Document text Recognition

Abstract

Text recognition is a process by which computer recognizes the written or printed characters newlinesuch as words, letters, numbers and to change them into a form that the computer can newlineunderstand and make use of it further. There is a great influence of computers on the lives of newlinehuman beings since the computer usage has an immense importance in the day to day activities. newlineText recognition can be treated as the fastest and very natural way of communicating computer newlinesystem by a human being. The bank cheque processing, automatic data entry, retrieval of text newlinefrom old documents are some of the important areas of applications. Researchers have already newlineproposed many recognition systems and approaches for printed and handwritten character newlinerecognition in English, Chinese etc. languages. In the Indian context, Devanagari script has newlinebeen given attention by the researchers for three decades. But there are many languages which newlineare yet to be attended in Devanagari script. In this thesis attention has been given to the very newlineancient Indian languages such as Pali - a language which is also referred as Aadi-Prakrut or newlineMagadhi and another language which has been focused in this thesis is the Marathi language. newlineA novel approach is devised for identifying isolated handwritten Pali characters. In this newlinework, the structural and the features of Gabor filter are considered. Using structural features, newlinea feature vector comprising of the density of total pixels, height, width and area is calculated. newlineAdditionally, 16 features of Gabor filter are considered and the decision tree is employed for newlinethe purpose of classification. It is observed that the proposed system achieves 96.25% newlinerecognition accuracy, tested over 26240 handwritten Pali characters. Another approach to newlinerecognize handwritten Devanagari characters is also performed using a combination of newlinequadratic and SVM classifiers. Here, features like directional features that are strength, angle newlineand histogram of gradient are employed. Gaussian filters are used to downsample the features newlineto obtain a