Date of Award:

2012

Document Type:

Thesis

Degree Name:

Master of Science (MS)

Department:

Computer Science

Advisor/Chair:

Xiaojun Qi

Abstract

With the rapid development of digital multimedia and internet techniques in the last few years, more and more digital images are being distributed to an ever-growing number of people for sharing, studying, or other purposes. Sharing images digitally is fast and cost-efficient thus highly desirable. However, most of those digital products are exposed without any protection. Thus, without authorization, such information can be easily transferred, copied, and tampered with by using digital multimedia editing software. Watermarking is a popular resolution to the strong need of copyright protection of digital multimedia. In the image forensics scenario, a digital watermark can be used as a tool to discriminate whether original content is tampered with or not. It is embedded on digital images as an invisible message and is used to demonstrate the proof by the owner. In this thesis, we propose a novel localized geometric-distortion resilient digital watermarking scheme to embed two invisible messages to images. Our proposed scheme utilizes two complementary watermarking techniques, namely, local circular region (LCR)-based techniques and block discrete cosine transform (DCT)-based techniques, to hide two pseudo-random binary sequences in two kinds of regions and extract these two sequences from their individual embedding regions. To this end, we use the histogram and mean statistically independent of the pixel position to embed one watermark in the LCRs, whose centers are the scale invariant feature transform (SIFT) feature points themselves that are robust against various affine transformations and common image processing attacks. This watermarking technique combines the advantages of SIFT feature point extraction, local histogram computing, and blind watermark embedding and extraction in the spatial domain to resist geometric distortions. We also use Watson’s DCT-based visual model to embed the other watermark in several rich textured 80×80 regions not covered by any embedding LCR. This watermarking technique combines the advantages of Harris feature point extraction, triangle tessellation and matching, the human visual system (HVS), the spread spectrum-based blind watermark embedding and extraction. The proposed technique then uses these combined features in a DCT domain to resist common image processing attacks and to reduce the watermark synchronization problem at the same time. These two techniques complement each other and therefore can resist geometric and common image processing attacks robustly. Our proposed watermarking approach is a robust watermarking technique that is capable of resisting geometric attacks, i.e., affine transformation (rotation, scaling, and translation) attacks and other common image processing (e.g., JPEG compression and filtering operations) attacks. It demonstrates more robustness and better performance as compared with some peer systems in the literature.

Comments

This work made publicly available electronically on May 10, 2012.

Share

COinS