Towards Robust and Hidden Image Copyright Labeling
E. Koch & J. Zhao
Fraunhofer Institute for Computer Graphics
Wilhelminenstr. 7, 64283 Darmstadt, Germany
email: {ekoch,zhao}@igd.fhg.de
Abstract –– This paper first presents a ”hidden label” approach for identifying the ownership and distribution of multimedia information (image or video data) in digital networked environment. Then it discusses criteria and difficulties in implementing the approach. Finally a method using a JPEG model based, frequency hopped, randomly sequenced pulse position modulated code (RSPPMC) is described. This method supports robustness of embedded labels against several damaging possibilities such as lossy data compression, low pass filtering and/or color space conversion.
1 Introduction
The electronic representation and transfer of digitized multimedia information (text, video, and audio) h
李安喜ave increased the potential for misuse and theft of such in-formation, and significantly increases the problems associated with enforcing copyrights on multimedia information [1,2]. These problems are rooted from the intrinsic features of the digitally formated information: (1) making copies is easy and inexpensive; (2) each copy is exactly identical to the original; and (3) dis-tribution of the copies (e.g. via network or floppy) is easy and fast. For this reason, creators or publishers of multimedia materials fear providing their works for usage in new multimedia services, and are seeking technical solutions to the problems associated with copyright protection of multimedia data.
These problems have recently raised attentions in na-tional IT (information technology) programmes, for example, NII (National Information Infrastructure) launched in the United States in 1993 established a working group on Intellectual Property Rights which is mainly concerned with copyright law and its ap-plication and effectiveness in the context of NII [3]. Several projects are currently or have recently been concerned with copyright and related issues in the dig-ital world, for example, the EC ESPRIT project CITED (Copyright in transmitted electronic data) [4] and COPICAT (Copyright Ownership Protection in Computer Assisted Training) [5], and the EC RACE project ACCOPI (Access Control and Copyright Protection for Images) [6].
In [2], we have summarized that the technical mecha-nism of copyright protection for information in d
igital form can be divided into three levels: access control, use right control, and labeling-based mechanism. This paper addresses the problems in developing the last level of mechanism, and presents a JPEG-based meth-od of labeling image for copyright protection. Although some attention has been given to stegano-graphic labeling and similar problems [7], there exists no technology designed to secretly embed a robust and invisible (hidden) copyright label in images. In partic-ular, no current method adequately addresses the pos-sibilities of using data compression, low pass filtering and/or simply changing the file format to remove an embedded code. Therefore, one of the main goals of this paper is to define a reasonable set of functional re-quirements and design criteria for an image copyright labeling method, and to furthermore demonstrate that the main difficulties involved in designing such a sys-tem can be solved.
The discussion begins with a section which outlines the functionality of the proposed system and general design criteria for the novel embedding technique. A specific method based on the JPEG compression stan-dard for embedding copyright labels in image data is then presented.
2 Requirements and possible attacks
In order to be effective and workable in a multimedia environment, the copyright label must be difficul
t to remove and survive processing which does not seri-ously reduce the value of the image. This encompasses a wide range of possibilities including format conver-sions, data compression, and low pass filtering. In addition to copyright labeling of broadcast images, ap-plication areas for steganographic labeling techniques include copyright and/or secure records labeling of electronic publishing, facsimiles, scientific imaging, and medical imaging.
Requiring the copyright label to be a reliable property identification tool imposes following basic functional requirements on the system:
(1)The image must contain a label or code, which
marks it as property of the copyright holder. (2)The image data must contain a user code, which
verifies the user is in legal possession of the data.
(3)The image data is labeled in a manner which allows
its distribution to be tracked.
协作
学习It is assumed, the main purpose of any attack would be to make the embedded label unverifiable. There are es-sentially two general ways to make the embedded label unverifiable: (1) alter the image data to render the copyright label unreadable, and (2) show that the label is not a reliable identification tool.
