Handwritten Devanagari Character Recognition using SVM and ANN

Abstract: Character recognition is a challenging research area due to its diverse applicable environment. It is able to solve complex problem and make easy task for human. This paper proposed a system for recognizing offline Handwritten Devanagari Characters using Artificial Neural Network and Support Vector Machines as classifiers and the results are compared. We discuss various characteristics of the classification methods that are applied successfully to handwritten Devanagari characters. It involves binarization, noise removable, and normalization. Statistical and structured based features are used for extracting the feature of characters. The feature extraction techniques: Chain Code, Zone Based Centroid, Background Directional Distribution and Distance Profile features are applied to the pre-processed images. Experiment is carried out by varying the image sizes: 30x30, 40x40, and 50x50 using MATLAB on more than 20,000 samples. The overall recognition accuracy is 97.61 % by using SVM.

Keywords: Optical Character Recognition (OCR), Artificial Neural Network (ANN), Pre-processing, Binarization, Feature Extraction, Support Vector Machine (SVM).