Elsevier

Signal Processing

Volume 66, Issue 3, 28 May 1998, Pages 385-403
Signal Processing

Robust image watermarking in the spatial domain

https://doi.org/10.1016/S0165-1684(98)00017-6Get rights and content

Abstract

The rapid evolution of digital image manipulation and transmission techniques has created a pressing need for the protection of the intellectual property rights on images. A copyright protection method that is based on hiding an ‘invisible’ signal, known as digital watermark, in the image is presented in this paper. Watermark casting is performed in the spatial domain by slightly modifying the intensity of randomly selected image pixels. Watermark detection does not require the existence of the original image and is carried out by comparing the mean intensity value of the marked pixels against that of the pixels not marked. Statistical hypothesis testing is used for this purpose. Pixel modifications can be done in such a way that the watermark is resistant to JPEG compression and lowpass filtering. This is achieved by minimizing the energy content of the watermark signal at higher frequencies while taking into account properties of the human visual system. A variation that generates image dependent watermarks as well as a method to handle geometrical distortions are presented. An extension to color images is also pursued. Experiments on real images verify the effectiveness of the proposed techniques.

Zusammenfassung

Die schnelle Entwicklung digitaler Bildmanipulationen und -über- tragungsverfahren hat eine drängende Notwendigkeit für den Schutz intellektueller Eigentumsrechte von Bildern erzeugt. Ein Schutz des Copyrights, der auf dem Verstecken eines ‘unsichtbaren’ Signals beruht, ist als digitales Wasserzeichen innerhalb des Bildes bekannt und wird in dieser Arbeit vorgestellt. Der Einschluß eines Wasserzeichens wird im räumlichen Bereich durch eine geringe Modifikation der Intensität zufällig ausgewählter Bildpunkte erreicht. Die Erkennung des Wasserzeichens erfordert nicht die Vorlage des Originalbildes und wird durch Vergleich der mittleren Intensität der markierten Bildpunkte mit derjenigen der nicht markierten Punkte erreicht. Zu diesem Zweck wird ein statistischer Hypothesentest benutzt. Die Punktmodifikation kann in einer solchen Weise durchgeführt werden, daß das Wasserzeichen gegenüber einer JPEG-Kompression und Tiefpaßfilterung resistent ist. Durch die Minimierung des Energieinhaltes des Wasserzeichensignals bei höheren Frequenzen wird dies erreicht, wobei die Eigenschaften des menschlichen Gesichtssinnes berücksichtigt werden. Es werden eine Variation, die bildabhängige Wasserzeichen erzeugt, sowie eine Methode präsentiert, die geometrische Verzerrungen behandelt. Ein Erweiterung auf Farbbilder wird auch verfolgt. Experimente mit echten Bildern bestätigen die Effizienz der vorgeschlagenen Methode.

Résumé

L’évolution rapide de la manipulation des images numériques et des techniques de transmission a généré un besoin pressant de protection des droits de la propriété intellectuelle sur les images. Une méthode de protection des droits d’auteur, basée sur le camouflage d’un signal ‘invisible’ dans l’image, connusous le nom de filigrane numérique (digital watermark) est présentée dans cet article. Le placement du watermark est opéré dans le domaine spatial par modification légère de l’intensité de pixels de l’image choisis aléatoirement. La détection du watermark ne requiert pas l’image originale et elle s’opère par comparaison entre l’intensité moyenne des pixels marqués et celle des pixels non marqués. Un test d’hypothèse statistique est utilisé à cet effet. Les modifications des pixels peuvent être faites de telle façon que le watermark soit résistant vis-à-vis de la compression JPEG et du filtrage passe-bas. Ceci est obtenu en minimisant l’énergie du signal de watermark dans les hautes fréquences tout en tenant compte des propriétés du système visuel humain. Une variante générant des watermarks dépendant de l’image ainsi qu’une méthode de prise en compte des distorsions géométriques sont également présentées. Une extension aux images couleur est également en cours de développement. Les expériences réalisées sur des images réelles permettent de vérifier l’efficience des techniques proposées.

Introduction

Digital media have revolutionized the way still images and image sequences are stored, manipulated and transmitted, giving rise to a wide range of new applications (digital television, digital video disc, digital image databases, electronic publishing, etc.) that are expected to have an important impact on the electronics and entertainment industry. One of the main features of digital technology is the ease with which images can be accessed and duplicated. However, this feature has an important side effect; it allows for easy unauthorized reproduction of information, i.e. data piracy. Due to this, protection of intellectual property rights, i.e. copyright protection of stored/transmitted digital images is a very important issue. One way to help protect images against illegal recordings and retransmissions is to embed an invisible signal, called digital signature or copyright label or watermark, that completely characterizes the person who applied it and, therefore, marks it as being his intellectual property. Obviously, the secure and unambiguous identification of the legal owner of an image requires that each individual or organization that produces, owns or transmits digital images (artists, broadcasting corporations, image database providers, etc.) uses a different, unique watermark.