In addition several properties of digital data and design constraints, which are related to preventing attack on the copyright label, should be considered carefully. First, forgery of a digital copyright can only be pre-vented, if a forger cannot produce a valid copyright code. Second, the basic nature of digital images ensur-es that the copyright label can be easily altered if an at-tacker can identify the label data. Third, most digital images found in a multimedia environment can be low pass filtered, transformed to a different format or color space, or carefully re-quantized and compressed with-out significantly altering the images appearance or af-fecting its value. Finally, the image data is the only ran-dom sequence available to mask the data, and the statistics of the images, although generally unknown, are not under the control of the copyright system.
In sum, each of these points represents a potential means of attacking the copyright label and the follow-ing functional specifications are designed to prevent these attacks:
(1)A secret key type encryption code must be created
using the unique identification of a work and used as the copyright label to prevent forgery of labeling.
(2)The image data must camouflage the copyright la-
bel code both visually and statistically to prevent an attacker from finding and deleting it. The functional requirement stating that the copyright label appears to be part of a normal image sequence and visually transparent is designed to prevent this attack. (3)The signals used to embed the copyright label must
contain a noise margin to resist damage if the image is processed or compressed.
(4)The system must be designed in such a way that the
copyright labels locations and the same copyright
code are not used repeatedly for embedding codes in different images to prevent the label from be found by comparing different images signed by the same owner.
The noise margin created by modeling the lossy com-pression allows for some loss of energy in the
pulse, before the pulse becomes unreadable. Therefore if the pulse energy is concentrated at low frequencies, the embedded code should be relatively robust. Unfortu-nately, the final consideration with regard to pulse de-sign and visual camouflaging, is in direct conflict with using low frequency pulse shapes. Specifically, it is widely accepted that noise in the low frequencies com-ponents of images is more noticeable than noise in the high frequency components. This is the basic concept behind the very efficient transform and sub-band cod-ing techniques [8-10]. A reasonable trade-off between protection against processing attacks and visibility of the embedded code, is to make the pulses bandpass processes. Some additional design criteria must be de-veloped to allow both requirements to be met simulta-neously.
3 System Framework
The proposed approach, called Randomly Sequenced Pulse Position Modulated Code (RSPPMC) copyright labeling, is rooted in the well-known fact that typical digital images of people, buildings and natural settings can be considered as non-stationary statistical pro-cesses, which are highly redundant and tolerant of noise [8]. Hence, changes in the image data caused by moderate levels of wideband noise or controlled loss of information are hardly visibly noticeable, even when the altered images are compared directly with the original images.
Furthermore, the statistics of image sequences are only locally stationary and apriori unknown. More impor-tantly, the process which produces such a sequence has random properties, which prevent the sequence from being reproduced exactly by a second experiment. This type of random signal is ideally suited for the pur-pose of statistically masking a sparse sequence of mod-erately large pulses.
The RSPPMC method consists of splitting the problem into two components. The first component produces the actual copyright code and a random sequence of locations for embedding the code in the image. This component is designed with the intention of imple-menting it, using existing encryption and pseudo ran-dom number generation techniques [11,12]. In fact,
麻醉剂these methods are only discussed to establish a frame-work for developing a novel technique for embedding data in images. The second component actually embeds the code at the specified locations, using a sim-ple pulsing method, designed to appear to be a natural part of the image, which yet resists being damaged through simple processing techniques. This compo-nent consists of four steps:
(1)The position sequence is used to generate a se-
quence of pixel mapped locations where the code will be embedded.
(2)The blocks of 2-D image data, y(k,l) where k,l are
the indices of discrete image points, are locally transformed and quantized at the locations selected in step 1, in a manner reflecting acceptable informa-tion loss in the image for the application to produce
a 2-D image residual, n(k,l), in which the RSPPMC
will actually be embedded.
(3)The code pulses, i.e. high or low, representing the
binary code being embedded, are superimposed on the signal n(k,l) selected locations.
