Mikhail PomeranetsEE3414
Kathy ZaslavskayaFinal Project ReportProf. Yao Wang
Survey of Digital Watermarking Technology
Watermarking is best known for official company letterheads or a hundred dollar bill. However, in today’s world of advanced technology, protecting copyrighted works gained a new urgency. Since the exponential rise of the Internet in the early 1990’s, illegal distribution of copyrighted material has followed the exponential rise. It is estimated that by 2005, book publishers and record label companies will lose 4.6 billion dollars a year (Forrester Research).
This paper will focus on three main issues. First, the paper will disclose the reason why digital watermarking has become so important. Second, the paper will focus on different technical approaches to watermarking technology and describe different techniques. Finally, commercial applications and new research will be discussed in order to gain understanding of what the future holds.
In order to proceed, some basic terms need to be defined. A watermark in its basic form is extra data that is inserted into a medium such as an image or audio file. A robust watermark is one that could withstand malicious or accidental attacks, such as cropping, scaling, file conversion/compression, printing, and rotation. A fragile watermark, on the other hand, is one that is destroyed once any changes to the document occur. Under ideal circumstances, a robust watermark would also pinpoint exact pixels that have been tampered with.
When working with images, watermarking can be implemented using the spatial or frequency domain. Several techniques for the procedure exist, to use in both domains, such as color separation in the spatial domain, and Fast Fourier Transform in the frequency domain. Watermarking using these methods, as well as other schemes, will be discussed in detail.
A simple watermarking algorithm uses metatags to embed secret code in the header of the file. However, this technique is very fragile, as any type of file conversion is likely to remove the watermark. Another simple technique takes advantage of the Contrast Sensitivity Function (CSF). CSF is an index characterizing how well the human could see at different frequencies. Particularly it shows that at high frequencies the eye is very insensitive to change. Therefore, encompassing a watermark in high frequency will make it transparent to the human eye. Unfortunately, a simple low pass filter will remove the watermark.
Watermarking research has yielded algorithms, not only for use with ordinary images, but with text as well. Though fragile, three proposed methods are text line coding, word space coding, and character encoding. These methods depend on the spacing between the lines of a text, spacing between words, and spacing between characters. While some techniques are more fragile (watermark is easily removed) than others, the purposes, potential use and security issues regarding each technique made its way into our research, and will be discussed in full detail.
A superior watermarking method has been developed by Kodak. Either a logo or a bitstream could be embedded in any image or movie. The bitstream is embedded using positive or negative Dirac Delta function. (x,y)= C(x,y)C(x,y), where is defined as cyclic correlation. In other words, the formula above is equivalent to multiplying the magnitudes of two signals, while subtracting their phases in the Fourier domain.
The process is designed to split the image into M x N blocks. A secret key is used to generate a ‘random’ phase that resembles noise. The ‘noise’ and watermark are convolved and scaled to a small amplitude. This result is then added to each block of the original image. Because it resembles noise, the key cannot be easily intercepted in the image. Mathematically, the process is
I’(x,y)= (M(x,y)*C(x,y))+I(x,y), where
* is cyclic convolution, I(x,y) is original image, I’(x,y) is the image embedded with a watermark, M(x,y) is the message image, C(x,y) is the carrier image, and is arbitrary constant to make image invisible and robust
The reverse is performed when extracting the watermark. The image is broken up into M x N pieces and all the pieces are added together to cancel the effects of possible noise. As a separate process, the same ‘random’ noise is generated using the same key. The noise is then correlated with the sum of all blocks. If the correct key has been used, the original watermark will appear. The image below, courtesy of Kodak, demonstrates the process.
This same technology could be applied to the digital movie industry. An invisible watermark could be embedded into the frames of a movie, containing theatre and distributor information. If a pirated copy should fall into the hands of law enforcement officials, the watermark could be extracted and used to prosecute the responsible parties. Kodak has demonstrated an accuracy of over 99.9%, given 23 or more frames which are watermarked (half a second).
Another method has been proposed by Wong and Memon, both of whom are professors at PolytechnicUniversity. In fact, Wong and Memon have two variants to their method. The first includes a secret key, and the second a public key. In the secret key scheme, only two people have the key, the owner and the verifier. The key must be passed along with the image, but on a different, yet secure channel. While this greatly increases security, it is not always desirable. For instance, if anorganization wants to broadcast copyrighted material, having a secret decryption key would negate the purpose.
The public key resolves the problem by allowing anyone to verify the authenticity of the document, while only one person is able to embed the watermark into the content. Regardless of whichever method is used, the image must be grayscale. Otherwise, the watermarking process has to be applied to each color plane separately (i.e., R, G, B).
The image is divided into N x M blocks where Xr denotes the rth block. First, all LSBs are set to 0. The user key along with the LSBs and other parameters are used to create a hash function output. In a good hash function, given x and y, where x is not equal to y, h(x) will never equal h(y). Once the hash output has been calculated it is XORed with the watermark to receive signals Cr. Cr is the inserted into the LSBs to make the final image. Since only LSBs have been modified, no significant degradation is visible. The figures below, courtesy of Wong and Memon, show embedding and extracting algorithms.
The remaining work for this paper consists of obtaining various commercial uses of watermarking technology, and how such technology would be implemented ina commercial environment. Suffice it to say, questions regarding the usefulness of such technology will be addressed.
New research has yet to be considered to see where digital watermarking is going. However, it is evident that the field contains much promise, and is undergoing active research. Possible uses of such technology include proof of occurrence, when a picture has been modified using software such as Adobe Photoshop, the watermark will reveal which part of the photograph has been tampered with. Another application currently under development is duplication control. In such an application, additional hardware is added to devices such as DVD players, where a watermark is read and modified each time the media is copied. After a pre-programmed number of copies, the hardware will not allow the creation of duplicates.
