Historical Document Image Binarization Based on Edge Contrast Information

The first step of historical document image recognition is the image binarization, however historical documents of poor quality, and there are all kinds of ink, stained and the background texture complexity, which brings huge challenge to the work. This paper proposes a novel binarization method for degraded historical document image based on contrast and location information of edge. Firstly, based on the fact that all pixels in a single edge are connected, a new text edge extraction method is proposed. The method combines the high and low contrast information in the image, avoids the shortcomings of noise and breakage in the edge extraction, and has high accuracy. Secondly, the Sauvola algorithm is improved. In the process of local threshold calculation, the contrast information of the edge is added to improve the adaptability of the algorithm to the brightness change and get more accurate local threshold. Finally, the binarization of image is carried out by using the local threshold and edge location information. The comprehensive experimental results on DIBCO database show that the proposed method can eliminate noise while preserving low contrast foreground, and performs better than the classical methods.