WO2005096614A2

WO2005096614A2 - Digital signal watermarking detection method and device

Info

Publication number: WO2005096614A2
Application number: PCT/FR2004/000484
Authority: WO
Inventors: Yann Bodo; Nathalie Laurent; Sébastien BRANGOULO
Original assignee: France Telecom
Priority date: 2004-03-02
Filing date: 2004-03-02
Publication date: 2005-10-13
Also published as: WO2005096614A3

Abstract

The invention relates to a method of detecting the watermarking of a digital signal (S') that has been watermarked with the application of a watermarking function to pre-determined attributes of the signal. The inventive method comprises the application of a watermarking key extraction function in order to: select (50) the watermarked attributes and to locate the co-ordinates thereof in a reference partition comprising as many zones (Z0, Z1) as values that can be taken by the marking key data; modify (52) the reference partition by inserting at least one new zone (Z2) between the aforementioned zones (Z0, Z1); assigning (56, 58) the extracted marking key datum with which each selected attribute is associated with the value of the zone containing the selected attribute or a weighted value between two neighbouring zone values as a function of the position of the attribute within said new zone if the attribute is positioned therein; and calculating (60) a correlation score between the initial marking key and the marking key thus extracted.

Description

Method and device for detecting watermarking of a digital signal

The present invention relates to a method of detecting watermarking of a watermarked digital signal by applying a watermarking function to at least a portion of predetermined attributes of the digital signal, and a device for implementing this method. . More specifically, the invention applies to a digital signal watermarked by the application of a watermarking function performed using a marking key comprising data with discrete values each of which data is associated with at least a selected attribute comprising the following steps:

- locate the coordinates of the selected attribute in a space comprising a reference partition defining as many zones as discrete values that the data of the marking key can take;

- assign a discrete value to each of the zones;

- possibly modify the coordinates of the selected attribute, so that it is located in the zone whose discrete value corresponds to the value of the data of the marking key with which the selected attribute is associated.

The watermarked digital signal is for example a fixed digital image or a video signal.

The document US Pat. No. 5,960,081 describes a method of watermarking motion vectors of a video signal. Watermarking consists in coding two bits of a marking key in the two coordinates of a selected motion vector. More precisely, a bit is carried by the parity of each coordinate. For a bit, coding consists in making the parity of the coordinate coincide with the parity of the bit to be coded, for example by incrementing the coordinate by one.

We can thus consider that the vector space is divided into blocks, in each of which two zones are defined (an even zone, an odd zone). One of the two zones is associated with a first binary value, while the other is associated with a second binary value.

On reception of such a video signal, in order to detect watermarking, it is then necessary to carry out the operation consisting in selecting the tattooed movement vectors and in positioning them in the reference space used during watermarking, in order to determine the binary values. they wear. A correlation calculation between the marking key detected in this way and the marking key used during watermarking finally makes it possible to determine whether the video is actually watermarked. Unfortunately, this solution is not without drawbacks, since the slightest attack can transform an even ordinate into an odd ordinate and vice versa. In this case, a change in parity between transmission and reception will have a direct consequence on the robustness of the watermarking detection method implemented. By "attack" is meant malicious attacks but also non-malicious attacks such as compression, or the change in spatial or temporal format of the video signal.

The invention aims to remedy this drawback and, more generally, aims to improve the robustness of existing tattoo detection methods, without increasing the complexity of the corresponding tattoo methods.

To this end, the subject of the invention is a method for detecting the watermarking of a digital signal of the aforementioned type, characterized in that it comprises the application of a function for extracting the marking key, the application of this function comprising the following steps: - select the tattooed attributes;

- Modify the reference partition by inserting at least one new zone between the zones of the reference partition, this new zone defining a neighborhood of the borders between the zones of the reference partition;

- allocate to the discrete value of the data of the extracted marking key with which each selected attribute is associated, the discrete value of the zone in which the selected attribute is located if it is located in one of the zones the reference partition, outside the new zone;

- allocate to the discrete value of the data of the extracted marking key with which each selected attribute is associated, a weighted value between two discrete values of neighboring areas, if it is in the new area, this weighted value being dependent the location of the attribute within this new zone according to a predetermined function;

- calculate a correlation score between the initial marking key and the extracted marking key to decide on the presence or absence of the marking key in the selected attributes.

