Enhanced speeded up robust feature with bag of grapheme (esurf-bog) for tamil palm leaf character recognition

Abstract

Palm leaf was one of the vital writing tools in ancient times. Coarse, and contrasting colored surface and different colored letters are the hallmarks of this manuscript. In this paper, feature recognition using Enhanced Speeded Up Robust Feature with Bag of Grapheme (ESURF-BoG) is used to recognize the characters in Tamil palm leaf manuscripts. This method aims to detect the strongest critical points from the input character with different orientations. These key point features were created for training image as a model called Bag of Grapheme (BoG) with code word creation. Hence, unsupervised key point features were extracted, and pattern matching is performed for the testing image. The proposed architecture is used to recognize the characters from a vast data set of optical palm leaf manuscripts. This method is compared with the classic feature detectors SURF, min-Eigen feature detector and HOG feature detectors in terms of accuracy of recognition. The proposed method outperformed the classic models in terms of the recognition rate.