Just as importantly, a watermark could be inserted into a photo ID or a passport, where the watermark would contain an index to the national database. Each time the passport is scanned, the national database will be contacted to validate the passport. Such system would make fake photo IDs almost impossible.
Sources
- Digimarc is a commercial provider of watermarking technologies. The site offers many solutions for modern business.
- “Digital Watermarking” by R. Chandramouli, Dept. of ECE, Stevens Institute of Tecnology, Nasir Memon, Dept. of CS, Polytechnic University, and Majid Rabbani, Imaging Research & Advanced Development, Eastman Kodak Co.
- “Data Embedding Using Phase Dispersion” by Chris Honsinger and Majid Rabbani, Copyright 2000, Eastman Kodak Co.
- “Secret and Public Key Imgae Watermarking Schemes for Image Authentication and Ownership Verificaiton” by Ping Wah Wong, and Nasir Memon, IEEE June 29, 1999.
- Cannon has developed an error correcting approach, and is planning to improve it further.
- ACM article describing different approaches and watermarking techniques.
- Paper published by two German researchers regarding a new watermarking technique.
------
What is Digital Watermarking?
Digimarc's digital watermarking technologies allow users to embed in audio, images, video and printed documents a digital code that is imperceptible during normal use but readable by computers and software. The science of creating these imperceptible codes is known as digital watermarking. Digimarc is a leading owner of intellectual property relating to digital watermarks and a pioneer in the commercial application of this technology.
Digimarc ImageBridge watermarking / copyright communication and tracking of digital imagesDigimarc Excalibur Copy Detection / counterfeit detection of ID cards and packaging
Digimarc Excalibur Secure Authentication / authentication and linking of packaging and documents
Digimarc MediaBridge technology / linking of packaging, tags and labels
------
------
Performance
In general, the performance of a data embedding algorithm is a tradeoff between three factors:
- Data capacity (number of embedded bits)
- Visibility of the embedded message
- Robustness (ability to withstand image-processing tasks)
Commonly, performance of a digital watermarking technology is evaluated by keeping (1) and (2) constant, and then measuring the message survival rate when the picture is subjected to varying degrees of image manipulation and degradation.
------
------
Data for still pictures is frequency-converted using Fourier conversion, where the converted data is manipulated according to the key information- embedding location and strength-and pulse signals are inserted for the inverse Fourier conversion that restores the picture data. The inserted pulse signals for different pieces of information are dispersed throughout the picture during the inverse conversion process and retained in an invisible state in the picture data, forming a digital watermark.
Canon's New Technologies Using Error-Correction EncodingDigital watermarking is one form of security technology to ensure copyright protection. This technology provides a high level of safety, because the location of embedded information is secret, and neither the watermark algorithm nor the location of embedding is made public.
Although effective when used within one organization, the secrecy required for this method is a hindrance when several users are involved, or when standardization is desired.
Working under the premise that making both algorithms for protecting image copyrights and embedding locations public would enable many people to use the technology for copyright protection, Canon developed its own error-correction encoding technology. This method uses mathematical processing to restore signals that are difficult to distinguish from noise to their original state. With this technology, the data in picture and voice contents that contain embedded watermark information is processed as a whole to correct errors. It is possible to restore embedded information in the contents by dispersing and embedding the watermark information and subjecting the entire contents to the error-correction encoding process. This works even if the information has been erased or illegally rewritten.
This new technology, developed by attacking the problem from a different angle, protects copyright holders and purchases by preventing embedded data from being illegally rewritten. The software that reads the watermark is commercially available to facilitate the detection of illegal activity.
In addition to the aforementioned method of protecting image copyrights, Canon is developing digital watermark using wavelet conversion adopted for JPEG 2000, as well as other digital watermark including those for high-quality printing.
------
Techniques, applications, links, etc
------
Despite claims that digital watermarks can survive image alteration and cannot be stripped without seriously affecting image quality, a recent CyberTimes report revealed that the digital watermarks on some images "may have been weakened or [may have] disappeared by the time the images were processed for the Internet." Resizing, compressing and converting images from one file type to another may add noise to an image or diminish its watermark in such a manner that the watermark becomes unreadable. Further, even when a digital watermark remains intact, tracking services are of limited use to copyright owners in searching for illegal copies of their works on the Internet when such copies are within sites protected by passwords.
------
An international meeting point for scientists, researchers and companies active in digital watermarking.
- Why not add the copyright information into the file format?
One could define a new audio file format, in which the watermark is a part of the header block but is not removable without destroying the original signal, because part of the definition of the file format requires the watermark to be therein. In this case the signal would not really be literally 'destroyed' but any application using this file format would not touch it without a valid watermark. Some electronic copyright management system propose mechanisms like this. Such schemes are weak as anyone with a computer or a digital editing workstation would be able to convert the information to another format and remove the watermark at the same time. Finally this new audio format would be incompatible with the existing one. Thus the watermark should really be embedded in the audio signal.
This is very similar to S.C.M.S. When Philips and Sony introduced the 'S/PDIF' (Sony/Phillips Digital Interchange Format), they included the S.C.M.S. (Serial Code Management System) which provides a way copies of digital music are regulated in the consumer market. This information is added to the stream of data that contains the music when one makes a digital copy (a 'clone'). This is in fact just a bit saying: digital copy prohibited or permitted. Some professional equipment are exempt for needing S.C.M.S.
With watermarking however, the copy control information is part of the audio-visual signal and aim at surviving file format conversion and other transformations.
------
estimates of losses
Page 1 of 8