EP1319219B1 - Method for preventing counterfeiting or alteration of a printed or engraved surface - Google Patents

Abstract

A 'fragile' signature in the form of a numerical watermark is incorporated in a part or whole of the document. The watermark has asymmetric modulation of amplitude of a visible or invisible component. To detect the presence or absence of the signature the surface has to be of uniform color. Pre-treatment of the image is carried out in order to eliminate visible printed or engraved data which does not contain the watermark. During searching for the signature the image defining the original surface is predicted from the image of the signed surface. The signature can also be printed so that it is visible on the surface. During searching for the signature the numerical watermark is digitized by a scanner, or portable device. The non visible property of the ink is used to form the asymmetric modulation. Different signatures can be hidden in different regions of the surface.

Description

Technical area

The present invention relates to a method for preventing the counterfeiting or alteration of a printed or engraved surface.

State of the art

• les hologrammes, les impressions de motifs spéciaux
• les impressions avec encres spéciales
• les codes utilisant des encres invisibles
• les systèmes à puce

The usual systems intended to prevent counterfeiting or alteration of printed or engraved documents can be classified in different groups:

• holograms, special pattern prints
• prints with special inks
• codes using invisible inks
• smart systems

Holograms, special patterns and other decorations are difficult to reproduce as they require special equipment. They are specially designed to interfere with conventional photocopying systems so that the copy is visibly different from the original. These systems can be visually controlled without the aid of particular devices but have the disadvantage of being expensive, sufficiently known to be reproduced without problems by counterfeit experts, and finally their visibility is detrimental to the aesthetics of the protected object. (perfume packaging for example). Their visibility is also the reason for their limited effectiveness in that an attacker can easily identify the security element, either to copy it or physically delete it.

Special ink prints use the particular chemical properties of the ink to provide a definite response to a particular action. Thus, fluorescent inks become very bright when illuminated by a particular wavelength, some inks are even invisible to natural light, other inks change color depending on their orientation or temperature (and may reveal by warming the paper with a finger), etc. Special inks have as a common point to be particularly expensive and require changes in the usual industrial production chain (additional mask for offset, for example). In addition, although more robust to counterfeiting than the previous group, it is also possible to reproduce their effects to the extent that the pirate can control by itself the fidelity of his copy compared to the original as soon as he has the device reacting the ink.

Codes using invisible inks, unlike the two previous groups, hide digital information. These codes can be characters, barcodes, 2D codes, etc. In addition to its high cost and inherent to invisible inks, this system has two major drawbacks. On the one hand, because of the nature of the codes used, it is located on a certain part of the document or the packaging and it is therefore possible to destroy it without altering the entire surface. On the other hand, the codes used always have geometric peculiarities (bars, geometric figures, characters, etc.) clearly identifying them as anti-copy devices. This greatly facilitates the task of the pirate seeking to reveal and reproduce the ink. Moreover as soon as the pirate knows how to realize this reproduction, ipso facto holds the means of reproducing the code.

Finally systems based on memories or embedded processors combine the disadvantages of being very expensive, unsightly and localized. Their main application is to secure a communication, or to dynamically store information rather than distinguish an original from a copy.

The document EP 0 961 239 discloses a solution for embedding a watermark in an image such as a banknote image. The proposed solution is to modify by slight alterations of the ink color, the density or the texture of the surface of the banknote. According to one embodiment, the watermark will modulate the width or location of a line component that forms the image of the bill.

One of the aims of the present invention is to overcome the disadvantages of known methods for preventing counterfeiting or alteration of printed or digitally engraved documents.

For this purpose, the present invention relates to a method for preventing the counterfeiting or alteration of printed or engraved documents characterized by embedding a digital watermark in part or all of the document.

• réaliser un filigrane numérique vba(k) de type spatial en fonction du signe du bit b={-1,1} de l'information à cacher et de l'amplitude v de la modulation d'une composante de couleur ainsi que d'un générateur aléatoire a(k) défini par une clé, tel que la densité du filigrane soit de 2% ou moins,
• imprimer l'image c(k),
• imprimer par-dessus l'image c(k) le filigrane vba(k) pour des valeurs ba(k) > 0 uniquement afin d'obtenir une modulation asymétrique d'amplitude sur la surface à une résolution comprise entre 300 et 1200 dpi.