(4)The quantized data is decoded; and then, inversely
transformed to produced the labeled image data. In order to comply with functional requirements re-lated to robustness, the transformation used in the se-cond step includes the color space transforma
tions and sub-banding and/or frequency transformations to al-low direct access to the appropriate frequency bands in the gray scale component of an image. A quantization process is included in this step to guarantee that the la-bel will survive a specific amount of information loss.数值
模拟 A JPEG Compression Standard Based RSPPMC Copyright Label will be described in the next section.
4 Embedding a RSPPMC in Quantized JPEG Coefficients
Considering the functional requirement of robustness in a multimedia environment, the loss model in step 2 of the label process should be based on an industrial standard. From this perspective, image compression schemes used in GIF, TIFF, MPEG and JPEG are of in-terest. However, the wide spread use and growth of the JPEG [9] and MPEG formats and their efficiency in compressing images make transform coding the ob-vious choice for designing a copyright labeling sys-tem. Also, transform coding and/or sub-band coding techniques have the advantage of allowing direct ac-cess to specific frequency bands in the image, where the RSPPMC is to be embedded. This eliminates the problem of designing and detecting bandpass wave-lets.
The basic characteristics of images, which make trans-form quantization a useful image data compr
ession tool are (1) images are generally low pass processes, and (2) high frequency image components have little visual impact.
The DCT representation of images has been widely re-searched [10]. The typical characteristics of image DCT’s are also well known. Readers unfamiliar with the DCT and image transform quantization should re-fer to [8-10] for details.
The second point allows the higher frequency coeffi-cients to be more coarsely quantized than the low fre-quency components by the transform quantizer. Pre-dictably, the JPEG transform quantizer utilizes this fact by increasing qs(k,l) as a function of the increasing frequency vector normal.
Using these assumptions, several signals can be derived from the image data Y(i,j), which naturally contain pulses meeting the requirements outlined in section 2. One of the simplest is the sub-block signal, N(k1,l1,k2,l2)+ȧY Q(k1,l1)ȧ–ȧY Q(k2,l2)ȧ (1) where Y Q(k1,l1), Y Q(k2,l2) are the quantized coeffi-cient values at the selected locations. This non-sta-tionary random process should have an expected val-ue of approximately zero if ȧk1,l1ȧis approximately equal to ȧk2,l2ȧ. Also, the signal should have a mod-erate variance level in the middle frequency ranges, i.e. 1.5tȧk,lȧt4.5, where scattered changes in the image data should not be noticeably visible. The spe-cific frequencie
s being used to embed the pulses will be ”hopped” in this range to increase the robustness of the signal and making it more difficult to find. The principle being employed here is identical to the con-cept of frequency hopped spread spectrum commu-nications [13].
A logical choice for the detection of ”highs” and ”lows”, based on the signal defined in (1) is decided high if:
创业王
论坛N(k1,l1,k2,l2)u0, (2a) and decided low if,N(k1,l1,k2,l2)t0. (2b) However, embedding the code in this signal must also take into account the JPEG quantization process and any noise margin added to the pulses in the code. Therefore, the test for a written high is set as:ȧY Q(k1,l1)ȧuȧY Q(k2,l2)ȧ+p, (3a)
where p is a noise margin factor. The corresponding equation for a written low is:
ȧY Q(k2,l2)ȧuȧY Q(k1,l1)ȧ+p. (3b) Standard JPEG compression uses a ”quality factor” to scale the quantization, allowing for different image qualities and compression factors. In order to guaran-tee that the copyright label will survive compressions up to a specific level compression, the quantization table should be scaled to the desired quality factor. Also, due to numerical problems (in
calculating quan-tization step size according to quality factor, and in the quantization process) which can occur if the image is quantized with a JPEG quality greater than the de-signed factor for embedding the copyright code, some conditions must be be met. They are not discussed in this paper because of the limited space.
