An OCR for separation and identification of mixed English — Gujarati digits using kNN classifier

摘要

This paper addresses the script identification problem of bilingual printed document images. We propose an OCR system that separates and identify mixed English-Gujarati digits. Here, first the system is trained with standard data samples. Then for testing, data samples are collected from different sources of paper like, news paper, book, magazine, etc. Random sized pre-processed image is normalized to uniform sized image. A statistical approach is used for feature extraction. For classification kNN classifier is used. The model gives average accuracy of 99.26% for Gujarati digits, 99.20% for English digits, and overall accuracy 99.23%.