FR2795585A1 - Determination method for a partition in order to insert a watermark into a digital image - Google Patents

Abstract

The method involves modulating coefficients (X) of the regions (R) representing a digital image (I). The partitioning into regions is effected by an adaptive partitioning as a function of a criterion of detectability of an information bit inserted on each region. The method involves generation of a centered pseudo-random sequence equal in size to the cardinal number (N) of the set of coefficients, formed from centered pseudo-random sub-sequences. The set of coefficients are modulated by the centered pseudo-random sequence in order to insert the same information bit on the set of coefficients. Finally the method involves checking, over each region, of a criterion of detectability of the information bit inserted by modulation. Independent claims are included for a method of decoding a watermarking signal in a digital image, for a partition determination device and for a decoding device.

Description

The present invention relates to a method for determining a partition of a digital image for inserting a marking signal and a method of inserting an associated marking signal.

It concerns correlatively a device for determining a partition of a digital image for inserting a marking signal and a device for inserting a marking signal.

The present invention is generally in the technical field of marking (English watermarking) or tattooing of digital images, and more particularly still images.

The marking of digital data makes it possible to protect such data for example by associating them with copyright information.

In its general principle, the marking consists of inserting an indelible mark in digital data, assimilated to the encoding of additional information in the data.

Decoding this additional information verifies the copyright information that has been inserted.

This inserted mark must therefore be both imperceptible, robust to certain distortions applied to the digital image and reliable detection.

Conventionally, a standard technique for inserting a marking signal in a digital image consists in using a linear modulation model in which at least one subset of coefficients representative of the digital image is modulated according to this linear model. using a weighting coefficient.

Noting X = {X ;, 1 <_ i <_ N} a set of representative coefficients at least a portion of a digital image and w = {wj, 1s j <_ P} a size marking signal P <_N, a pseudo-random signal of known distribution and of zero mean, the linear insertion formula is X'j = Xj + bajwj, with 1 <_j <_P, where {Xj, 15 j _ <P} is a subset of the set of coefficients X, b is an information bit, and aj is a weighting coefficient, also called modulation amplitude.

The detection of the marking signal then consists in detecting whether the pseudo-random sequence w has been inserted or not in a set of coefficients. This detection is done without using the original image and can be based on a standardized statistical test that calculates a correct detection probability.

Such an insertion technique makes it possible, by inserting a marking signal, to insert a single bit of information since the response of the detector is binary (yes / no).

To insert a larger number of information bits into the digital image, particularly when a Q-bit code is desired indicating, for example, the name or address of the owner or author of the image, It is necessary to repeat the previously described insertion method as many times as there are information bits to be inserted. Typically, to insert a binary signal, we use either b = 1 or b = -1.

In other words, we must choose Q subsets of coefficients and operate the modulation of these subsets by choosing Q marking signals.

Subsets of distinct coefficients are preferably chosen so that the modulations are not superimposed on each other, which could disturb the detection or cause annoying visual effects. It is therefore a question of choosing a partition of the coefficients representative of the digital image in Q distinct subsets, each carrying an information bit.

There are known methods of arbitrary partitioning of the set of coefficients, in blocks, independently of the content of the digital image. However, the content of the images is not spatially homogeneous. Indeed, the response of the detector is dependent on the image signal itself. The larger the variance of the representative coefficients of the image, the lower the response of the detector.

In addition, for a given modulation amplitude, the greater the length of the sequence of the inserted marking signal, and therefore the larger the region and hence the number of scalable coefficients, the more reliable the detection.

In conclusion, for a given modulation amplitude and a given image, there is a minimum length of the marking signal necessary to ensure a given probability of detection, this minimum length also depending on the local content of the image signal in the region in which is inserted. the marking signal.

The object of the present invention is to propose a method for determining a partition that makes it possible to take account of the image signal itself and is inexpensive from the computational point of view.

The present invention thus provides a method for determining a partition in distinct regions of a set of representative coefficients of at least a portion of an image for inserting a marking signal.

This method comprises the following steps: generating a pseudo-random sequence centered equal in size to the cardinal of said set of coefficients, consisting of centered pseudorandom sub-sequences; modulating said set of coefficients by said pseudo-random sequence centered to insert a same bit of information on said set of coefficients; and - validating a detectability criterion on each region of the information bit inserted by modulation.

