A Fast Handwritten Digit Recognition Algorithm Based on Improved SVM

Handwritten digit recognition is of great value for application in the field of Image Processing and Pattern Recognition. For ensuring better recognition accuracy and speeding up classification process, this paper proposes a fast handwritten digit recognition method based on improved SVM. The new method uses the Separability Measure (SM) between classes in a high dimensional feature space to determine the best kernel parameters, it can fast train SVM classifiers to recognize handwritten digits. The computation of Separability Measure is a simple iterative process, thus the time required for computing SM is far less than that for training SVM classifiers in traditional parameter optimization methods. Therefore, the time for kernel parameters selection will be reduced greatly, the training process will be speeded up accordingly, and the recognition speed will be improved finally. Our experiments in the MNIST database demonstrate that the improved algorithm is feasible and effective.