Performance Evaluation of a Document Image Watermarking Approach With Enhanced Tamper Localization and Recovery

Abstract

Digital transmission of sensitive images and documents over unsecure networks, such as the Internet, has become a general practice. As a result, the digital content has become vulnerable to intentional and unintentional modifications during transmission. Prior to considering the reliability of such digital content, it is important that the authentication and the integrity of the content can be confirmed. Such issues were largely considered in the literature for natural and texture-based images, with only minimal work found to address the challenge of sensitive document images with known constraints. In this paper, we present an evaluation of a non-blind robust-watermarking approach with linear interpolation for tamper-detection, localization and recovery. Performance of the proposed approach and its resistance to random paint-based and Stirmark-based attacks for sensitive documents are investigated. Throughout this paper, a sensitive Arabic scripture was used as a case study of a sensitive document image, in which such operations were performed. The proposed model presents a superior tamper detection and recovery capability in comparison to other models in the related literature. Simulation results had demonstrated that the proposed method was robust to malicious attacks and was capable of localizing and correcting tampered regions with a high degree of accuracy. Significantly, it was noted that an average of 43 dB was obtained for the peak signal-to-noise ratio results, while generally low-BER results were achieved, with a rate of 0% in some cases. Finally, the proposed approach possessed a further advantage in its broad applicability to other sensitive digital image content.