It is thus possible to partition a set of coefficients into distinct regions by performing adaptive partitioning according to a detectability criterion of an information bit inserted on each region.

Such adaptive partitioning has the advantage of adapting to the content of the image signal and of ensuring the detectability of the marking signal inserted on the different regions determined by this partition, unlike the choice of an arbitrary partitioning.

In practice, for each region of the partition, a criterion of detectability of an inserted information bit is validated by modulation of the coefficients of said region. This criterion can consist of a comparison of a result of a standardized statistical test with a threshold value corresponding to a probability of detection fixed in advance.

With the method according to the invention, it is possible to generate a pseudo-random sequence centered on the meeting of smaller sized pseudo-random centered sequences and to modulate all the available coefficients to carry out the insertion of a marking signal. .

By modulating all the coefficients with the same information bit and a centered pseudo-random sequence which is decomposed into pseudo-random subsequences also centered on each region of the partition made on the set of coefficients, the generation steps a pseudo-random and modulation sequence do not need to be repeated at each new region tested.

According to a preferred characteristic of the invention, the method for determining a partition comprises a single step of applying a distortion to said set of modulated coefficients, before the step of validating a detectability criterion.

It is known to apply to a set of modulated coefficients a predetermined distortion and to carry out the validation of the detectability criterion after this pre-distortion, so that it is possible to guarantee the detection of the marking signal inserted on a partition coefficients for a predetermined distortion.

Thanks to the invention, a pre-distortion can be applied to all the modulated coefficients, in a unique way, and does not need to be repeated with each new region tested in the score.

In a particularly practical embodiment of the invention, the set of coefficients is a set of spectral coefficients resulting from a spectral transformation of a digital image, and the step of applying a distortion comprises the sub-elements. following steps - inverse spectral transformation of said set of modulated coefficients to obtain a marked image <B>; </ B> - application of a distortion to said marked image; and spectral transformation of the marked image to obtain a set of modulated coefficients after distortion.

The gain obtained in terms of computational complexity is all the more important in this case since the application of a distortion requires several processing steps of all the coefficients, and in particular the application of an inverse spectral transformation. to obtain the marked image and to apply a distortion, such as a compression I decompression to it, and then again, the application of a direct spectral transformation of the image in order to verify the criterion of detectability in the domain spectral.

According to another particularly practical feature of the invention, the centered pseudo-random sequence is formed of pseudo-random subsequences of identical size.

Such a pseudo-random sequence is particularly well suited to the application of an iterative upward or downward partitioning of the quaternary tree (quadtree) type which makes it possible to iteratively merge regions of identical sizes or to divide iteratively between regions in sub-regions of identical size.

Preferably, the pseudo-random subsequences have a size greater than or equal to a minimum number of coefficients forming a region of said statistically significant size partition to validate said detectability criterion on said region.

It is thus possible theoretically to operate a maximum partition of the set of coefficients, with the same single pseudo-random centered sequence, each pseudo-random subsequence being centered on the smaller statistically acceptable regions for insertion of a modulation marking signal in the set of coefficients.

According to a practical embodiment, which is inexpensive in terms of calculations, at the generation stage, each pseudo-random centered subsequence is created by generating half of the pseudo-random samples of said subsequence by a generator of pseudorandom numbers and symmetrizing said generated pseudo-random samples.

Preferably, in this embodiment, in the generation step, each pseudo-random centered subsequence is created by further performing on said pseudo-random samples a permutation using a secret key.

This permutation operation makes it possible to better preserve in a subsequence created by symmetrization of pseudorandom samples its very nature of random sequence and not to weaken the imperceptibility of the inserted marking signal.

According to another practical embodiment of the invention, in the generation step, each pseudo-random centered subsequence is created by generating pseudo-random samples by a pseudorandom number generator and redistributing on each sample the sum of said pseudo-random samples generated.

According to a preferred feature of the invention, an iterative fusion partition is performed from an initial division of said set of coefficients into regions of statistically significant size to validate said detectability criterion on said regions and, for each region of said partition for which said detectibility criterion is not validated, said region is fused with another region of said partition. This iterative merging partitioning mode is particularly simple to cover and makes it possible to directly obtain a partition of the set of coefficients in a number of distinct regions which corresponds to the maximum of information bits of the marking signal that the it is possible to insert in the image, for a given detection reliability and possibly a given robustness to certain distortions. This number of regions thus determined is called the practical capacity of the digital image.

