

A graph theoretic approach is proposed for object shape representation in ahierarchical compositional architecture called Compositional Hierarchy of Parts(CHOP). In the proposed approach, vocabulary learning is performed using ahybrid generative-descriptive model. First, statistical relationships betweenparts are learned using a Minimum Conditional Entropy Clustering algorithm.Then, selection of descriptive parts is defined as a frequent subgraphdiscovery problem, and solved using a Minimum Description Length (MDL)principle. Finally, part compositions are constructed by compressing theinternal data representation with discovered substructures. Shaperepresentation and computational complexity properties of the proposed approachand algorithms are examined using six benchmark two-dimensional shape imagedatasets. Experiments show that CHOP can employ part shareability and indexingmechanisms for fast inference of part compositions using learned shapevocabularies. Additionally, CHOP provides better shape retrieval performancethan the state-of-the-art shape retrieval methods.