Intelligent Multi-Level Regions-of-Interest (ROI) Document Image Encryption Using an Online Learning Model

Image-based document management systems have become increasingly popular for handling documents that contain both text and graphical elements. When such systems are used to store confidential information, document security is a key concern. Conventional encryption techniques used for images fail to provide the level of flexibility required by such document management systems. Newer image encryption techniques provide improved flexibility at the cost of backwards compatibility and ease of use. This paper presents a novel approach to document image encryption using an online learning model. The proposed system provides backwards-compatible document image encryption in regions of interest (ROI) with support for multiple levels of authority. Furthermore, the proposed system is capable of learning from user feedback to improve the ROI selection used during the semi-automatic document encryption process. Experimental results from the encryption of test documents demonstrate the effectiveness of the proposed system.