Classification of deformable geometric shapes using radial-basis function networks and ring-wedge energy features

This paper describes a system for automatic classification of geometric shapes based on radial-basis function (RBF) neural networks even in the existence of shape deformation. The RBF network model is built using ring-wedge energy features extracted from the Fourier transform of the spatial images of geometric shapes. Using a benchmark dataset, we empirically evaluated and compared the performance of the proposed approach with two other standard classifiers: multi-layer perceptron neural networks and decision trees. The adopted dataset has four geometric shapes (ellipse, triangle, quadrilateral, and pentagon) which may have deformations including rotation, scaling and translation. The empirical results showed that the proposed approach significantly outperforms the other two classification methods with classification error rate around 3.75% on the testing dataset using 5-fold stratified cross validation.