Document Image Binarization Using Adaptive Image Contrast for Degraded Documents

Abstract

Binarization of document images which are highly degraded is a challenging task due to the issues like irregular contrast, irregular illumination etc. among the foreground text pixels and document background pixels. In this thesis, we propose an image binarization technique that uses adaptive image contrast to address these challenges. Adaptive image contrast is the weighted sum of the image gradient and the local contrast which makes it more tolerant to text and background variation that occur due to various kinds of degradations found in document images. In the proposed method, first the source image is de-noised using DWT as a pre-processing step. Then the de-noised image is used to construct an adaptive contrast map. The contrast map thus constructed is then binarized to form a binary contrast image. To determine the text stroke edge pixels, an edge map generated by Canny’s edge detector algorithm is combined with the binary contrast image. The grayscale values of the text stroke edge pixels are used to segment the text from the image background by local thresholding in sub-images independent of each other .The proposed method is easy to use, efficient, and requires minimal parameter tuning. It has been compared with other promising technique for image binarization on datasets from DIBCO 2009, 2011 and HDIBCO 2010 and achieves notably better results than other competing methods.