Separating text and background in degraded document images - a comparison of global thresholding techniques for multi-stage thresholding

Before any processing of the textual content of a document image can be performed the text must be separated from the background of the image. Several thresholding algorithms have previously been proposed and are widely used in document processing. None have been shown effective at thresholding difficult documents where the background and foreground are non-uniform. In this paper we investigate the use of three global thresholding algorithms (Otsu's, Kapur's entropy and Solihin's quadratic integral ratio (QIR)) as the first stage in a multi-stage thresholding algorithm for use in degraded document images. It is concluded that Otsu's and Kapur's algorithms do not work well for difficult documents as they tend to over-threshold the image, thus losing much of the useful information. The QIR algorithm is more accurate in separating the foreground and background in these images, leaving a range of undecided, fuzzy, pixels for later processing in a subsequent stage.