This approach effectively improves the smoothness of the digital watermarking detection process by defining an area of least confidence, within which the contribution of the data extracted from the marking key is weighted so as to take account of the greatest sensitivity to attacks, attributes located in this area. It will be noted that this improvement in the tattoo detection method does not add any complexity to the tattoo method itself. In particular, this approach makes it possible to significantly improve the detection of watermarking of a video signal, when the attributes chosen are its motion vectors.

A method for detecting the watermarking of a digital signal according to the invention may also include one or more of the following characteristics: - the predetermined function is a sigmoid function;

- the space is divided into several parts, in each of which, at least two zones are defined, each of these zones being situated inside another of these zones, with the exception of an external zone whose external contour corresponds to the contour of the part in which it is located, each new zone inserted during the extraction step surrounding an internal zone;

- the marking key being with binary values, the reference partition comprises a first zone and a second zone complementary to the first, and, during the extraction of the marking key:

The reference partition is modified by inserting at least a third zone between the first and the second zone, this third zone defining a neighborhood of the border between the first and the second zone,

• the bit of the extracted marking key to which each selected attribute is associated is allocated a weighted value between 0 and 1, if it is in the third zone, this weighted value being dependent on the location of the attribute within this third zone according to a predetermined function;

the surfaces of the first and second zones are each substantially equal to 45% of the surface of the part in which they are defined, the surface of the third zone inserted during the extraction step being substantially equal to

10% of the surface of this part;

- The reference partition produced during watermarking comprises a reference grid comprising blocks of predefined dimensions, each block comprising as many zones as discrete values that the data of the marking key can take;

- At least one sub-block centered inside the block is defined in each block of the reference grid, the areas of this block being defined by the spaces between the sub-blocks and the block;

- the blocks and sub-blocks of the reference grid are rectangular; the method of watermarking the digital signal further comprising the following steps: • hierarchical calculation of several successive levels of attributes, the attributes of a given level each being associated with several attributes of the level directly below;

• selection of at least some of the attributes belonging to the highest level;

• application of the watermarking function to each selected attribute leading to the calculation of a parameter for modifying this attribute;

• application of the modification parameter of the selected attribute to the attributes of a lower level associated with this attribute, for each attribute of the highest level selected during the watermarking:

• we extract the tattooed attributes associated with this attribute;

• an average attribute is calculated equal to the average of the tattooed attributes associated with this attribute; and

• the function of extracting the marking key is applied to the calculated average attribute;

- the attributes of a given level are each equal to the average of the attributes of the directly lower level with which they are associated.

The invention also relates to a device for detecting watermarking of a digital signal, for implementing a method as described above, characterized in that it comprises:

- means for selecting the tattooed attributes;

means for modifying the reference partition by inserting at least one new zone between the zones of the reference partition, this new zone defining a vicinity of the borders between the zones of the reference partition;

means for allocating, to the discrete value of the data of the extracted marking key with which each selected attribute is associated, the discrete value of the zone in which the tattooed attribute is located if it is located in one of the zones of the reference partition, outside the new zone;

means for allocating, to the discrete value of the data of the extracted marking key with which each selected attribute is associated, a weighted value between two discrete values of neighboring zones, if this is found in the new zone, this weighted value being dependent on the location of the attribute within this new zone according to a predetermined function; means for calculating a correlation score between the initial marking key and the extracted marking key to decide on the presence or absence of the marking key in the selected attributes.

