A Numeral Script Identification from a Multi-lingual Printed Document Image

Abstract

India is a multi-lingual multi-script country, where a printed document which contains information in the form of texts, images, etc.; the texts part may have composed with characters and numerals of one or more scripts. So, it is necessary Identify the scripts of numerals/characters from multilingual document before feeding them to their individual script OCR systems. In this paper, the system made an attempt to recognize the script of numerals belongs to Kannada, Devanagari, and English based on structural features like water reservoir, aspect ratio, horizontal and vertical strokes. Initially, Bi-script and tri-script numerals script identification experiments are conducted on a dataset of 2100 numerals string(word), by taking 700 samples for each script and noticed average accuracy for tri-script numerals is 93.62%.