According to the invention, there is provided a method of printing a watermark and an image on a printed surface, this watermark being self-correlated and containing information to be hidden, this method being characterized by the following steps:

• to realize a spatial watermark vba (k) according to the sign of the bit b = {- 1,1} of the information to be hidden and the amplitude v of the modulation of a color component as well as of a random generator a (k) defined by a key, such that the density of the watermark is 2% or less,
• print the image c (k),
• print over the image c (k) the watermark vba (k) for values ba (k)> 0 only in order to obtain asymmetrical amplitude modulation on the surface at a resolution between 300 and 1200 dpi.

The digital watermark technique, also known as digital tattooing, is a technique for hiding information in a robust and imperceptible manner in multimedia data such as music, video, images, documents, etc. The information that is hidden is called the signature. This signature can be for example a number, a name or even an image. After the protection of multimedia data with a digital watermark we speak of signed image, signed video, etc.

So far, the technique of the digital watermark has only been used for the purpose of finding the signature on a possible copy, to prove the origin of the information.

"Hide" has a very specific meaning in this context: for example in the case of an image, we will slightly change the color of some pixels, and in the case of music we will change the sound a little from time to time.

"Imperceptible" means that the modifications introduced are such that an individual can not distinguish the original data from the data signed by his own senses. For example, a signed image must look exactly the same as a normal image, a signed music must look perfectly normal, as well as a video or any other data. The whole problem is to make sure that a computer is able to detect this hidden information when it escapes our senses. There are also applications where a visible watermark is acceptable or even desirable. This allows in particular to further increase the robustness and visual control of the presence of a watermark. The principle that remains is that the watermark must not be visually disturbing.

The "robustness" of a watermark means that one must be able to find the signature after any manipulation of signed data. Take for example the case of a signed image: it must be able to compress, print, scan or rotate without ever losing the signature.

Numerous publications have been made on the different techniques for hiding a watermark in an image, video or audio signal. As far as the images are concerned, the latter can be classified according to the technique used for the marking: some operate modifications directly in the spatial domain (see for example [1] M. Kutter, F. Jordan, F. Bossen, "Digital watermaking of color images using amplitude modulation," Journal of Electronic Imaging, Vol. 7, No. 2, pp. 326-332, April 1998 .), others operate these modifications in a transformed domain (for example the frequency domain) even intermediate domains like the wavelets (see [2] Shelby Pereira, Svyatoslav Voloshynovskiy and Thierry Pun, Optimized wavelet domain watermark embedding strategy using linear programming, Harold H. Szu and Martin Vetterli eds., Wavelet Applications VII (part of SPIE AeroSense 2000), Orlando, Florida USA, April 26-28 2000 .).

These techniques can also be used for video tagging, with some adaptations. Other techniques specifically dedicated to video tagging are also possible by defining new transformed domains such as 3D subbands or motion vectors (for example, see [3] patent US 5,960,081 Video watermarking using motion vectors and [4] patent application EP 0762417 A2 , Watermarking video in the compressed domain).

As already mentioned, the technique of the digital watermark has so far been used for the purpose of finding the signature on a possible copy to prove the origin of the information present on the copy, thanks to the presence of the watermark that the we find on the copy. This involved in all cases the use of a robust watermark.

In the method of the present invention, the purpose of incorporating the digital watermark on the surface is different, since its presence is intended to prevent counterfeiting or alteration of the surface concerned, that is to say to allow to prove that it is the authentic surface if the watermark is present or that it is a copy or that the surface has been altered if the watermark is missing. In the case where the watermark is incorporated for surface authentication purposes with respect to copies, the robustness of the watermark must be reduced so that a copy of the surface will result in a failure of the watermark reading. digital. This is called "fragile" watermark. A typical application is to prevent counterfeiting of securities such as banknotes. In the case where the watermark is incorporated in order to avoid the alteration of all or part of the surface, the watermark may be strong or fragile.

