Bleed-Through Removal in Document Images

Abstract

When documents are written on both sides, quite often ink bleeds through the paper. This is a common phenomenon with old documents and low quality paper. With the presence of increased bleed-through, reading and deciphering the text becomes tedious. This thesis implements algorithms for reducing bleed-through distortion using techniques in digital image processing. A comparative study of three methods has been performed with the first being basic enhancement through thresholding. The next two methods are founded on a registration process which aims at working on both the recto and verso sides simultaneously. Firstly, both sides of the documents are digitized. The verso is then flipped and corrected so as to correspond to the coordinates of one side exactly with the coordinates of the original writing on the other which is done using an affine transformation of six parameters. The parameters are found by optimizing the alignment process. A restoration algorithm is then applied to remove bleed though areas on the desired side page. The third method proposes the use of cross correlation to handle the registration process. The accuracy of bleed-through correction is found to be largely dependent on the accuracy of alignment of the documents.