IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 16, NO. 4, MAY 1998 525Image-Adaptive Watermarking Using Visual ModelsChristine I. Podilchuk, Member, IEEE, and Wenjun ZengAbstract— The huge success of the Internet allows for thetransmission, wide distribution, and access of electronic datain an effortless manner. Content providers are faced with thechallenge of how to protect their electronic data. This problem hasgenerated a flurry of recent research activity in the area of digitalwatermarking of electronic content for copyright protection.Unlike the traditional visible watermark found on paper, thechallenge here is to introduce a digital watermark that does notalter the perceived quality of the electronic content, while beingextremely robust to attack. For instance, in the case of imagedata, editing the picture or illegal tampering should not destroyor transform the watermark into another valid signature. Equallyimportant, the watermark should not alter the perceived visualquality of the image. From a signal processing perspective, thetwo basic requirements for an effective watermarking scheme,robustness and transparency, conflict with each other.We propose two watermarking techniques for digital imagesthat are based on utilizing visual models which have been de-veloped in the context of image compression. Specifically, wepropose watermarking schemes where visual models are used todetermine image dependent upper bounds on watermark inser-tion. This allows us to provide the maximum strength transparentwatermark which, in turn, is extremely robust to common imageprocessing and editing such as JPEG compression, rescaling, andcropping. We propose perceptually based watermarking schemesin two frameworks: the block-based discrete cosine transformand multiresolution wavelet framework and discuss the merits ofeach one. Our schemes are shown to provide very good resultsboth in terms of image transparency and robustness.Index Terms—Copyright protection, DCT’s, image watermark-ing, perceptual models, wavelets.I. INTRODUCTIONTHE success of the Internet introduces a new set ofchallenging problems regarding security. One of manyissues that has arisen is the problem of copyright protection ofelectronic information. Specifically, the idea of digital water-marking of electronic data has become an area of increased re-search activity over the last several years. Here we address theproblem of watermarking digital image content. Current workon watermarking falls into two broad categories: source-basedand destination-based schemes. Source-based schemes fo-cus on ownership identification/authentication where a uniquewatermark identifying the owner is introduced to all thecopies of a particular image being distributed. A source-based watermark could be used for authentication and todetermine whether a received image or other electronic datahas been tampered with. An important constraint to considerManuscript received March 1997; revised July 1997.C. Podilchuk is with Bell Laboratories, Lucent Technologies, Murray Hill,NJ 07974 USA (e-mail: [email protected]).W. Zeng is with Sharp Laboratories of America, Inc., Camas, WA 98607USA (e-mail: [email protected]).Publisher Item Identifier S 0733-8716(98)03272-7.for many source-based applications is the ability to detect thewatermark without the original image. The watermark couldalso be destination based where each distributed copy getsa unique watermark identifying the particular buyer or end-user. The destination-based watermark could be used to tracethe end-user in the case of illegal use such as reselling. It isreasonable to assume that the content provider has the originalimage available for watermark detection in destination-basedapplications. There may be applications where we wouldlike to attach multiple watermarks, source based as well asdestination based, to one image.We begin by reviewing some of the requirements thatare necessary to provide a useful and effective watermark-ing scheme. These requirements apply to any data type ingeneral but we focus on requirements that are most usefulfor destination-based rather than source-based applications.The three features that we examine for our application are:transparency, robustness, and capacity. Transparency refers tothe perceptual quality of the data being protected. For thecase of image data, the watermark should be invisible overall image types. Such a requirement is most challenging forimages composed of large smooth areas. The digital water-mark should also be robust to signal processing. Ideally, theamount of signal distortion necessary to remove the watermarkshould degrade the desired image quality to the point ofbecoming commercially valueless. Typical signal processingincludes intentional transformations of the image data as wellas illegal attempts to remove or transform the watermarkinto another valid watermark. Typical image transformationsinclude compression (in particular JPEG), resampling, requan-tization, image enhancements, cropping, and halftoning. Fordestination-based watermarking, capacity may be a criticalissue for widely distributed content. By capacity we meanthe ability to be able to detect the watermarks with a lowprobability of error as the number of watermarks increases.The watermarking technique should provide a framework toinsert the maximum number of distinguishable watermarks.There has been some interesting work on source-basedwatermarking for authentication and alteration detection of theoriginal data. For this application as well as several others,it is desirable to be able to extract the watermark withoutthe original image. The requirement of being able to detectthe watermark without the original image introduces a verychallenging problem especially if robustness is also desirable.Here we focus on applications where the original image isavailable for watermark detection. Such a scheme is practicalfor destination-based applications such as identification of end-users (customers) where the content provider would like toidentify the watermark in case of illegal use and trace thewatermark back to the appropriate end-user.0733–8716/98$10.00 1998 IEEE526 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 16, NO. 4, MAY 1998The requirements of transparency, robustness, and capacityintroduce a challenging problem from the signal processingperspective. The most straightforward way to introduce atransparent watermark results in a watermark that is