Handwritten Digit Recognition Based on LS-SVM

In this paper we present a method of handwritten digit recognition based on least squares support vector machines (LS-SVM). Principal component analysis (PCA) is used to extract 10-dimension features from the original digit images for handwritten digit recognition. The performance of LS-SVM on handwritten digit recognition tasks is compared with three typical classification methods, including linear discriminant classifiers (LDC), the nearest neighbor (NN), and the back-propagation neural network (BPNN). The experimental results on the popular MNIST database indicate that LS-SVM obtains the best accuracy of 87.5% with 10-dimension features, outperforming the other used methods.