• Invisibilité
  Le filigrane est imprimé en utilisant des couleurs et des résolution imperceptibles à l'oeil nu. Cela permet donc de protéger par exemple un emballage sans que son graphisme en soit altéré, ce qui est important pour des raisons marketing.
• Non localité
  Le filigrane peut recouvrir la totalité de la surface d'un document imprimé. Il n'est donc pas possible de l'effacer sans altérer le document, par exemple en grattant la surface. En pratique, cette propriété permet par exemple d'éviter les marchés gris, c'est à dire les produits revendus par des distributeurs non autorisés. En effet, ces derniers effacent parfois le code (code 2D invisible par exemple) identifiant leur revendeur en « fraisant » la surface de l'emballage où le code est imprimé.
• Prix
  Le filigrane est imprimé en utilisant des systèmes d'impression traditionnels. En ce qui concerne l'impression industrielle (offset, etc), il s'intègre totalement dans la chaîne de production et n'induit aucun coût supplémentaire. En ce qui concerne l'impression personnelle (jet d'encre, laser, etc), il est totalement compatible avec les imprimantes du commerce. Dans les deux cas, la lecture se fait avec un scanner digital standard. Ce faible prix ouvre des marchés nouveaux : d'une part en ce qui concerne l'impression industrielle, les emballages de produits de luxe ou pharmaceutiques, ainsi que les certificats, les chèques, les billets d'entrées, etc. D'autre part pour l'impression personnelle, il permet à n'importe qui possédant un équipement standard de créer et vérifier des documents sécurisés et personnalisés. Par exemple un médecin peut cacher le nom des médicaments prescrits dans le papier utilisé pour imprimer son ordonnance. Il est possible de programmer une imprimante pour qu'elle cache un filigrane dans tout document imprimé permettant ainsi d'identifier plus tard la date de l'impression, le nom de l'utilisateur, etc.
• Stockage d'information
  En plus d'authentifier l'original, le filigrane contient une information numérique (typiquement de plusieurs dizaines de bits par centimètres carrés) qui est encodée ou retrouvée à l'aide d'une clé. En pratique ce stockage permet par exemple de sécuriser une information imprimée en texte visible (donc susceptible d'être modifiée). En effet il est alors possible d'encoder la même information dans le filigrane et donc de pouvoir détecter toute modification dans le texte du document (date, montant, identité, etc.). Une application concerne les contrats dont on veut s'assurer de la date d'émission. Un autre exemple avec les billets de banque : le numéro de série peut être caché dans chaque billet, ainsi il est impossible de créer des faux billets avec des numéros différents car il faudrait pouvoir générer à chaque fois le filigrane correspondant.
• Système de lecture et d'écriture à clé
  Afin de pouvoir créer et lire le filigrane, il faut utiliser la même clé. En contrôlant l'accès aux clés, on peut donc contrôler quand et par qui chaque filigrane est créé ou lu, ce qui est essentiel : en effet cela complique beaucoup le piratage consistant à créer un nouveau filigrane (le plus simple restant de copier un filigrane existant). D'autre part le pirate sera dans l'incapacité de vérifier qu'un filigrane a été copié avec succès (car pour le lire, il est nécessaire de connaître la clé utilisée pour le cacher). La sécurité offerte par ce système de clé est donc supérieure à celle d'une information imprimée par exemple avec une encre invisible et révélée par ultraviolet, où le pirate peut aisément vérifier et donc améliorer l'encre qu'il contrefait.
• Difficile à identifier visuellement
  Même en utilisant des dispositifs (filtres, microscopes) particuliers, il est difficile d'identifier la présence d'un filigrane car son aspect visuel se rapproche de celui du grain du papier. Il n'a pas de caractéristique géométrique simple et n'a de sens que pour le programme de détection et muni de la bonne clé. Pour tous les papiers de valeur qui font l'objet d'une analyse fine de la part des pirates, cette qualité est primordiale.
• Difficile à copier
  L'utilisation combinée de certaines couleurs (jaune sur fond blanc par exemple) et de hautes résolutions d'impression (1200 dpi par exemple) permet d'obtenir un filigrane difficile ou impossible à reproduire sur un équipement de copie classique.

The following list describes a set of features present only in isolation in known systems intended to prevent counterfeiting or alteration of printed or engraved documents mentioned above:

• Invisibility
  The watermark is printed using colors and resolutions imperceptible to the naked eye. This allows to protect for example a package without its graphics is altered, which is important for marketing reasons.
• Not locality
  The watermark can cover the entire surface of a printed document. It is therefore not possible to erase it without altering the document, for example by scraping the surface. In practice, this property makes it possible, for example, to avoid gray markets, ie products resold by unauthorized distributors. Indeed, these sometimes erase the code (invisible 2D code for example) identifying their dealer by "milling" the surface of the package where the code is printed.
• Price
  The watermark is printed using traditional printing systems. As far as industrial printing (offset, etc.) is concerned, it fully integrates into the production line and does not entail any additional cost. With regard to personal printing (inkjet, laser, etc.), it is fully compatible with commercial printers. In both cases, reading is done with a standard digital scanner. This low price opens new markets: on the one hand with regard to industrial printing, luxury or pharmaceutical packaging, as well as certificates, checks, tickets, etc. On the other hand for personal printing, it allows anyone with standard equipment to create and verify secure and personalized documents. For example, a doctor may hide the names of the medications prescribed in the paper used to print his prescription. You can program a printer to hide a watermark in any printed document so that the date of printing, the name of the user, and so on can be identified later.
• Storage of information
  In addition to authenticating the original, the watermark contains digital information (typically several dozen bits per square centimeter) that is encoded or retrieved using a key. In practice, this storage makes it possible, for example, to secure information printed in visible text (which can therefore be modified). Indeed it is then possible to encode the same information in the watermark and thus to be able to detect any modification in the text of the document (date, amount, identity, etc.). An application concerns contracts whose date of issue is to be determined. Another example with banknotes: the serial number can be hidden in each ticket, so it is impossible to create fake banknotes with different numbers because it would be necessary to generate each time the corresponding watermark.
• Key reading and writing system
  In order to create and read the watermark, you must use the same key. By controlling access to keys, we can control when and by whom each watermark is created or read, which is essential: indeed it complicates much the piracy of creating a new watermark (the simplest remaining to copy a watermark existing). On the other hand the hacker will be unable to verify that a watermark has been copied successfully (because to read it, it is necessary to know the key used to hide it). The security offered by this key system is therefore greater than that of information printed for example with an invisible ink and revealed by ultraviolet, where the hacker can easily check and thus improve the ink it counterfeits.
• Hard to identify visually
  Even using particular devices (filters, microscopes), it is difficult to identify the presence of a watermark because its visual appearance is similar to that of paper grain. It has no simple geometric feature and only makes sense for the detection program and has the correct key. For all the papers of value that are the subject of a thorough analysis by hackers, this quality is paramount.
• Hard to copy
  The combined use of certain colors (yellow on a white background for example) and high print resolutions (1200 dpi for example) makes it possible to obtain a watermark that is difficult or impossible to reproduce on conventional copy equipment.

Methods that are done entirely in the digital domain generally hide the watermark by increasing and decreasing the color intensity of certain points, which means that some pixels are brightened while others are darkened, as shown in the illustration. Figure 1 : the graph shows the variation of luminance of the pixels of an image according to their position X and for an identical position Y. The four peaks illustrate the effect of a symmetrical modulation of this signal obtained by increasing or decreasing its intensity locally.

There are, however, some cases where symmetrical modulation is impossible to achieve, either for mathematical reasons (image to be signed entirely white or black) or practical (related to the printing technique).

The Figure 2 gives an example of asymmetric modulation obtained by darkening the color of some pixels. This modulation can then be positive or negative depending on whether color is added or removed. The graph of the figure shows the luminance variation of the pixels of an image as a function of their position X and for an identical position Y. The two peaks illustrate the effect of an asymmetrical modulation of this signal obtained by decreasing only its intensity. The Figure 3 gives some examples of digital watermark images.

Thus, another object of the present invention is to propose a method for hiding and / or retrieving a digital watermark, characterized by the use of an asymmetrical modulation of the amplitude of a visible or invisible light component.

detailed description

• la Figure 1 illustre un exemple de modulation symétrique;
• la Figure 2 illustre un exemple de modulation asymétrique;
• la Figure 3 illustre des exemples de filigranes asymétriques;
• la Figure 4 illustre la mise en oeuvre du procédé intégré avec une technique d'impression offset standard;
• la Figure 5 illustre la mise en oeuvre du procédé avec une étape d'impression offset séparée;
• la Figure 6 illustre la mise en oeuvre du procédé avec une étape d'impression offset séparée;
• la Figure 7 illustre la mise en oeuvre du procédé avec une imprimante à jet d'encre;
• la Figure 8 est un schéma bloc du procédé de signature d'un matériau en trois étapes;
• la Figure 9 est un schéma bloc du procédé de lecture d'une image uniforme signée en trois étapes; et
• la Figure 10 est un schéma bloc du procédé de lecture d'une image non-uniforme signée en trois étapes.