Alternatively, according to another preferred feature of the invention, a division is performed by iterative division of said set of coefficients and, for each region of said partition, a partition is validated in subregions of said region if and only if it exists. at least one subregion for which said detectability criterion is validated.

This iterative division partitioning mode is also relatively simple to cover.

According to another aspect of the invention, a method of inserting into a digital image a marking signal comprising one or more information bits, comprises a method of determining a partition according to the first aspect of the invention. , and a step of inserting the information bit or bits by modulating the coefficients of one or more regions of said partition.

This insertion method has advantages similar to those of the method for determining a partition that it covers, and in particular makes it possible to implement the insertion of a marking signal in an adaptive and reliable manner vis-à-vis of its detection.

Correlatively, the present invention also relates to a device for determining a partition in distinct regions of a set of representative coefficients of at least a portion of an image for inserting a marking signal.

According to the invention, this device for determining a partition comprises: means for generating a centered pseudo-random sequence equal in size to the cardinal of said set of coefficients, formed of pseudo-random centered subsequences; means for modulating said set of coefficients by said pseudo-random sequence centered to insert a same bit of information on said set of coefficients; and means for validating a detectability criterion on each region of the information bit inserted by modulation.

The present invention also relates to a device for inserting into a digital image a marking signal comprising one or more information bits, comprising a device for determining a partition according to the invention, and insertion means. the information bit or bits by modulating the coefficients of one or more regions of said partition.

The present invention also relates to a computer, a digital image processing apparatus, a digital printer, a digital camera and a digital camera adapted to implement the partition determining method and the insertion method according to the present invention. to the invention and / or comprising a partition determining device or an insertion device according to the invention.

These partitioning and inserting devices, this computer, this digital image processing apparatus, this digital printer, this digital camera and this digital camera have features and advantages similar to those described with reference to FIGS. methods of partitioning and inserting that they implement.

The present invention also relates to a storage medium or information carrier, readable by a computer or a microprocessor, incorporated or not to the latter, optionally removable, comprising portions of software codes or program instructions adapted to implement the steps of the method of determining a partition or insertion according to the invention, when said storage means or information carrier is implemented by a microprocessor or a computer. Other features and advantages of the invention will become apparent in the description below of various embodiments of the invention.

In the accompanying drawings, given by way of nonlimiting examples - FIG. 1 is a general diagram illustrating a device for inserting a marking signal in a digital image. FIG. 2 schematically illustrates FIG. a spectral decomposition step implemented during the insertion of a marking signal; FIG. 3 is a block diagram illustrating a device adapted to implement the method of determining a partition and / or insertion according to the invention.

FIG. 4 illustrates the application of the method for determining a partition of a digital image in accordance with the invention; FIG. 5 is an algorithm of the method for determining a partition according to a first embodiment of the invention; and FIG. 6 is an algorithm of the method for determining a partition according to a second embodiment of the invention.

First of all, with reference to FIG. 1, the insertion of a marking signal into a set of coefficients representative of a digital image I will be described first.

This digital image I can be represented by a set of coefficients, either in the spatial domain or in a transformed domain, the coefficients being in the latter case hybridized, that is to say located both in the spatial domain and in the space domain. in the frequency domain. Such a representation of the image is for example obtained by using a subband resulting from a spatio-frequency decomposition of the image, for example a discrete wavelet decomposition.

Here, the insertion of the marking signal is carried out by a technique of insertion by spread spectrum, by modulation of coefficients of a space-frequency representation of the image, obtained by a spatio-frequency transformation of the digital image. I: It is possible to use, by way of example, a discrete wavelet decomposition S shown schematically in FIG. 2. This spatio-frequency decomposition is well known in the field of image processing and only the principle below is recalled. It makes it possible to split the image into frequency sub-bands and to obtain hybrid coefficients, that is to say spectral coefficients also located in space, here in the plane of the image.

Image I consists of a series of digital samples. The image I is for example represented by a series of bytes, each byte value representing a pixel of the image I, which can be a black and white image, with 256 gray levels.

The multi-resolution spectral decomposition means consist of a subband decomposition circuit, or analysis circuit, formed of a set of analysis filters, respectively associated with decimators by two. This decomposition circuit filters the image signal I along two directions, in sub-bands of low frequencies and high spatial frequencies. The circuit comprises several successive analysis blocks for decomposing the image I into sub-bands according to several resolution levels.

