Abstract
This thesis introduces the novel concepts of Intermedian Curve Sets (IMCS) and Intercentroidal Curve Sets (ICCS). They are used here as means for the foundation of an IMCS/ICCS based shape analysis framework, comprised of IMCS/ICCS shape representation structures, a novel feature extraction and object alignment system, and probabilistic multiple classifier strategies. The proposed model is tested according to the MPEG-7 Core Experiment CE-Shape-1 Part B standards to evaluate both its retrieval and classification accuracy. Almost 85% correct retrieval and 98% classification accuracy are recorded. Given a qualitatively produced dataset consisting of digital images illustrating distinctive leaves of the tree genus Tilia, several versions of the proposed and other competitive shape analysis approaches are applied to automatically extract complete sets of morphological/shape features. The aim is to successfully perform plant taxonomic identification. Extensive comparative studies prove that the proposed IMCS/ICCS based shape analysis framework is both efficient and promising, as it outperforms other published shape analysis schemes, performing approximately 74% and 98% correct identification, for the 4-species and 14-species Tilia datasets examined respectively.