Shape-based image retrieval using support vector machines, Fourier descriptors and self-organizing maps

Image retrieval based on image content has become an important topic in the fields of image processing and computer vision. In this paper, we present a new method of shape-based image retrieval using support vector machines (SVM), Fourier descriptors and self-organizing maps. A list of predicted classes for an input shape is obtained using the SVM, ranked according to their estimated likelihood. The best match of the image to the top-ranked class is then chosen by the minimum mean square error. The nearest neighbors can be retrieved from the self-organizing map of the class. We employ three databases of 99, 216, and 1045 shapes for our experiment, and obtain prediction accuracy of 90%, 96.7%, and 84.2%, respectively. Our method outperforms some existing shape-based methods in terms of speed and accuracy.