An Experimental Study on Various Combinations of Shape Descriptors and Matching Methods Applied in the General Shape Analysis Problem

The aim of the General Shape Analysis (GSA) is to find one or a few most similar general templates for a processed object — exact identification is not performed, but the general shape features are obtained. The GSA approach may be applied for coarse separation of objects in the database prior to their classification or in the case when the data are incomplete or there is a high variability within them. In this paper, the GSA problem is investigated using various combinations of shape descriptors and methods for estimating the similarity or dissimilarity between shape representations. The experiments involved the use of five shape descriptors, namely the Two-Dimensional Fourier Descriptor, Generic Fourier Descriptor, UNL-Fourier descriptor, Zernike Moments and Point Distance Histogram, as well as four matching methods, that is the Euclidean distance, Mahalanobis distance, correlation coefficient and C1 metric. The effectiveness of the experiments was calculated as a coincidence between experimental results and results provided by people through inquiry forms. The experiments made it possible to determine the influence of various matching methods on the final effectiveness when a particular shape descriptor was applied. The best result was obtained for the combination of UNL-Fourier descriptor and C1 metric.