Malayalam OCR: N-gram approach using SVM classifier

Optical Character Recognition which could be defined as the process of isolating textual scripts from a scanned document, is not in its 100% efficiency when it comes to a complex Dravidian language, Malayalam. Here, we present a different approach of combining n-gram segmentation along with geometric feature extraction methodology to train a Support Vector Machine in order to obtain a recognizing accuracy better than the existing methods. N-gram isolation has not been implemented so far for the curvy language Malayalam and thus such an approach gives a competence of 98% which uses Otsu Algorithm as its base. Highly efficient segmentation process gives better accuracy in feature extraction which is being fed as the input of SVM. The proposed tactic gives an adept output of 95.6% efficacy in recognizing Malayalam printed scripts and word snippets.