Histogram of oriented gradients for English-Bengali script recognition

Abstract

Script recognition plays a key role in successful implementation of Optical Character Recognition (OCR) in multi-script document analysis environment. In this paper, we propose a new method for English-Bengali script recognition based on histogram of gradient orientation. We observe that both the scripts of consideration have distinct nature of edge orientations. English and Bengali script can be distinguished based on frequency of different edge directions. A modified version of the HOG feature is suggested towards script recognition. Experimental results show very high recognition accuracy by the proposed method in case of printed and handwritten document image databases.