An Efficient and Scalable Approach for Script Recognition in Online Bilingual Handwritten Text

Abstract

Script identification of handwritten text is one of the most captivating and complex applications for recognizing of different patterns of text in the field of pattern recognition. There's been a lot of progress in recognizing handwriting in monolingual environment, very few have explored in bilingual or mixed-script environment. In this paper, a Handwriting Recognition System is proposed for the code-mixed text for bilingual words of two different scripts (Devanagari for Hindi Language and Roman for English language). To recognize the script of a segmented handwritten word, different classification algorithms have been used and then comparative analysis has been performed to measure the recognition performance rate of MLP, SVM, HMM and DNM (Deep Net Model). First pass of the proposed system identified or segmented the words based on their connectivity. After that these segmented words, further segmented at the level of stroke points. From the identified types of these stroke points the idea about the script associated with the complete word has been recognized. In this proposed system, 97.025% accuracy have been achieved as per stroke identification process and 98.07% with script identification. After script identification, on the basis of these results an appropriate handwriting recognition engine can be applied to the input to convert it to its digital form.