The invention will be better understood on reading the description which follows, given solely by way of example and made with reference to the appended drawings in which:

- Figure 1 shows the successive steps of a watermarking method according to the invention;

- Figure 2 shows a reference grid used for the application of a watermarking function according to a first embodiment of the invention; - Figures 3a, 3b, 3c and 3d illustrate rules for applying the watermarking function according to the first embodiment;

- Figure 4 ^"shows the successive steps of a tattoo detection method according to the first embodiment of the invention;

- Figure 5 shows a device for implementing the method of Figure 4; - Figure 6 shows a detail of the reference grid of Figure 2, after modification thereof, by application of a detection method according to the first embodiment;

- Figure 7 shows a detail of a modified reference grid according to a second embodiment of the invention; - Figure 8 shows the general appearance of a sigmoid function.

Before describing a method for detecting the watermarking of a video signal according to the invention, it is necessary to detail a possible way of applying watermarking to this signal which is consistent with this detection method.

The watermarking method shown in FIG. 1 is implemented by computer means of the conventional type. It is, for example, a microcomputer which comprises a central computing unit associated with memories of the RAM and ROM type.

The tattooing process includes a first step 10 of motion estimation.

During this step, a video source signal S is supplied at the input of the microcomputer and, conventionally, a matrix of motion vectors 12 is associated with any image of the video signal S. This matrix of motion vectors makes it possible to generate a predicted image of the image considered from, for example, the previous image of the video signal, by moving blocks of pixels thereof, as a function of the calculated motion vectors. The calculation of motion vectors is carried out for example by the application of a "block matching" process. This process consists, for each block of pixels in the image considered, to evaluate the best motion vector making it possible to reconstruct this current block from a block of the same size of the previous image moved using the motion vector. To do this, we carry out a search around the current block in the previous image, in order to determine the motion vector which minimizes a conventional cost function known as DFD for "Displaced Frame Difference", representing the difference between the block of the previous image moved and the current block in the image considered.

The motion vector matrix 12 represented in this figure comprises thirty-six motion vectors denoted V _J , with 1 <i <9 and 1 <j <4. Of course, the number of motion vectors will generally be higher. Only thirty-six have been shown, for the sake of simplicity in the following description.

Then, during a step 14, a higher level of motion vectors is calculated 16. In this higher level, each motion vector is associated with four motion vectors of the lower level 12. In fact, the blocks corresponding to each motion vector of the lower level 12 are grouped by four into macroblocks. The upper level therefore has 9 vectors denoted V _λ to V _g . Each vector

movement V _t of the upper level is representative of a macro-block and is calculated as follows:

_ _> 1 4 „

V ^• = - A Y-> v '- ^J

⁴ 7 = 1 Step 14 could be repeated several times, so that a hierarchy of motion vector levels could be created. In this example, we limit ourselves to a single iteration, that is to say to two levels of motion vectors.

Then, during a step 18, a set 20 of motion vectors among the motion vectors 16 of the upper level is selected in a deterministic manner, or in a pseudo-random manner from a secret key, or from a mask. . In this example, the set 20 of selected motion vectors consists of the vectors

'6' 5 '2 9 "

During the next tattooing step 22, the selected vectors 20 are modified by applying a tattooing function to them, using a marking key W. In this example, the number of motion vectors 20 selected is related to the size of the marking key W and W is a binary word comprising four bits. This is the reason why four motion vectors were selected during step 18. Thus, during step 22, the watermarking function is configured to insert the first bit of the marking key W in the motion vector V ₆ , the second bit

in the motion vector V ₅ , the third bit in the motion vector V ₂ and the

fourth bit in the motion vector V ₉ . The watermarking function chosen will be described more precisely with reference to FIGS. 2 to 6.