Copyright protection is just one of the potential applications of embedding invisible data within images or other types of signal (e.g. audio signals), a technique usually referred to as data hiding or steganography. Other applications include authentication control, tamper-proofing (i.e. checking whether the content of an image has been altered or not) and insertion of invisible image annotations (e.g. scene/object description). For each of these applications, the embedded signal should possess a different set of properties. In this paper we would limit our discussion to digital watermarks serving the purpose of copyright protection. Digital watermarks of this type should be [14]:

  • undeletable by an ‘attacker’;

  • easily and securely detectable by their owner;

  • perceptually and statistically invisible;

  • resistant to lossy compression, filtering and other types of processing.

The creation of an algorithm capable of producing watermarks that fulfil all these contradicting requirements is not an easy task. A number of attempts to introduce copyright labelling techniques that comply with some or all of the above specifications have been reported lately in the literature 1, 3, 4, 6, 7, 9, 10, 12, 13, 15, 16, 17, 18, 21, 22, 23, 25, 26, 29, 30. However, research on copyright protection of images is still in its early stages and none of the existing methods is totally effective against attacks. The techniques proposed so far can be classified in two broad categories: (i) methods that embed the watermark by directly modifying the intensity of certain pixels 1, 4, 22, 25, 15, 26. (ii) methods that act upon selected coefficients of a properly chosen transform domain (DCT domain, DFT domain, etc.) 3, 7, 9, 13, 17, 18, 23, 30. Watermarking techniques can be alternatively split into two distinct classes depending on whether the original image is necessary for the watermark detection or not. Although the existence of the original image facilitates to a great extent watermark detection, such a requirement is rather difficult to be met in most real life applications. It would be, e.g., totally impractical for the owner of a large image database to keep double copies of its images for authentication and copyright protection purposes. Furthermore, searching within the database for the original image that corresponds to a given watermarked image would be very time consuming. In this paper we propose a watermarking technique that belongs to the class of intensity domain techniques, i.e. it embeds copyright information by modifying the intensity of a subset of the image pixels. The proposed method is actually an extension and continuation of the method reported by the authors in 16, 21, 22. The watermark casting algorithm allows for a flexible choice of the intensity modifications. This flexibility can be exploited to design watermarks that possess desirable properties like robustness against lossy compression and lowpass filtering. Watermark detection is carried out by using hypothesis testing and does not require the original image. Another important feature of the algorithm is its mathematical tractability that allows a thorough investigation of algorithm performance. Furthermore, the proposed watermarking technique can be easily combined with noise masking techniques to yield watermarks that are invisible.

The basic operating principle of the proposed algorithm presents certain similarities to the so-called Patchwork technique that has been independently developed in MIT Media Lab [1]. However, the two methods differ in the statistical approach adopted for the watermark detection. Furthermore, an extensive part of our paper is devoted to important extensions of the basic method (image dependent watermarks, robust watermark design by means of optimization techniques, handling of geometric distortions) that are not addressed in [1]. The proposed algorithm bears also certain similarities with other methods 6, 9, 17 developed independently at about the same time or afterwards. However, the differences between our method and the above-mentioned techniques are rather important. For example, the algorithms proposed by Cox and Ruanaidh require the original image during the watermark detection whereas our method does not. The outline of this paper is the following. The basic algorithm is described in Section 2. Design of watermarks that are robust to filtering and compression is discussed in Section 3. Section 4deals with ways to incorporate properties of the human visual system in order to generate invisible watermarks. A method to handle geometrical distortions is presented in Section 5. Image dependent watermarks are proposed in Section 6as a means of further robustifying the algorithm. Extension to color images is treated in Section 7. Experiments on real images are presented in Section 8.

Section snippets

Basic watermarking algorithm

Consider an image I of dimensions N×M:xnm,0⩽n<N,0⩽m<M.A watermark pattern S is a binary pattern of the same size where the number of ‘ones’ equals the number of ‘zeros’:snm,0⩽n<N,0⩽m<M,snm∈{0,1}.

Using S we can split I into two subsets of equal size:A={(n,m)|snm=1},B={(n,m)|snm=0},|A|=|B|=12|I|=12N×M=P,I=A∪B.The digital watermark is superimposed on the image as follows:xsnm=xnm⊗fnm,where ⊗ is a superposition law (in our case addition), xsnm are the pixels of the watermarked image Is and fnm is

Immunity to subsampling

An important issue that should be examined about the proposed watermarks is their immunity to subsampling. Only the case of mean value subsampling is considered here. In this case, if the original image I was of size N×M, the subsampled image Isub is (N/2)×(M/2) pixels large, and the intensity levels xnm of its pixels are given byxnm′=14(x2n,2m+x2n+1,2m+x2n,2m+1+x2n+1,2m+1), 0⩽n<N/2,0⩽m<M/2.

