ANALYSIS OF VARIOUS BINARIZATION TECHNIQUES

Abstract

Enhancement of deteriorated document images is one of the striking and demanding researches in present time. Deteriorations in an image are due to uneven illumination scattered over document including smudging of text, bleeding through of ink to the other side of page, aging of document causing degradation of paper ink. Binarization of an image is a method of separating pixel intensities into two sets, white fall in background and black are termed as foreground. Thresholding is a significant phenomenon for image binarization. On the large scale thresholding is classified into two categories: global thresholding and local thresholding. Images in which contrast is uniformly distributed between foreground and background e.g. document images, global thresholding is preferred. Document images in which discrepancy in illumination and contrast exists, there lies various pixels that are difficult to classify as foreground or background. Binarization incorporating local thresholding is better suitable for these kinds of cases. In the past many binarization techniques are proposed, but a single technique suitable for all types of degradation. For choosing the best method of binarization for various degraded images there is no automatic and robust system available. Each binarization technique used to enhance the degraded image has some pros and cons. This thesis holds a comprising survey of binarization techniques and explains a new binarization method using Gradient Correlation Similarity decolourization method with standard log based Nick’s thresholding algorithm. The results of various methods are compared with respect to parameters e.g. Precision Rate, recall rate, DRD, MRM, fMeasure etc. This thesis also highlights the various issues encountered related to document image binarization.