Following this step 22, four new motion vectors V, V are obtained,

V and V. These vectors verify for example the following relationships:

^ = ^ ₊ ^, v = V ₂ + âf ₂ , v = V ₉ + ^~ δv ₉ , in the case where the transformation is additive. One can however imagine other more general non-additive transformations. During the following step 25, the watermarking function with the parameters δV _t calculated during the previous step is applied to the motion vectors of the lower level which are associated with the selected vectors 20. So for example, the transformation

SV ₆ applied to the motion vector V ₆ , is also applied to the vectors

correspondents of the lower level, that is to say the vectors V ₆ ,, V _{6 2} , V _{6 <3} , V _{6 Λ} . Sixteen new motion vectors 26 are thus obtained at the lower level, satisfying the following equation:

V / e {6; 5; 2; 9}, V / e {1,2,3,4} ^ = V _tJ + όΥ,

These sixteen new motion vectors replace the corresponding initial vectors to provide a new matrix 28 of motion vectors. This new matrix 28 makes it possible to obtain a watermarked version S 'of the video signal S.

In fact, during a last step 30, a motion compensation is carried out using this matrix of motion vectors 28, from the image which served as a reference for the motion estimation during the step 10, that is to say the image preceding the image considered. During this step, two methods are possible. A first method consists in performing the motion compensation using all the motion vectors of the matrix 28 to find a watermarked version of the image considered.

A second method consists in carrying out the motion compensation only for the blocks corresponding to the motion vectors modified by the watermarking operation and in leaving the other blocks of the image considered unchanged, in order to obtain a better image quality. .

Steps 10 to 30 previously described can be repeated on several images of the video signal S. The redundancy thus obtained makes it possible to increase the robustness of the watermarking.

One can also achieve redundancy of watermarking by associating each bit of the marking key W with several motion vectors selected in the upper level.

A watermarking method such as that described above does not fit into a coding / decoding system. But the modifications to be made so that it can be implemented in such a system, implementing an analysis of movement, are obvious. Indeed, the video coding methods, such as the methods using the MPEG2 standard or the MPEG4 standard, generate motion vector matrices on which it is possible to apply the method described above, when coding the video signal. Note that step 14 is optional. Without the application of this step, the selected motion vectors are the motion vectors of the lower level, and we can compensate for the less significant spreading of the watermark in the image, by the selection of several motion vectors for each bit of the key. marking. This however has the effect of limiting the performance of the tattoo in terms of invisibility. Indeed, an advantage of the hierarchical approach is to apply the same deformation on neighboring motion vectors in the lower level.

In a particular embodiment of the invention, the insertion of the marking key in the video signal S can be distributed over several pairs of images. In other words, it is possible to insert only part of the marking key W in selected motion vectors from a matrix of motion vectors calculated between two successive images, then insert the other part of the marking key on one or more several motion vector matrices calculated between other pairs of successive images of the video signal S

The watermarking function applied to the selected movement vectors 20 will now be described with reference to FIGS. 2 to 6. According to a first embodiment of the invention (binary mode), for the application of the watermarking function, a reference grid is used which is represented in FIG. 2. This grid is placed in a Cartesian coordinate system (O, x, y) orthogonal and consists of blocks 40 of dimension Ko on the abscissa and H ₀ on the ordinate. In our example, Ko is equal to H ₀ and is worth 7 pixels.

We define a sub-block 42 of smaller dimensions and centered inside each block. In our example, sub-block 42 has the following dimensions:

K = H _T = 5 pixels.

Thus, inside each block of the reference grid, two zones Zo and Z _{L are} defined, the first zone Z ₀ being between the sub-block 42 and the block 40, and the zone Z being the zone inside the sub-block 42. Preferably, the dimensions Ko, H ₀ ,

and Hi are chosen so that the zones Zo and Z are substantially of the same surface. In this example, the total area of each block 40 is 49 pixels, the area of the Z area is 25 pixels and the area of the ZQ area is 24 pixels. Let K (V _X , V _y j be a motion vector selected for the application of the watermarking function using the marking key W. More precisely, the application of the watermarking function on the vector V consists in transform this so that it carries information making it possible to determine the value of one of the bits of the marking key, for example the i-th bit denoted Wj. We will denote V '{y, V' _y ) the resulting vector. The tattoo function is defined by the application of the following rules:

- if W | = 0 and if the point V with coordinates (V _x , V _y ) in (O, x, y) is located in the zone Z ₀ , then V = V;

- if W | = 0 and if the point V with coordinates (V _x , V _y ) in (O, x, y) is located in the zone Z _u then V undergoes a transformation, so that the point V with coordinates (V ' _x , V ' _y ) in (O, x, y) is located in Z ₀ (This transformation will be detailed with reference to Figure 3);

- if Wι = 1 and if the point V with coordinates (V _x , V _y ) in (O, x, y) is located in the zone Zi, then V = V;

- if Wj = 1 and if the point V with coordinates (V _x , V _y ) in (O, x, y) is located in the zone Z ₀ , then V undergoes a transformation, so that the point V with coordinates (V ' _x , V' _y ) in (O, x, y) is located in Z. (This transformation will be detailed with reference to Figures 4, 5 and 6).

FIG. 3a illustrates the case where Wj = 0 and where the point V is located in the zone Z, that is to say inside the sub-block 42 whose vertices are denoted A, B, C and D . The vertex closest to V is determined, here point A, then a weighted central symmetry of center A is applied to bring point V to V located in zone Z ₀ . Of course, the weighting parameter is calculated so that V is always in the zone Zo. For example, given the dimensions of blocks 40 and 42,

2 -rr. here we define the central symmetry weighted by the following relation: A V = - AV.

If Wj = 1 and if V is located in the zone Z ₀ we obtain V by applying a weighted axial symmetry whose axis is the side of the sub-block 42 closest to the point V, if this transformation makes it possible to obtain a point V located in zone Z. This is the case illustrated in Figures 3b and 3c. In this case, if we note H the projection of V on the axis considered, and with the dimensions chosen in our example, the weighted axial symmetry is defined by the

following relation: HP = - HV

2

In the other cases represented in FIG. 3d, that is to say when the image of the point V by an axial symmetry as described previously does not make it possible to obtain a point V in the zone Z, we apply to the point V a weighted central symmetry, the center of which is the nearest vertex A, B, C or D. In the example of figure 3d, it is

point A. The weighted central symmetry is defined here by the relation A V = - AV. This

case arises when point V is located in the vicinity of one of the vertices of block 40. It is possible to detect and extract the mark which has been inserted in a video signal according to the watermarking method described above.

As shown in FIG. 4, the method of detection and extraction of the mark is a method which comprises the application of a function of extraction of the binary marking key W consisting of:

- select the tattooed vectors (step 50); - Modify the reference partition by inserting a third zone between the first and the second zone, this third zone defining a neighborhood of the border between the first and the second zone (step 52);

- for each tattooed vector:

• determine in which zone each tattooed vector is located (step 54); • allocate to the bit of the extracted marking key with which each selected motion vector is associated, the binary value of the zone in which the tattooed vector is located if it is located in the first or second zone, outside the third zone (step 56); • allocate to the bit of the extracted marking key with which each selected motion vector is associated a weighted value between 0 and 1, if it is in the third zone, this value being dependent on the location of the motion vector within this third zone according to a predetermined function, in particular a sigmoid function (step

58);

calculating a correlation score between the initial marking key and the extracted marking key to decide on the presence or absence of the marking key in the selected vectors (step 60). More precisely, during step 50, if step 14 was implemented during watermarking of the video signal, for each motion vector 20 of the highest level 16 selected during the application of the watermarking method :

- the tattooed movement vectors associated with this movement vector are extracted;

- an average vector is calculated equal to the average of the tattooed movement vectors 26 associated with this movement vector; and

- The extraction function of the marking key described above is applied to the calculated average vector.

FIG. 5 shows a device for implementing the detection method described above. Such a device comprises for example a microcomputer 70 provided with a user interface comprising a display screen 72 and an input keyboard 74, connected to a central unit 76.