The following description, given by way of example, refers to the accompanying drawing in which:

• the Figure 1 illustrates an example of symmetric modulation;
• the Figure 2 illustrates an example of asymmetric modulation;
• the Figure 3 illustrates examples of asymmetrical watermarks;
• the Figure 4 illustrates the implementation of the integrated process with a standard offset printing technique;
• the Figure 5 illustrates the implementation of the method with a separate offset printing step;
• the Figure 6 illustrates the implementation of the method with a separate offset printing step;
• the Figure 7 illustrates the implementation of the method with an ink jet printer;
• the Figure 8 is a block diagram of the three-stage material signature process;
• the Figure 9 is a block diagram of the process of reading a uniform image signed in three steps; and
• the Figure 10 is a block diagram of the process of reading a non-uniform image signed in three steps.

An example of symmetric modulation is illustrated in Figure 1 . The graph shows the luminance variation of the pixels of an image as a function of their X position and for an identical Y position. The four peaks illustrate the effect of a symmetrical modulation of this signal obtained by increasing or decreasing its intensity locally.

An example of asymmetric modulation is illustrated in Figure 2 . The graph shows the luminance variation of the pixels of an image as a function of their X position and for an identical Y position. The two peaks illustrate the effect of an asymmetrical modulation of this signal obtained by decreasing only its intensity.

Watermark printing

Different strategies are possible for printing an asymmetrically modulated watermark, depending on whether it is positive or negative. In addition, it is possible to choose either a separate print or a simultaneous print when printing another visual pattern (background, text or graphic).

One way to achieve positive asymmetric modulation is to use an overprint technique by printing the watermark over the actual colors of the material and other information already printed and thus ignoring local color variations on the surface of that material. This approach imposes that the values of the color components of the material can only be darkened during the signature since additional ink is added. Mathematically, this corresponds to a positive asymmetric modulation of the color of the points. In principle, this approach can be applied to any printing process. Some specificities of watermark printing may depend on the printing process. The particular cases of offset and inkjet type printing for performing a positive modulation are detailed below.

The Figure 4 illustrates the implementation of the above method using positive modulation with an offset type industrial printing technique in the case of simultaneous printing of the watermark. In this example, four-color printing 45 is performed (for example for a package 40) which means that four different ink colors are used for each of the yellow masks 41, cyan 42, magenta 43 and black 44. The digital watermark can With a single color, it is generally desirable to use for the watermark one of the colors already selected for standard printing. The Figure 4 shows how the different masks can then be applied. In this case, the watermark printing fully integrates into the standard industrial printing line and therefore does not induce any additional cost. For example, the yellow mask can be used simultaneously for two different things: on the one hand it contains the yellow component of the image to print and on the other hand it contains the image of the watermark. The computer tools used during the flashing of the offset film make it easy to achieve this integration.

Another possible alternative is to use a separate mask for the watermark as shown on the Figure 5 . In this case, the watermark is overprinted during an additional step with a mask and possibly an ink of its own (here magenta). The mask 51 then defines the points of the watermark that are printed over the previously printed material 50. This method, although more expensive to implement by the printer, has the advantage of being able to easily change the watermark during the production. This allows for example to apply a watermark identifying different countries of sale to a series of identical packaging. It should be noted that when non-covering type inks are used, it is also possible to print the final image over the digital watermark, as shown in the illustration. Figure 6 . In this case, it is the inverse method that is used, the watermark being printed beforehand on the material, the final image being superimposed during an additional step. The yellow masks 61, cyan 62, magenta 63 and black 64 are used to overprint the pattern. Since the ink is transparent, the watermark 60 located under the pattern can still be detected in the final result 65.

Another usable printing method is of inkjet type as illustrated by the Figure 7 . The illustration shows an example of an ink jet printing system using four colors, yellow 71, cyan 72, magenta 73 and black 74, their print heads 75 and the printed material 70. The watermark is overprinted on the material. The implementation of an ink jet printer for printing a watermark is particularly simple in that the vast majority of printer drivers automatically manage the color mixing to obtain a particular hue. The step of quadrichromic decomposition is therefore most often useless. It should be noted, however, that, depending on the drivers and printers, it may sometimes be desirable to choose a watermark color corresponding to the fundamental colors of the printer, in order to avoid getting raster colors or problems with the printer. alignment between dots of different colors. As with offset printing, watermark printing can be simultaneous with information or patterns intended to be printed normally. It is also possible to print the watermark separately, over or under the final pattern. In particular, text can be overprinted on the signed material itself, this text possibly being linked to the watermark. For example, the key figures of a contract can thus be hidden in the watermark of the paper and thus make it possible to guarantee its integrity.

