An Adaptively Iterative Method of Document Image Binarization

Document image binarization is difficult when the image is affected by background noise or nonuniformly illuminated. In this paper, an adaptively method is proposed to address the above problems and thus get expected binarization results. This method begins by defining a filter window length with initialized stroke width, and then transforms the document image into a feature space with Gaussian kerneled Bhattacharyya distance and sets up a threshold with Ostu’s method to temporarily binarize the image, and in the end, it uses the stroke width extracted from the newly obtained image to update the filter window length until the iteration convergent. The proposed method has been tested on the DIBCO(2009-2012) databases and get acceptable results on most images in these databases. In addition, expected results have been achieved when we tested this method on several other seriously noised and shadowed document images, even with various resolutions, which indicates our method is effective. Key–Words: adaptive binarization, document image, Bhattacharyya distance, stroke width, run-length histogram