This central unit 76 comprises a conventional microprocessor and memory space allowing the implementation: of means 78 for selecting tattooed vectors;

- Means 80 for modifying the reference partition by inserting the third zone between the first and second zones of the reference partition;

means 82 for allocating, to the bit of the extracted marking key with which each selected vector is associated, the binary value of the area in which the tattooed vector is located if the latter is located in the first or second area, outside the third zone;

means 82 for allocating, to the bit of the extracted marking key with which each selected vector is associated, a weighted value between 0 and 1, if it is in the third zone, this value being dependent on the location the motion vector within this third zone according to a predetermined function, in particular a sigmoid function; means 84 for calculating a correlation score between the initial marking key and the extracted marking key to decide on the presence or absence of the marking key in the selected vectors.

The modified reference partition for the application is represented in FIG. 6. As previously, with reference to FIG. 2, this partition is placed in a Cartesian coordinate system (0, x, y) orthogonal and consists of blocks 40 of dimension Ko on the abscissa and H ₀ on the ordinate. We define a sub-block 42 of smaller dimensions K ^

H, and centered inside each block 40. Around the border between block 40 and sub-block 42, which consists of the outer edge of sub-block 42, a band Z ₂ is defined comprising all of the points located in a predetermined vicinity of this border.

There are thus defined, inside each block of the reference grid, three zones Zo, Z _\ and Z ₂ , the first zone Z ₀ being between the outer border of the strip Z ₂ and block 40, the zone Z being the zone inside the sub-block 42 not covered by the band Z ₂ , and the zone Z ₂ being constituted by the band Z ₂ itself. The points located in the vicinity of the border between block 40 and sub-block 42 are therefore excluded from zones Z ₀ and Z

Preferably, the dimensions Ko, H ₀ , Ki, W _\ and the thickness of the strip Z ₂ are chosen so that the zones Zo and Z are of the same surface, substantially equal to 45% of the total surface of the block 40 , and so that the surface of the third zone Z ₂ represents approximately 10% of the surface of the block 40.

FIG. 7 represents a reference partition modified in accordance with a second embodiment of the invention. In this embodiment, the marking key W comprises data which can take more than two discrete values, for example three. The initial reference partition, before modification, made up of blocks 40 as in FIG. 2, therefore comprises in each block 40 as many zones as values that the data of the marking key can take, that is to say three (Z ' ₀ ,

For this, one defines in a block 40 two centered sub-blocks 44 and 46 (the sub-block 46 being located inside the sub-block 44) which make it possible to define these three zones. The first zone Z ′ ₀ is between the outer border of the sub-block 44 and the block 40, the zone Z is the zone located in the sub-block 44 outside the sub-block 46, and the zone

Z ' ₂ is the area inside sub-block 46.

The modification of the reference partition, carried out during the detection of watermarking consists in this example of inserting a first band Z ' ₃ between the zones Z' ₀ and Z near their border materialized by the sub-block 44 and inserting a second strip Z ' ₄ between zones Z and Z' ₂ near their border materialized by sub-block 46.

It is easily understood that it is possible to generalize the method for applications in which the marking key comprises data with N discrete values, N being any natural integer greater than or equal to 2.

Finally, FIG. 8 shows the shape of several conventional one-dimensional sigmoid functions FF ₂ , F ₃ and F ₄ .

These functions are suitable for the first embodiment with a binary marking key. They allow you to progress more or less progressively from a value "1" to a value "0". Thus, one of them can be used to allocate to the bit of the extracted marking key with which each selected motion vector is associated a weighted value between 0 and 1, if it is in the third zone Z ₂ .

The allocated value depends on the situation of the motion vector in the zone Z ₂ . The closer it is to zone Z ₀ , the more it tends towards "0", and the closer it is to zone Z the more it tends to "1".

Of course, other progressive transition functions between the values "0" and "1" can be chosen.