The realization of a negative modulation can be carried out during a simultaneous printing by following the same principle as described above because it is always possible to subtract from the color at the level of the electronic file: on the pattern to be printed, the points corresponding to the watermark are then cleared up. To achieve a separate print with a negative modulation, it is necessary to use a particular ink: for example, in the case of a visible ink, one solution is to use a type of ink covering. The synthesis of different possibilities for printing the watermark is presented in the table below: Simultaneous printing Separate printing Modulation Possible Possible positive asymmetric by overprinting or under pressure Modulation Possible Possible

Parameters controlling the visibility of the watermark

• Taille des points : il s'agit du diamètre des points du filigrane obtenus après impression. La taille minimale des points est fixée par la technologie d'impression. Des valeurs entre 300 et 1200 points par pouces sont courantes. Plus la taille des points est petites, moins le filigrane est visible.
• Couleur des points : En fonction de la couleur, de la texture et des motifs éventuels appliqués au matériaux, certaines couleurs peuvent être plus ou moins visibles. Par exemple, il est usuel d'utiliser une couleur jaune pour des fonds blancs (modulation positive séparée ou simultanée).
• Densité du filigrane : Cette dernière définit le ratio entre le nombre de points imprimés par unité de surface (mesurée également en points). Des valeurs typiques de 0.02 ou moins peuvent être utilisées. Une taille de point très fine permet d'augmenter la densité du filigrane.
• Quantité d'encre : Lorsque le procédé d'impression le permet, il est intéressant de jouer sur la quantité d'encre utilisée pour imprimer chaque point.
• Tramage : La technique de tramage (demi-teintes) permet de reproduire n'importe quelle couleur à partir des différentes couleurs fondamentales. Il est alors préférable que la taille du tramage soit suffisamment fine par rapport à la taille des points.
• Type d'encre : Des substances non visibles peuvent également être utilisées.

Whatever the type of modulation or printing chosen, the final visibility of the watermark as well as its fragility to duplication is controlled by a common set of parameters:

• Size of the points: this is the diameter of the points of the watermark obtained after printing. The minimum size of the points is fixed by the printing technology. Values between 300 and 1200 dots per inch are common. The smaller the point size, the less the watermark is visible.
• Color of points: Depending on the color, texture and possible patterns applied to the materials, some colors may be more or less visible. For example, it is customary to use a yellow color for white backgrounds (separate or simultaneous positive modulation).
• Watermark Density: The latter defines the ratio between the number of dots printed per unit area (measured also in points). Typical values of 0.02 or less can be used. A very fine point size makes it possible to increase the density of the watermark.
• Ink Quantity: When the printing process allows it, it is interesting to play on the amount of ink used to print each dot.
• Screening: The halftoning technique allows to reproduce any color from the different fundamental colors. It is then preferable that the size of the screening is sufficiently fine compared to the size of the points.
• Ink type: Non-visible substances can also be used.

The influence of some of these parameters is illustrated on the Figure 3 . Watermark 1 is visible. The lower visibility of the watermark 2 is obtained by simultaneously decreasing the density and the size of the points. The watermark 3 further includes a lightening.

Reading the watermark

The main difficulty lies in the ability to find the asymmetrical watermark. In general, most tattooing techniques can extract information from the signed image without using the original image. Some techniques first make a prediction of what the original image was from the signed image and can then infer what the signature is. This technique is still applicable in the present case. In the case where the material has a uniform and known initial color, it is possible to suppress this prediction. This is particularly the case of a blank sheet of paper. This makes it possible to increase the reliability of the detection and thus to reduce the visibility of the watermark to the extreme limit of sensitivity of an optical scanner. Consequently, it makes it very difficult to duplicate the signed material, for example by photocopying: in fact, the losses inherent in any reproduction system generally weaken. this signature below the detectability threshold. An application consists of including such a watermark on papers that you want to avoid copying, such as banknotes for example.

In order to increase the reliability of the detection, it is also possible to code the signature using the difference between pairs of pixels and then to calculate the average of these differences. From a statistical point of view, this will increase the correlation of the detection resulting in a more reliable signature.

