Script Identification for printed document images at text-line level using DCT and PCA

Abstract

The progress of information technology and the wide reach of the Internet are drastically changing all fields of activity in modern days. As a result, a very large number of people would be required to interact more frequently with computer systems. To develop the human–machine interaction more effective in such situations, it is desirable to have systems capable of handling inputs in a variety of forms such as printed/handwritten paper documents. In a multi-lingual country like India, where more than 22 official languages and 12 different scripts are used for these languages. it is an utmost essential & complicated for designing an OCR system and it became more difficult if the document consist of multiple languages so for an automated multilingual environment such document processing systems relying on OCR would clearly need to identify the script type of the document files, so that specific tool of OCR can be selected. In this paper, a script identification approach for Indian scripts is proposed at text-line level. It is a Visual appearance-based script recognition method. The recognition is based upon features extracted using Discrete Cosine Transform (DCT) and Principal Component Analysis (PCA) algorithm and for further extraction we use Modified-KNN. The proposed method is tested on printed document images in 11 major Indian languages, 95% recognition accuracy is obtained.