Digital watermarking by DCT coefficient manipulation

• Main Author: Sun, Xiaodi
• Language: English
• Published: 2009
• Online Access: http://hdl.handle.net/2429/11880

The advent of digital multimedia and worldwide area networks such as the Internet facilitates efficient distribution, reproduction and manipulation over networked information systems. To discourage the unauthorized copying and distributing of electronic documents, new tools are needed for their tracking and copyright enforcement. Digital watermark, which is an invisible mark embedded in a digital medium, for example a serial number or copyright information and can be retrieved even after attacks such as image processing and lossy compression, is one of the most promising ways for this purpose. This thesis is concerned with analyzing and developing several digital watermarking schemes based on manipulating DCT coefficients of images. In the first scheme, a variant of the watermarking scheme proposed by Langelaar, et al. for JPEG/MPEG streams is implemented based on selectively discarding high frequency DCT coefficients. The watermark bits can be recovered by comparing the high frequency energy in difference DCT blocks. The second scheme embeds a watermark into the DCT coefficients of an image during the process of quantization; by introducing a self-reference pattern, the robustness of an watermark can be improved. In the third scheme, pairs of mid-frequency DCT coefficients are selected and modified to embed a watermark. The last scheme is built on the foundation of spread-spectrum communication. After presenting a generic spread spectrum based watermarking scheme, we derive the expectation and variance of the correlation between the investigated image and the watermark, from which the performance of our spread spectrum watermarking algorithm is analyzed and is found to be consistent with the experiment results. The watermarking schemes can detect the embedded watermarks in the DCT domain without the help from the original images. Experiment results show that the embedded watermarks in images can be retrieved successfully even after the JPEG compression and certain other image processing attacks. In addition, our spread spectrum watermarking scheme is capable of embedding multiple watermarks into an image and is robust to cropping attack as well as the JPEG compression with a very low quality factor.