Superquadric-based geons recognition utilizing support vector machines

Recognizing geons is one key research issue in developing 3D object recognition system based on Recognition by Components (RBC) theory. We present a novel approach to recognize the set of superquadric-based geons by support vector machines (SVMs) classifier. A new feature set derived from superquadric parameters is proposed for SVM-based classification. And the knowledge-based feedback of SVM network has been introduced for improving the classification performance. Comparison experiments have been performed for feature sets selection, kernel functions and related issues of SVM classifier for geons classification. Experimental results are presented and discussed, which show the efficiency and advantages of our approach.