Effective Printed Tamil Text Segmentation and Recognition Using Bayesian Classifier

Text segmentation and recognition of Indian languages have gained a lot of research interest in the recent years. The existence of a huge number of symbols and varying characteristics in these languages makes segmentation and extraction of text a challenging task. The Tamil language has a wide variety of the literature, and printed text is available in various forms such as newspaper, books, and magazines. In this paper, extraction of printed Tamil text from an image is done irrespective of the characteristics of the text such as font style, color, and size. The proposed work uses scanned printed Tamil text as the input image. This input image is binarized since text is always available in the foreground, and histograms can be used to segment them into lines and words. The morphological operator, dilation, is used to remove outliers such as dots and commas present in an underlying object and segment the printed text into words to facilitate text detection. Further, each character is identified using bounding box technique. Classification of Tamil letters is done by extracting features such as gradient information and curvature-based information obtained from grayscale and binary images. These features are trained, and characters are classified using Bayesian classifier. The recognized characters are documented as text using Unicode format. The performance of the approach is evaluated using precision, recall, and F-measure.