By way of example, the image I is decomposed here into sub-bands at a decomposition level d equal to 3.

A first analysis block receives the image signal I and the filter through two digital filters respectively low-pass and high-pass, in a first direction, for example horizontal. After passing through decimators in pairs, the resulting filtered signals are in turn filtered by two filters, low-pass and high-pass respectively, in a second direction, for example vertical. Each signal is again passed in a decimator by two. At the output of this first analysis block, four sub-bands LLI, LH ,, HL, and HH, of the highest resolution in the decomposition, are thus obtained.

The sub-band LL comprises the low frequency components according to the two directions of the image signal I. The sub-band LH comprises the low frequency components in a first direction and the high frequency in a second direction of the image signal I. The sub-band HL comprises the high frequency components in the first direction and the low frequency components in the second direction. Finally, subband HH comprises the high frequency components in both directions.

A second analysis block in turn filters the low-frequency sub-band LL, to similarly provide four intermediate resolution level LL2, LH2, HL2 and HH2 sub-bands in the decomposition. A third analysis block then filters the LL2 low frequency subband to provide four lowest resolution LL3, LH3, HL3 and HH3 subbands in this decomposition.

This gives 10 subbands and three levels of resolution. Of course, the number of resolution levels, and therefore of subbands, can be chosen differently.

Of course, other types of spectral transformation could be used such as discrete Fourier transform, discrete cosine transform or Fourier-Mellin transform.

Frequency sub-bands forming a set of spectral coefficients in which a marking signal can be inserted are generally obtained.

It is thus possible to consider here the high frequency sub-band of higher resolution HH, to achieve the insertion of a marking signal. This sub-band HH thus provides a set of coefficients X, of size for example equal to N.

Let X = {Xi, <B> 1: 9 </ B> i <N} be the set of representative coefficients of digital image I.

Of course, one could consider a subset of coefficients of this set X, representative only of a portion of the digital image I.

To insert a marking signal in this set of coefficients X, a technique consists in inserting a pseudo-random signal, by spreading its spectrum, in order to make this signal undetectable by spectral or statistical analysis. Let w = {wj, <B> 1 </ B> <_ j <_ P) be a marking signal of size P <_N, a pseudo-random signal of known distribution and of zero average. The most common distributions are the binary distribution {-1, 1), the uniform distribution on [-1, 1] and a normalized Gaussian distribution centered N (0,1).

The linear insertion formula is X) = Xj + bajwj, with 1 <_j <_P, where {Xj, <B> 1 </ B> <_ j <_ P) is a subset of the set coefficients X, b is an information bit, and aj is a weighting coefficient, also called modulation amplitude.

The set S = {j, 1: g j <_ P) is also called the insertion medium of an information bit.

To insert a signal that can be formed of several bits of information, it is useful to partition the set of coefficients to determine the number of regions or the number of insertion media available in this set.

For this, the insertion device 10 according to the invention comprises a device for determining a partition 11 and insertion means 12.

The device for determining a partition 11 is adapted to produce an adaptive partition of the set of coefficients as a function of the detectability of an inserted marking signal.

This adaptive partitioning method consists in varying the number of modulated coefficients to insert an information bit of the marking signal, in order to simultaneously satisfy the fixed invisibility and detection probability criteria.

Such adaptive partitioning according to a detectability criterion is described in particular in the patent application FR 99 04462 cited with reference in the present description. In principle, this patent application FR 99 04462 describes a method for determining a partition in which the size of the insertion medium is varied for each bit of information until a correct detectability criterion is satisfied.

The device for determining a partition 11 according to the present invention comprises for this purpose means for validating a detectability criterion on each region of an information bit, inserted by means of a pseudo-random sequence of zero average by modulating the coefficients of the region.

It also comprises means for generating a pseudo-random sequence centered equal in size to the cardinal of the set of coefficients representative of an image, here equal to N. As will be described in more detail below, this sequence pseudo-random is formed of pseudo-random sub-sequences centered. The operation of these generation means will be described with reference to the method for determining a partition according to the invention.

Means for modulating the set of coefficients X by the pseudo-random sequence centered to insert a same bit of information on this set of coefficients are also incorporated in the device for determining a partition 11.

