A Hybrid Method for Multiclass Classification and Its Application to Handwritten Character Recognition

The support vector machine (SVM) is an effective pattern classification method. However, solving N(N-1)/2 binary classifications in the training phase makes it too costly to use SVM in applications with a high number N of class types. In this paper, we propose a new prototype classification method that can be combined with SVM for pattern recognition. This hybrid method has the following merits. First, the learning algorithm for constructing prototypes determines both the number and the location of the prototypes. This algorithm not only terminates within a finite number of iterations, but also assures that each training sample matches in class type with the nearest prototype. Second, SVM can be used to process the top-rank candidates obtained by the prototype classification method, which saves time in both the training and testing processes. We apply our method to recognizing handwritten numerals and handwritten Chinese/Hiragana characters. Experiment results show that the hybrid method saves a great deal of training and testing time in large scale tasks and achieves comparable accuracy to that achieved by using SVM solely. Our results also show that the hybrid method performs better than the nearest neighbour method.