In order to apply the detection algorithm on Isub, we generate a subsampled version S′ of the N×M

Robust watermark design

Unfortunately the method outlined in Section 2is not robust to compression using the well-established JPEG standard which achieves efficient image compression by combining the Discrete Cosine Transform (DCT) with appropriate DCT coefficient quantization schemes. This is due to the fact that the watermark signal fnm (8) is essentially low-power, white noise. As a consequence, it is heavily distorted by JPEG. Furthermore, the watermarks can be easily deleted by other operations such as mean or

Invisible watermarking using properties of the human visual system

Since watermarks are essentially additive noise superimposed on the image, they should be invisible so as not to affect the image quality to a great extent. Another important motivation for generating invisible watermarks is the fact that such watermarks would be difficult to be detected (and destroyed) by visual inspection. From this point of view, watermarking exhibits significant resemblance to lossy image compression [18], an application where invisible distortions are also highly

Handling geometrical distortions

A weak point of the watermarking methods described in the previous sections is that the pixels that form the subset A are specified only in terms of their spatial location and therefore, any geometrical distortion (e.g. line removal) can ‘fool’ the detection algorithm. However, this type of distortion can be treated using a correlation-based preprocessing module. The output of this module is an index d(n) that gives the translation that line n has undergone. d(n) can be fed into the detection

Image dependent watermarks

Using a single watermark on all images of equal size whose copyright is owned by the same individual is very convenient in terms of implementation simplicity and speed. However, such a practice is not a safe one. If, for example, someone can manage to obtain both the original and the watermarked version of the same image, he can easily recover the watermark signal by performing a simple subtraction of the two images and then eliminate the watermark from all images in which it exists. Another

Watermarks for color images

The watermarking techniques presented in the previous sections can be easily extended to handle color images. In this case, watermark casting is done by generating three different watermark patterns SR,SG,SB, one for each RGB channel, and modifying, for each channel, the intensity of the pixels that belong to the corresponding sets AR,AG,AB. The watermark casting and detection procedures for color images are exactly the same as for the corresponding procedures for grayscale images. Sets A, B

Experimental results

The resistance of the proposed watermarking algorithms to various distortions was studied in a series of experiments on grayscale images. The first set of experiments dealt with the resistance of the various techniques to JPEG compression. As it was mentioned in Section 4, when an image is subject to a certain distortion (in our case compression), both the probability 1−β of correct detection in a watermarked image and the probability α of erroneous watermark detection in an image that bears no

Conclusions

A new method for the copyright protection of images using digital watermarks has been proposed in this paper. The proposed method produces watermarks that are not detectable by visual inspection but at the same time are robust to JPEG compression and lowpass filtering. Variations that produce watermarks that are immune to geometric transformations and also image-dependent were also presented. An extension to color images was also presented.

References (30)

  • W. Bender et al.

    Techniques for data hiding

    Proc. SPIE

    (1995)
  • V. Bhaskaran, K. Konstantinides, Image and video compression standards: algorithms and architectures, Kluwer,...
  • A. Bors, I. Pitas, Image watermarking using DCT domain constraints, Proc. 1996 IEEE Int. Conf. on Image Processing...
  • O. Bruyndonckx, J.-J. Quisquater, B. Macq, Spatial method for copyright labeling of digital images, Proc. IEEE Workshop...
  • C. Chou, Y. Li, A perceptually tuned subband image coder based on the measure of just-noticeable-distortion profile,...
  • I.J. Cox, J. Kilian, T. Leighton, T. Shamoon, Secure spread spectrum watermarking for multimedia, NEC Research...
  • I. Cox, J. Kilian, T. Leighton, T. Shamoon, Secure spread spectrum watermarking for images, audio and video, Proc. 1996...
  • B. Girod

    The information theoretical significance of spatial and temporal masking in video signals

    SPIE, Human Vision, Visual Processing, and Digital Display

    (1989)
  • F. Hartung, B. Girod, Digital watermarking of raw and compressed video, Proc. SPIE 2952, Digital Compression...
  • C. Hsu, J. Wu, Hidden-signatures in images, Proc. 1996 IEEE Int. Conf. on Image Processing (ICIP 96), vol. III, 1996,...
  • N. Jayant, J. Johnston, R. Safranek, Signal compression based on models of human perception, Proc. IEEE 81 (10)...
  • H. Kinoshita, An image digital signature system with ZKIP for the graph isomorphism, Proc. 1996 IEEE Int. Conf. on...
  • E. Koch, J. Zhao, Towards robust and hidden image copyright labelling, Proc. IEEE Workshop on Nonlinear Signal and...
  • B.M. Macq, J.J. Quisquater, Cryptology for digital TV broadcasting, Proc. IEEE, June 1995,...
  • K. Matsui, K. Tanaka, Video-steganography: how to secretly embed a signature in a picture, IMA Intellectual Property...
  • Cited by (466)

    • Robust speech watermarking by a jointly trained embedder and detector using a DNN

      2022, Digital Signal Processing: A Review Journal
      Citation Excerpt :

      Digital watermarking emerged during the 1990s with a cornerstone publication [1] reviewing some of the most important spread spectrum watermarking techniques. Since then, a vast number of techniques have been developed for watermarking all types of digital multimedia [2,3], such as images [4–10] and videos [11]. Digital image watermarking remains the most researched part of the field, with many ideas being borrowed from it and applied elsewhere.

    • An Extensive Analysis of Digital Image Watermarking Techniques

      2024, International Journal of Intelligent Systems and Applications in Engineering
    View all citing articles on Scopus
    View full text