The method used to embed the copyright label in the sequence, N(k1,l2,k1,l2), is not complicated. The high/ low pulse pattern of the copyright label code is forced on the natural sequence at the selected group locations using a minimum mean square error approach, if it does not occur naturally. More complicated pulse pat-tern may be developed for representing the high/low bit, e.g. to use combinations (i.e. relationships) of three quantized elements Y Q(k1,l1), Y Q(k2,l2), Y Q(k3,l3) to replace equation (3).
In summary, the random pulse signal and conditions for detecting naturally occurring highs and lows de-scribed in equations (1) - (3) are designed to survive a JPEG compression down to a specified quality level. Clearly, decreasing the quality factor for the copyright code will make the signal more robust. However, this will also reduce the number of naturally occurring bits in the sequence. In addition, a lower quality factor will increase the likelihood that the changes necessary to superimpose the embedded code on the signal will be noticeably visible.
5 Conclusions
Using the prototypes we have developed, the exper-imental results indicate that the design requirements, developed in sections 2 for embedding a copyright la-bel in image data, can be met, using the JPEG model based RSPPMC method developed in section 4. In par-ticular, it was demonstrated that a copyright label code could be embedded in several images, using pulses with sufficient noise margins to survive common pro-cessing, such as lossy compression, color space con-version, and low pass filtering.However, these results also indicate significant room for improvement in the method. One possibility for im-provement could be to use different frequency band sets for encoding the high and low pulses. Also, meth-ods could be developed to utilize image restoration techniques and pattern recognition techniques for veri-fying copyright labels. For example, pattern recogni-tion techniques could be used to read copyright labels from images which have been cropped. In addition, methods suitable for applications with special require-ments, such as cartography and medical imaging, are currently being investigated.
The authors would like to acknowledge Scott Burgett and Jochen Rindfrey, for their contributions to this work.
References
[1] B. Kahin, ”The strategic environment for protecting multi-media”, IMA Intellectual Property Project Proceedings, vol. 1, no.1, 1994. pp.1-8.
[2] E. Koch, J. Rindfrey, J. Zhao. ”Copyright Protection for Multimedia Data”, Proceedings of the International Confer-ence on Digital Media and Electronic Publishing (6-8 Decem-ber 1994, Leeds, UK).
生物技术的应用
[3] B.A. Lehman, R.H. Brown, ‘‘Intellectual Property and the National Information Infrastructure’’. Preliminary draft of the working group on intellectual property rights, July 1994. [4] G. Van Slype. Natural language version of the generic CITED model. ESPRIT II CITED Project 5469, June 28, 1994. [5] A.J. Kitson and D.T. Seaton (eds.). Copyright Ownership Protection in Computer Assisted Training (COPICAT), Esprit Project 8195, Workpackage 2 (Requirements Analysis), De-liverable 1, June 2, 1994.
[6] RACE M 1005: Access control and copyright protection for images (ACCOPI), Workpackage 1 Deliverable, July 1994.
[7] K. Mantusi and K. Tanaka, ”Video-Steganography: How to secretly embed a signature in a picture,” IMA Intellectual Property Project Proceedings, vol. 1, no. 1, 1994.
[8] A. K. Jain, Fundamentals of Digital Image Processing, Prentice Hall, Englewood Cliffs, NJ, 1989.
[9] G.K. Wallace, ”The JPEG still picture compression stan-dard”, Communications of the ACM, vol. 34, no. 4, April 1991. pp.30-40.
[10] K.R. Rao and P. Yip Discrete Cosine Transform: Algo-rithms Advantages, Applications, Academic Press. 1990. [11] G. J. Simmons, Contemporary Cryptology: The Science of Information Integrity, IEEE Press, New York, 1994. [12] B. Schneier, Applied Cryptography: Protocols, Algo-rithms, and Source Code in C, John Wiley & Son, Inc., New York et al., 1994.
[13] R. C. Dixon, Spread Sprectrum Systems, 2nd ed., Wiley, New York, NY, 1984.
豫公网安备41160202000603
站长QQ:729038198
关于我们
投诉建议