Joint restoration and compression of document images with bleed-through distortion

Abstract

This thesis presents research on the restoration of documents suffering bleed-through and the joint compression of the original document and its bleed-through corrected version. This topic as presented in this thesis requires four steps: (i) registration, (ii) segmentation, (iii) inpainting and (iv) compression. First, a review of existing methods for registration and segmentation are presented, followed by a review of some existing image compression formats for the purpose of compressing the images. Then, the development of efficient registration and segmentation algorithms that are based on previous research are presented, along with a new method for joint document compression. The methods presented in this thesis are compared with previous methods and are shown to have improvements over them. In terms of registration, the method presented here is faster in reaching a more accurate final solution than previous registration methods presented for this problem. The segmentation algorithm presented here is shown to accurately segment portions of bleed-through away from the rest of the document. On the topic of joint compression, it is shown in this thesis that jointly compressing a document with its bleed-through corrected version will reduce the bandwidth required to transmit those document images, when compared to simply compressing the two images individually.