An Approach towards Malayalam Handwriting Recognition Using Dissimilar Classifiers

The handwriting recognition in Malayalam is a challenging as well as emerging area of pattern recognition. It is a tedious process mainly due to its enormous character set. Here we propose a novel method for handwriting recognition by using two dissimilar classifiers. It can also be called as an ensemble method in which multiple classifiers are combined to solve a particular problem and thereby improve the performance of the system. The experiment is conducted in 2 phases. In the first phase, 33 isolated characters in Malayalam were used. In the second phase, Malayalam sentences were used. From the preprocessed image, we were extracted two features: SURF feature and Curvature feature. These features were fed as input to a neural network and an SVM classifier. Finally, the result of both the classifiers was combined to get the final results. The system showed an accuracy of 89.2% in the first phase. An accuracy of 81.1% was exhibited in the second phase.