These transition functions can then be easily adapted to key-key embodiments comprising data with discrete values. This detection and extraction process makes it possible to extract an estimate

W = (w,, W ₂ , W ₃ , W ₄ ) of the marking key W = (Wι, W ₂ , W ₃ , W ₄ ), from a watermarked image of the video signal.

Once this marking key has been extracted, the calculation of the correlation score makes it possible to determine a confidence threshold characterizing the presence or absence of the marking key in the video signal.

A first correlation rule to estimate whether or not this marking key is present in an image is given by the following formula:

or

- 1 4 _ 1 4 W = -∑W,, and W = -∑W,

4, = ι 4, ₌ | In order to validate the detection phase, it is possible to achieve an accumulation of the coefficients C over the entire video signal S ', if all the images of this signal are watermarked.

Alternatively, a second correlation rule for estimating the presence of the marking key in a plurality of images l ' _n of the signal S' is given by the following recurrent formula:

C n. - \ '("-1) + 1-- Q

C _n -, where d (w _n , w) is \ the Hamming distance. not

It clearly appears that a tattoo detection method according to the invention improves the robustness to attacks of existing methods. Indeed, for a video signal watermarked using a binary marking key, the modification of the reference partition during the step of extracting the marking key makes it possible to weight the bits extracted from the key marking according to the situation of the motion vectors which carried them.

Another advantage of a watermarking method as described above resides in the hierarchical approach described with reference in particular to step 14 of the process described above, spreading the watermarking over several motion vectors of a lower hierarchical level from by selecting a corresponding motion vector from a higher hierarchical level, reduces the impact of an attack, malicious or not, on the watermarking of the video signal. Finally, it should be noted that the invention is not limited to the embodiment described above.

Indeed, the digital signal can be a fixed digital image or even a digital audio signal. In the case of a digital image, some of the DCT coefficients of the image represented in the DCT domain could be chosen as attributes.

Claims

1. Method for detecting the watermarking of a digital signal (S ¹ ) watermarked by the application of a watermarking function to at least part (20) of predetermined attributes of the digital signal, the application of the watermarking function watermarking being performed using a marking key (W) comprising data with discrete values each of which data is associated with at least one selected attribute and comprising the following steps: locating the coordinates of the selected attribute (V ) in a space comprising a reference partition defining as many zones (Z ₀ , Zi;

Z ' ₀ Z'i Z' ₂ ) that of discrete values which the data of the marking key can take; assign a discrete value to each of the zones; possibly modify the coordinates of the selected attribute, so that it is located in the zone whose discrete value corresponds to the value of the data of the marking key with which the selected attribute is associated, characterized in that it includes the application of a function for extracting the marking key (W), the application of this function comprising the following steps: selecting (50) the tattooed attributes; - Modify (52) the reference partition by inserting at least one new zone (Z ₂ ; Z ' ₃ , Z' ₄ ) between the zones of the reference partition, this new zone defining a neighborhood of the borders between the zones of the reference partition; allocate (56) to the discrete value of the data of the extracted marking key with which each selected attribute is associated, the discrete value of the zone in which the selected attribute is located if it is located in one of the zones of the reference partition, outside the new zone; allocate (58) to the discrete value of the data of the extracted marking key with which each selected attribute is associated, a weighted value between two discrete values of neighboring areas, if the latter is in the new area, this weighted value being dependent on the location of the attribute within this new zone according to a predetermined function; calculating (60) a correlation score between the initial marking key and the extracted marking key to decide on the presence or absence of the marking key in the selected attributes.

2. Method for detecting the watermarking of a digital signal according to claim 1, characterized in that, the digital signal (S ¹ ) being a video signal, said predetermined attributes are its motion vectors.

3. Method for detecting watermarking of a digital signal (S ¹ ) according to claim 1 or 2, characterized in that the predetermined function is a sigmoid function.