Production

One embodiment consists in using as a basis a symmetric amplitude modulation spatial type digital watermark algorithm, for example that described in [1]. We speak of symmetrical amplitude modulation of a signal when the value of the signal is increased at certain points and decreased at other points. In this technique, a color component of a set of pixels c (k) is modified by a value v corresponding to the amplitude of the modulation and according to the sign of the bit b = {- 1,1} to be hidden as well as a random generator a (k) defined by a key and giving two values {-1,1}. $$\mathit{vs}\left(\mathit{k}\right)\mathit{\text{'}}\mathit{=}\mathit{vs}\left(\mathit{k}\right)\mathit{+}\mathit{v}\mathit{.}\mathit{b}\mathit{.}\mathit{at}\left(\mathit{k}\right)$$

In equation (1), the set of points defined by vba (k) constitutes the watermark ( Figure 8 , step 84) which is added to the original image c (k) to give the signed image c (k) '. It is the latter which is then printed according to the present invention.

In the case of a positive asymmetric modulation (watermark superimposed for example), it is no longer the image c (k) ' but the watermark itself vba (k) which is printed over an image c ( k) . Indeed, the component c of the support (blue, luminance, etc.) already has an initial value o (k) and can only be increased during the overprinting. The following formula is then applied: $$\mathit{Yes}\mathit{b}\mathit{.}\mathit{at}\left(\mathit{k}\right)\mathit{>}\mathit{0}\mathit{so}\mathit{vs}\left(\mathit{k}\right)\mathit{\text{'}}\mathit{=}\mathit{o}\left(\mathit{k}\right)\mathit{+}\mathit{v}\mathit{.}\mathit{b}\mathit{.}\mathit{at}\left(\mathit{k}\right)\mathit{if\; not}\mathit{vs}\left(\mathit{k}\right)\mathit{\text{'}}\mathit{=}\mathit{o}\left(\mathit{k}\right)$$

The Figure 8 gives a block diagram of the complete process: the set of points constituting the watermark 85 is calculated 84 from the value of the bit to hide 81 and the key 82 defining the random sequence a (k). The value of the points being positive or negative, as defined by equation (1). The equation (2) is equivalent to thresholding the values of the watermark 85 by keeping only the positive values and then adding these values 87 to the image to be signed 83 to obtain the signed image 89. By comparison with the formula (1) corresponding to a symmetrical modulation of the amplitude according to the sign of ba (k), this technique is called "asymmetrical amplitude modulation". In addition, the sign of the modulation ba (k) being positive, the modulation is said to be positive.

$$\mathit{c}\left(\mathit{k}\right)\mathit{ʹ}\mathit{=}\mathit{M\; sinon}$$

où M est la valeur maximum autorisée par le masque, c'est-à-dire la valeur correspondant à la couleur du document avant signature. L'équation montre clairement la modulation positive par rapport à zéro et illustre également le fait qu'aux positions où le filigrane est caché, l'image sous-jacente n'est pas prise en compte (domination du filigrane sur les valeurs originales). Ce procédé présente l'avantage que le nombre effectif de points contribuant au filigrane augmente, pouvant atteindre un facteur 2 dans le meilleur cas.In the case where the watermark is realized by simultaneous printing, the method can be further improved by operating in such a way that the watermark dominates over the values of the original mask. Mathematically, this concept is formalized as follows: $$\mathit{vs}\left(\mathit{k}\right)\mathit{\text{'}}\mathit{=}\mathit{0}{}_{}\mathit{if\; b}\mathit{.}\mathit{at}\left(\mathit{k}\right)\mathit{<}\mathit{0}$$

$$\mathit{vs}\left(\mathit{k}\right)\mathit{\text{'}}\mathit{=}\mathit{M\; otherwise}$$

where M is the maximum value allowed by the mask, that is to say the value corresponding to the color of the document before signature. The equation clearly shows the positive modulation with respect to zero and also illustrates the fact that at the positions where the watermark is hidden, the underlying image is not taken into account (dominance of the watermark on the original values). This method has the advantage that the effective number of points contributing to the watermark increases, up to a factor of 2 in the best case.

It is also possible to obtain a negative modulation by printing a uniform color u "pierced" by the watermark. Equation (2) then becomes: $$\mathit{Yes}\mathit{b}\mathit{.}\mathit{at}\left(\mathit{k}\right)<\mathit{0}\mathit{so}\mathit{vs}\left(\mathit{k}\right)\mathit{\text{'}}\mathit{=}\mathit{o}\left(\mathit{k}\right)\mathit{+}\mathit{u}\mathit{-}\mathit{v}\mathit{.}\mathit{b}\mathit{.}\mathit{at}\left(\mathit{k}\right)\mathit{if\; not}\mathit{vs}\left(\mathit{k}\right)\mathit{\text{'}}\mathit{=}\mathit{o}\left(\mathit{k}\right)$$

In all cases (positive or negative asymmetric modulation), if the random generator. a (k) generates the same number of positive and negative values, it follows that, statistically, half of the pixels c (k) is modified. If the value of v is chosen sufficiently weak and that the fineness of impression is high enough, the printing of these points can be done in an invisible way.