The latter may also include means for applying a distortion D which makes it possible to predict and. guarantee the robustness of the marker signal inserted for certain distortions, typically a compression of the digital image for its storage.

These means for applying a distortion make it possible to actually simulate on the set of modulated coefficients a predetermined distortion. They comprise here, since the coefficients of the set X are spectral coefficients, inverse spectral transformation means of the set of modulated coefficients to obtain a marked image, means of applying a distortion on this image. labeled, and means of spectral transformation of the marked image to find all the modulated coefficients after distortion.

In this example, the inverse spectral transformation means are adapted to implement a multi-resolution spectral recomposition in wavelets and the direct spectral transformation means are adapted again to repeat the wavelet decomposition of the digital image I. The device In a first embodiment, a partition 11 also comprises iterative fusion partitioning means, based on an initial division of the set of coefficients X.

Preferably, this initial division makes it possible to cut the set of coefficients X into a set of regions of statistically significant size to validate the detectability criterion on these regions. This initial division will be described later with reference to the method of determining a partition.

The iterative merging partitioning means are adapted, for each region of the partition for which the detectability criterion is not validated, to merge the region with another region of the partition.

It is possible to use an iterative quaternary tree fusion technique which is relatively simple to implement.

Of course, the partitioning means could also be adapted to achieve this partition of the image by iterative division of the digital image according to a second embodiment of the invention.

Several methods for achieving this partition of a set of coefficients based on a detectability criterion are described in detail in the patent application FR 99 04462 cited herein by reference.

This partition determination device 11 makes it possible to obtain, at the output, a partition in regions of the set of coefficients X, an information bit that can be inserted on each of these regions with a reliable probability of detection and a robustness to a given distortion. The maximum number of regions that can be obtained by this adaptive partitioning of the set X corresponds to the practical capacity of insertion of this set X.

As illustrated in FIG. 1, an insertion device 10, incorporating the partition determination device 11 previously described, further comprises reading means 13 for a marking signal to be inserted, containing several bits of information. . By way of non-limiting example, a binary marking signal may be 1000111 representing for example the coding of the name of the author of the image to be marked. The insertion means 12 themselves are conventional and make it possible to insert by modulation the different information bits of the marking signal on the various insertion supports determined by the partition determination device 11.

All of this insertion device 10, and in particular the device for determining a partition 11 that it incorporates, can be integrated in a computer 10, or a camera or a digital camera.

Such a computer 10 is illustrated in FIG.

All of the aforementioned means of the insertion device 10 and in particular the partition determining device 11 according to the invention are incorporated in a microprocessor 100, a read only memory 102 (in English Read Only Memory or ROM) being adapted to memorize a program for determining a partition and insertion of a modulation marking signal, and a random access memory 103 (in English Random Access Memory or RAM) including registers adapted to store modified variables during the first time. execution of the program.

Of course, the device for determining a partition 11 and the insertion means 12 associated with the reading means 13 could be incorporated in separate computers, similar to that illustrated in FIG.

The microprocessor 100 integrated with the computer 10 can be connected to different peripherals, for example, a digital camera 107 or a microphone 111 via an input / output card 106 to receive and store digital documents to mark.

The digital camera 107 makes it possible, in particular, to provide images to be authenticated by insertion of a marking signal.

This computer 10 comprises a communication interface 112 connected to a communication network 113 to optionally receive images to be marked.

The computer 10 further comprises document storage means, such as a hard disk 108, or is adapted to cooperate by means of a floppy disk drive 109 with removable document storage means such as floppy disks 110. These fixed or removable storage means may further comprise the code of the insertion method according to the invention, which, once read by the microprocessor 100, will be stored in the hard disk 108.

As a variant, the program enabling the insertion device to implement the invention can be stored in the read-only memory 102.

In the second variant, the program can be received to be stored as previously described via the communication network 113.

The computer 10 also has a screen 104 for example to interface with an operator using the keyboard 114 or other means.

The central unit 100 will execute the instructions relating to the implementation of the invention. When powering up, the programs and methods relating to the invention stored in a non-volatile memory, for example the read-only memory 102, are transferred into the random access memory 103 which will then contain the executable code of the invention as well as the variables necessary for the implementation of the invention.

