Combined mRMR-MLPSVM Scheme for High Accuracy and Low Cost Handwritten Digits Recognition

this paper presents a novel algorithm for handwritten digit recognition which in addition to its high accuracy, enjoys low implementation complexity. The proposed algorithm sorts all features using mRMR (Minimal-Redundancy and Maximal-Relevance) method and selects the best top features by evaluating the training data. The selected features are then used as input of our classifier. The classifier that we used is an MLPSVM classifier which combines the good properties of MLP (Multi-Layer Perceptron) and SVM (Support Vector Machines). The performance of the proposed scheme is then evaluated against ORHD and MNIST datasets, which shows that despite lower complexity compared to existing methods, it can get to high accuracy of 96.1 and 98.14 on the datasets respectively

Keywords:

handwritten digit recognition , mRMR , SVM , MLP , ORHD , MNIST ,