The new value of the points c (k) ' can be measured on the printed sheet using an optical scanner. Two cases then arise depending on whether the color of the material is uniform and known or not.

In the first case, the information b is then easily found to the extent that o (k) = Constant, v and a (k) are all known in advance. The multiplicity of the modified points creates a redundancy making it possible to ensure the robustness of the technique to noise by statistical correlation. The Figure 9 is a block diagram describing the process: the signed image obtained by scanner is subtracted from the original image in order to restore the watermark. The bit constituting the signature is then calculated, optionally, an additional filtering step may be performed if visible information has been printed over the signed uniform image. The signed image 91 is previously filtered 92 in order to eliminate any noises (scratches, stains, text printed over the watermark, etc.). The resulting image 93 is then subtracted 94 from the original image 95 in order to extract the watermark 96. The value of the bit b is then found according to the standard watermark detection techniques, as described in the article [5] M. Kutter, "Watermarking resisting to translation, rotation, and scaling.", Proceedings of SPIE International Symposium on Voice, Video, and Data Communications, November 1998 , which essentially consists of inverting equation (2) and statistically correlating the value of bit b found on several pixels k in order to guarantee good robustness to possible errors that may occur, for example, during the digital acquisition of the image .

This method can be generalized to several bits b and can then code any digital information such as a number or a string of characters.

The second case is illustrated by the block diagram of the figure Figure 10 where the original image is predicted from the signed image, the signed image is subtracted from the predicted image in order to render the watermark, the bit constituting the signature is then calculated. A denoising filter 105, for example of the Wiener type, is used to make a prediction 106 of the original image o (k) from the signed image 101. The difference 102 between these two images then constitutes the watermark 107 which can be decoded 103 using the key 108 to find the bit 104 in the same manner as before ( Figure 9 ). Since the prediction error is significantly greater than in the first case, the number of bits b coded in this way is systematically lower.

In practice it can also be useful to print visible information over the digital watermark. This is the case for example of a blank sheet of paper that has a digital watermark and over which is printed a text. This is achievable by choosing different colors or intensities for the watermark and for visible information. It is then possible to filter the image before the watermark is detected ( Figure 9 , step 92) to differentiate the watermark from the printed text and thus eliminate the parts not having the watermark. One method is for example to use the blue component for the watermark and to print the document text in black.

Finally, the implementation of the detection requires an optical scanner capable of scanning the document on which the watermark is printed. The positioning on the scanner is never perfect, it is necessary to be able to find the information coded by the watermark after possible translations and rotations. A suitable technique is to use the method described by [5] which is based on a self-correlated watermark (to compensate for rotations) and an inter-correlation technique (to compensate for translations).

Other applications

The method can also be applied to other sectors than printing. For example, it is possible to use a laser to engrave metal surfaces, stones, ceramics, etc., and thus encode a digital watermark. The applications concerned are, for example, parts of the automotive or aerospace industry or luxury articles in the jewelery or objects sectors. of values. It is also possible to hide watermarks on CD-ROMs or audio CDs, on the screen-printed side or on the engraved side (ink or laser).

Claims (4)

1. Method of printing a watermark and a image on a printed surface, this watermark being auto-correlated and containing an information to hide, this method being characterized in the following steps :

   - realizing a digital watermark vba(k) of the spatial type in function of the sign of the bit b={-1;1} of the information to hide and of the amplitude v of the modulation of a color component as well as a random generator a(k) defined by a key, so that the density of the watermark is 2% or less,

   - printing the image c(k),

   - printing over the image c(k) the watermark vba(k) for the values ba(k) >0 only in order to obtain an asymmetrical amplitude modulation over the surface at a resolution comprised between 300 and 1200 dpi.
2. Method of printing a watermark according to claim 1, characterized in that the watermark covers the entire surface.
3. Printed surface comprising a image and a watermark according one of the claim 1 or 2.
4. Printed surface according to the claim 3, characterized in that the watermark is printed on a material having an uniform color.

Images