EDGE ORIENTED FILTER AND FEATURE INVARIANT CODING FOR MULTI-LINGUAL CHARACTER SEGMENTATION AND RECOGNITION

Abstract

In today’s computer world, Automated Character Recognition has gained a significant interest in building intelligent indexing system. This paper proposes to develop automated handwritten character recognition for Multi-lingual Scripts. Total system consists of two phases, Segmentation and Recognition. In the segmentation phase, the characters are segmented from a handwritten word image and in the recognition phase, they are processed for recognition. For the segmentation phase, a new filter called, Edge Density Filter is proposed and for recognition phase, the Gabor filter is accomplished for Scale and Rotation invariant Features Extraction. k-Nearest Neighbor Classifier (k-NN) is accomplished for the recognition of character, which has less computational complexity. Simulations are conducted over three datasets among which two are publicly available and one is Self-created. Kannada and Devanagari datasets are captured from Chars74 dataset and Telugu is created voluntarily. Performance evaluation is also carried out through Recall, Precision and Accuracy and the obtained results show an outstanding performance compared to the state-of-art techniques.