A handwritten digit recognizer using ensemble method

Handwritten digit recognition(HDR) is a field of image processing as well as feature extraction and pattern recognition. We propose a handwritten digit recognizer based on the ensemble method and some unique feature extraction techniques in this paper. The experiment is performed on the database of MNIST which contains 60,000 digits ranging from 0 to 9 for training , and another 10,000 digits as test data. We employ AdaboostM2, SVMs and neural network algorithm as comparison and find that the proposed method outperforms the other contrastive method and the proposed four feature extraction techniques work poorly when separated but improve greatly when putted together. We also analysis the reason why some misclassification occur in high probability with the confusion matrix as well as the shortcomings of our feature extraction methods.