TEXT SEGMENTATION AND BINARIZATION USING THE DIFFERENCE THEORETIC TEXTURE FEATURES

Abstract

Text Segmentation and Binarization is a very difficult and challenging task in case of unclear document images because of the difference between text and background of the text. Threshold play a vital role for text binarization because it separates the foreground and background of images and very difficult to find a good threshold. Many approaches have been available to find the threshold for text segmentation and binarization. All of them have its own advantage and disadvantage. In this project report, an approach has been proposed to text segmentation and binarization from challenging and sophisticated document images using difference theoretic texture feature. In this proposed algorithm, the adaptive threshold, contrast stretching and gamma correction have used for enhancement and binarization of document images. An adaptive global threshold algorithm merges with contrast stretching and gamma correction to enhance the document image. The difference theoretic texture feature is used for text extraction from an image. The difference theoretic feature technique has also used to segment text area from document images. This technique provides better results for all unclear and clear images as compare to other techniques. The proposed technique has been implemented using MATLAB 2012. This proposed technique has been tested on five recent document image binarization contest (DIBCO) 2009, 2011 and 2013 and handwritten-DIBCO 2009, 2010, 2011 and 2013 datasets [1] [2] [3] [4] and achieves performance 81.22%, 82.86%, 85.10%, 83.40%, 76.04% respectively. It is also tested on many different challenging document images which is captured by mobile phone. This technique provides better results for all unclear and clear images as compared to other techniques.