4. Method for detecting watermarking of a video signal (S ') according to any one of claims 1 to 3, characterized in that the space is divided into several parts, in each of which at least two zones are defined (ZQ, Z,; Z ' ₀ Z Z' ₂ ), each of these zones being located inside another of these zones, with the exception of an external zone (Z ₀ ) whose external contour corresponds at the contour of the part in which it is located, each new zone (Z ₂ ; Z ' ₃ , Z' ₄ ) inserted during the extraction step surrounding an interior zoneβi

5. Method for detecting the watermarking of a digital signal (S ') according to any one of claims 1 to 4, characterized in that, the marking key being with binary values, the reference partition comprises a first area ( Z ₀ ) and a second zone (Z ^ complementary to the first, and in that, when extracting the marking key: the reference partition is modified by inserting at least a third zone (Z ₂ ) between the first and second zone, this third zone defining a neighborhood of the border between the first and the second zone; the bit of the extracted marking key with which each selected attribute is associated is allocated a weighted value between 0 and 1, if is in the third zone, this weighted value being dependent on the location of the attribute within this third zone according to a predetermined function ,.

6. Method for detecting the watermarking of a digital signal (S ') according to claims 4 and 5, characterized in that the surfaces of the first and second zones (Z ₀ , Z) are each substantially equal to 45% of the surface of the part in which they are defined, the surface of the third zone (Z) inserted during the extraction step being substantially equal to 10% of the surface of this part.

7. Method for detecting watermarking of a digital signal (S ¹ ) according to any one of claims 1 to 6, characterized in that the reference partition produced during watermarking comprises a reference grid comprising blocks (40) of predefined dimensions, each block comprising as many zones (Zo, Z) as discrete values that the data of the marking key can take.

8. A method of detecting the watermarking of a digital signal (S ') according to claim 7, characterized in that in each block (40) of the reference grid is defined at least one centered sub-block (42) inside the block, the zones of this block being defined by the spaces between the sub-blocks (42) and the block (40).

9. A method of detecting the watermarking of a digital signal (S) according to claim 8, characterized in that the blocks (40) and sub-blocks (42) of the reference grid are rectangular.

10. Method for detecting watermarking of a digital signal (S ') according to any one of claims 1 to 9, characterized in that, the process for watermarking the digital signal (S') further comprising the following steps: hierarchical calculation (14) of several successive levels (12, 16) of attributes, the attributes of a given level each being associated with several attributes of the level directly below; selecting (18) at least some of the attributes belonging to the highest level (16); applying (22) the watermarking function to each selected attribute leading to the calculation of a parameter for modifying this attribute; application of the modification parameter of the selected attribute to the attributes of a lower level associated with this attribute, for each attribute (20) of the highest level (16) selected during tattooing: the tattooed attributes (26) associated are extracted to this attribute; a mean attribute is calculated equal to the mean of the watermarked attributes (26) associated with this attribute; and the function of extracting the marking key is applied to the calculated average attribute.

11. Method for detecting watermarking of a digital signal (S ') according to claim 8, characterized in that the attributes of a given level (16) are each equal to the average of the attributes of the level directly below (12) with which they are associated.

12. Device (70) for detecting the watermarking of a digital signal (S '), for implementing a method according to any one of claims 1 to 11, characterized in that it comprises: means (78) selecting the tattooed attributes; means (80) for modifying the reference partition by inserting at least one new zone (Z ₂ ) between the zones of the partition reference, this new zone defining a neighborhood of the borders between the zones of the reference partition; means (82) for allocating, to the discrete value of the data of the extracted marking key with which each selected attribute is associated, the discrete value of the zone in which the tattooed attribute is located if the latter is located in one of the zones of the reference partition, outside the new zone; means (82) of allocating, to the discrete value of the data of the extracted marking key with which each selected attribute is associated, a weighted value between two discrete values of neighboring zones, if this is found in the new zone, this weighted value being dependent on the location of the attribute within this new zone according to a predetermined function; means (84) for calculating a correlation score between the initial marking key and the extracted marking key to decide on the presence or absence of the marking key in the selected attributes.