An Approach to Improve Efficiency and Accuracy of Sophisticated and Inefficient Shape Matching Techniques

In this paper, we propose an approach to significantly speed up complex but accurate shape matching approaches. Existing shape matching approaches concentrate on the accuracy perspective of shape matching without giving much consideration on the efficiency. Consequently, such approaches although accurate but can not meet the online shape retrieval and classification demands. They are also useless for datasets containing extremely large number of shape samples. We handle this problem by presenting an extremely efficient shape matching approach based on compressed fourier coefficients. Fourier descriptors are further indexed by a hierarchical tree-based indexing structure to achieve fast pruning of distant shapes. We then employ a given sophisticated whilst inefficient shape matching approach on the pruned dataset which makes them applicable to large database settings. We further combine our proposed Fourier descriptor based shape matching with the sophisticated shape matching approach to further enhance its accuracy. Experimental evaluation demonstrates the effectiveness of our proposed approach using different shape datasets.