This random access memory 103 comprises different registers for storing the variables necessary for the execution of the program, and in particular an X.sub.i register, X.sub.j for storing the coefficients of the regions which are modulated or demodulated temporarily in order to check the detectability criterion and a register w. to store the generated pseudo-random sequence to determine the partition of the set X.

A communication bus 101 allows communication between the different sub-elements of the computer 10 or linked to it. The representation of the bus 101 is not limiting and in particular the microprocessor 100 is able to communicate instructions to any sub-element directly or via another sub-element.

We will describe a method of determining a partition and insertion according to the invention, firstly with reference to FIG. 4. The method for inserting a marking signal is implemented on an image. digital I.

This insertion method comprises first of all a method for determining a partition of a set of coefficients representative of the digital image I.

This partition is made according to a detectability criterion of an information bit inserted on each region of the partition.

Note that in practice, one can search for a partition of a digital image in a maximum of regions satisfying a detectability criterion in order to know the practical capacity of this image with respect to the insertion of a marking signal. This maximum capacity of the digital image corresponds to the number of information bits that can be inserted into the digital image while respecting the detectability criterion.

It is also possible to search for a given marking signal, comprising a number of bits of given information smaller than the maximum capacity of the image I, the best possible partition of the image I which makes it possible to insert this marking signal of given length, for a given probability of detection, and in regions of the largest possible sizes in order to best ensure the invisibility of the inserted marking signal.

The method for determining a partition firstly comprises a spectral decomposition step E10 of the type described previously in Wavelet Transform (DWT).

A set of spectral coefficients X representative of the digital image I is thus obtained. These spectral coefficients are subdivided into frequency sub-bands as illustrated schematically in FIG. 4, at the end of the decomposition step E10. The size of the set of coefficients X is here equal to N = M × M for a square image. Of course, the same method could apply to a rectangular image.

In this embodiment of the invention, the method comprises a generation step E11 of a pseudo-random centered sequence of size N equal to the cardinal N of the set of coefficients X representative of the image I. Let w = {w ,,,, 0 <_ m, n <_M} this pseudo-random sequence.

This pseudo-random sequence w is formed of pseudo-random sub-sequences centered wk.

The pseudo-random sequence can be seen as the union of pseudo-random subsequences wk.

Here, the pseudo-random sequence centered w is formed of pseudo-random sub-sequences wk of identical size which correspond to a block, called unit block, of the set of coefficients X.

Of course, the pseudo-random subsequences wk could have different sizes from each other.

Here, the size of these unit blocks corresponds to the minimum number of coefficients adapted to form a region of statistically significant size to validate the detectability criterion on this region.

As described below, when a standardized statistical test of detection is applied to a set of coefficients, a minimum number of coefficients is necessary for this test to be significant.

The pseudo-random subsequences wk have a size A = L x L greater than or equal to this minimum number of coefficients.

Preferably, pseudo-random sub-sequences wk of size equal to this minimum size are chosen. This characteristic is particularly advantageous when performing an iterative merge partition by merging blocks from an initial division of the image into unit blocks of size L x L. The initial division then corresponds to the maximum partition that the it is possible to make an image, beyond which the generated coefficient blocks are no longer statistically significant for the insertion and detection of an information bit.

For example, L = 8 is chosen.

The pseudo-random sequence w is then the union of N / 64 pseudo-random subsequences wk forming unit blocks.

On each unit block, a pseudo-random predefined distribution signal (Gaussian, uniform, ...) centered on a function dependent on a secret key K and on the index k must be generated in accordance with the invention. of each unit block. This key can be denoted f (K, k). It is then necessary to ensure that in practice the average of the signal wk is exactly equal to zero on each unit block.

A first method for generating such a pseudo-random centered sequence consists in centering the sequence in a deterministic manner, by symmetrization. This method is adapted to generate a pseudo-random subsequence of size A pair.

In this case, each pseudo-random centered subsequence wk is created by generating half A / 2 of the pseudo-random samples of the subsequence by a pseudo-random number generator of known centered distribution, and by symmetrizing the samples. pseudo-random thus generated to obtain the other half A / 2 of the samples of the subsequence wk.

For a one-dimensional sequence, we thus generate a set of samples of size A / 2 {wj, 1: 9 j <_A / 2} and we symmetry this set to obtain the other half of the samples {wj = -wA_j, A / 2 <j <A}.

In order to reinforce the randomness of the sequence thus obtained, it is possible to carry out on the set of samples obtained {wj, 1 <_j <-A} a permutation using a secret key.

