Gosciewska,K., Frejlichowski,D.
Abstract:
This paper presents the problem of coarse classification in an application to teeth shapes. Coarse classification allows to separate a set of objects into several general classes and can precede more detailed identification or narrow the search space. Object"s features are mainly determined by its geometrical aspects therefore we investigate the use of shape description algorithms, namely the Two-Dimensional Fourier Descriptor, UNL-Fourier Descriptor, Generic Fourier Descriptor, Curvature Scale Space, Zernike Moments and Point Distance Histogram. During the experiments we examine the accuracy of classification into two classes: single-rooted teeth and multi-rooted teeth - each class has five representatives. We also employ an additional step for data reduction. Reduced representations are obtained in three ways: by taking a part of the original representation, by predefining a shape description algorithm parameter or by applying an additional step of data reduction technique, i.e. the Principal Component Analysis or Linear Discriminant Analysis. Euclidean distance is used to match final feature vectors with class representatives to indicate the most similar one. The experimental results proved the effectiveness of this approach.