Local features-based script recognition from printed bilingual document images

ISSN： 0952-8091

Source： International Journal of Computer Applications in Technology, Vol.38, Iss.4, 2010-08, pp. : 283-297

Disclaimer: Any content in publications that violate the sovereignty, the constitution or regulations of the PRC is not accepted or approved by CNPIEC.

Previous Menu Next

Abstract

Classification and identification of language in a biscript document is one of the important steps in the design of an OCR system for successful analysis and recognition. This paper presents architecture for script recognition of bilingual document images (Tamil, English), which specifically takes the challenges of recognition at character level by predicting the script of word image using its initial character, thereby adapting to various font faces and sizes. This recogniser models every character as Tetra bit values (TBV), which corresponds to the spatial spread, derived from the segmented grids of the character. We employed a decision tree classifier (DTC) for the classification of script on over the patterns generated from TBV. A spatial features-based script recogniser (SFBSR) is trained and tested with bilingual document images, consisting of various Tamil and English words, to show its effectiveness towards script identification. Classification accuracy in training and testing sets is promising. An evaluation of the system performance with various techniques shows a significant performance improvement in SFBSR. This can be embedded with OCR prior to its recognition stage.