974 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 10, NO. 6, SEPTEMBER 2000 Embedding Image Watermarks in DC Components Jiwu Huang, Yun Q. Shi, and Yi Shi Abstract—Both watermark structure and embedding strategy affect robustness of image watermarks. Where should watermarks be embedded in discrete cosing transform (DCT) domain in order for the invisible image watermarks to be robust? Though many pa- pers in the literature agree that watermarks should be embedded in perceptually significant components, dc components are explic- itly excluded from watermark embedding. In this letter, a new em- bedding strategy for watermarking is proposed based on a quanti- tative analysis on the magnitudes of DCT components of host im- ages. We argue that more robustness can be achieved if watermarks are embedded in dc components since dc components have much larger perceptual capacity than any ac components. Based on this idea, an adaptive watermarking algorithm is presented. We incor- porate the feature of texture masking and luminance masking of the human visual system into watermarking. Experimental results demonstrate that the invisible watermarks embedded with the pro- posed watermark algorithm are very robust. Index Terms—DC components, embedding strategy, image wa- termarking, robustness, visual masking. I. INTRODUCTION Digital watermarking for images, video, and audio has re- cently drawn extensive attention. Robustness is one of the most basic requirements for invisible image watermarks. Both water- mark structure and embedding strategy affect robustness of wa- termarks. Cox et al. [1] claim that the watermarks composed of Gaussian random sequences are more robust. Our experiments support this statement [2]. The image watermarking algorithms fall into two groups. The first group of methods works in spatial domain by changing the gray levels of some pixels. Another group of techniques modi- fies the coefficients in transform domain. The methods in trans- form domain, especially in the discrete cosine transform (DCT) domain, are more popular for the following reasons. 1) The features of human visual system (HVS) can be incor- porated into watermarking in the transform domain more effectively. 2) The energy of embedded signal in the transform domain will be spread over all pixels in the spatial domain. This is advantageous to invisibility. 3) They can be implemented in compressed domain since most international image and video compression stan- Manuscript received June 25, 1998; revised January 11, 2000. This work was supported by the Natioinal Science Foundation (NSF) of China and the NSF of Guangdong, China. J. Huang is with the Department of Electronics, Zhongshan University, Guangzhou, 510275, China, and is also with the Institute of Automation, Chinese Academy of Science (e-mail:isshjw@zsu.edu.cn). Y. Q. Shi and Y. Shi are with the Department of Electrical and Computer En- gineering, New Jersey Institute of Technology, Newark, NJ 07102 USA (e-mail: shi@tesla.njit.edu). Publisher Item Identifier S 1051-8215(00)07568-6. dards, such as JPEG, MPEG, H. 261, and H. 263 are DCT-based. In this paper, we address a problem related to the embedding strategy for invisible image watermarking in the DCT domain. Most of the early image watermarking schemes modify the least significant bits (LSB) of original images or their representation in transform domain [3] to meet the requirement of invisibility. These approaches are not robust, since the LSB data are highly sensitive to various noises and common signal processing pro- cedures. Cox et al. [1] argue that the watermarks should be em- bedded in those perceptually significant components in order for the watermarks to be robustness. Some researches [4], [5] make a tradeoff. They embed the watermarks in middle-band coefficients in the DCT domain. Where should watermarks be embedded in the DCT domain? Though the viewpoint that wa- termarks should be embedded in perceptually significant com- ponents has now been well accepted, dc components are explic- itly excluded from watermark embedding [1], [2], [4], [5]. The consideration behind this is to avoid block artifacts in water- marked images. Indeed, an unreasonable amount of change in dc coefficients will result in a blocking effect in watermarked images because the average gray level of a block is proportional to the magnitude of the dc component. However, our quantita- tive analysis and experimental results on the magnitude of DCT components indicate that the watermarks may be more robust if they are embedded in dc components. Based on this observation, an adaptive watermarking algorithm is presented. Simulation re- sults support our novel embedding strategy and demonstrate that the watermarks generated with the proposed watermarking al- gorithm are invisible and very robust against noise and common image processing techniques such as compression, low-pass fil- tering, clipping, and subsampling. This paper is organized as follows. In Section II, we present a quantitative analysis on the magnitudes of DCT components for a few commonly used images with different texture fea- tures. Based on the analysis, we propose an embedding strategy that is different from the existing watermarking schemes. As an application of the novel embedding strategy, an adaptive wa- termarking algorithm utilizing the feature of texture masking of HVS is presented in Section III. The simulation results are shown and the conclusion is drawn in Section IV. II. WHERE SHOULD WATERMARKS BE EMBEDDED? In the DCT domain, watermarks should be embedded in those coefficients that meet the following requirements in order for the watermarks to be invisible and robust: 1) having large perceptual capacity [1] that allows strong watermarks to be embedded without perceptual distor- tion; 1051–8215/00$10.00 © 2000 IEEE Authorized licensed use limited to: BEIJING UNIVERSITY OF POST AND TELECOM. Downloaded on March 23, 2009 at 02:07 from IEEE Xplore. Restrictions apply.