A second method of generating a pseudorandom-centered sequence consists in creating each pseudo-random centered subsequence wk by generating pseudo-random samples by a pseudorandom number generator and redistributing the sum of these samples on each sample. pseudo-random samples generated.

Thus, using a secret key f (K, k), we first generate a sequence of pseudo-random samples of the same size A as the unit block, ie {w ;, 1 <_i <A}.

Puis, on redistribue cette somme Sk sur les échantillons afin d'obtenir une séquence pseudo-aléatoire centrée, c'est-à-dire dont la somme est exactement égale à zéro. The exact sum of these samples is then calculated

Then, we redistribute this sum Sk on the samples to obtain a pseudo-random sequence centered, that is to say whose sum is exactly equal to zero.

We thus obtain wk = fw; - SkIA, 1 <i <4}.

The pseudo-random sequence w thus consists of the union of sub-sequences wk centered so that the partitioning of a set of coefficients modulated by the pseudo-random sequence centered w provides subsets of coefficients modulated also by pseudo-random sub-sequences centered.

Of course, other techniques can be used to generate pseudo-random sequences centered.

As illustrated in FIG. 4, a modulation step E12 of the set of coefficients by the pseudo-random sequence centered w is then implemented so as to insert an information bit on the set of coefficients resulting from the decomposition. spectral.

In practice, the set of coefficients is copied into a working memory, for example the random access memory 103 in FIG. 3, so as not to directly mark the image I.

The same information bit is modulated on the set of coefficients, for example here b = 1, according to the modulation formula X '= X; + a; w;, with 0 <_i <_N.

The weighting amplitude a; is chosen in the usual way to guarantee the invisibility of the inserted information bit.

A constant weighting coefficient can be used for all coefficients, such as a; = av for all i, the value of the weighting coefficient av being equal to a maximum value ensuring the imperceptibility of the marking signal for the modulation of all the coefficients representative of the digital image.

It may be wise to exploit the fact that the image signal itself makes it possible to mask the modulation.

For this purpose, for each coefficient X, it is possible to use; to modulate, a weighting coefficient according to a law of the form a; = k; .av, where k; is a modulation factor dependent on the coefficients located in a neighborhood of the considered coefficient and #v is equal to the maximum value of the weighting coefficient ensuring the imperceptibility of the marking signal for the modulation of the set of coefficients.

Thus, each coefficient will be modulated according to the local content of the image, which makes it possible to locally increase the amplitude of the modulation in favor of detectability.

Preferably, in this embodiment of the invention, it is desired to perform a partitioning on a post-processed image signal in order to obtain a priori robustness of the inserted marking signal to the possible post-treatments suffered by the image I.

For this, in a distortion step E13, a distortion is applied to all the modulated coefficients.

Thanks to the one-step modulation of the coefficients, from a pseudo-random centered sequence, the distortion step E13 can be implemented only once during the determination of a partition, contrary to the techniques in which each coefficient subregion was modulated independently of each other and required the application of a distortion at each rank of the partition.

This is all the more advantageous since here, the application step E13 of a distortion comprises the following sub-steps as illustrated in FIG. 4 - inverse spectral transformation E13a of the set of modulated coefficients X; to obtain a marked image <B>; </ B> - application <B> El 3b </ B> itself of a distortion on the marked image and - spectral transformation E13c of the marked image to obtain a set of modulated coefficients after distortion.

Suitable distortions are distortions that alter the value of the pixels without changing the geometry of the image, such as the addition of noise, contrast modification, compression, low-pass or high-pass filtering. We consider here a compression of the image according to the JPEG standard (in English Joint Photographic Expert Group) of quality factor Q specified, with typically q = 75 by default for JPEG compression.

In practice here, a DWT- 'inverse wavelet transformation is applied, then a compression followed by decompression of the image, and finally again a direct wavelet transformation to be able to estimate the detectability of the information bit inserted into the image. spectral domain.

Iterative partitioning of the set of coefficients is then applied to find a partition of the set of coefficients in which each bit of information inserted can be correctly detected according to a fixed probability of detection. The number of regions thus determined during the partition corresponds to the practical capacity of insertion of the image I.

Various iterative division / merge segmentation techniques to which quaternary block partitioning (quadtree) belongs can be used. In general, we distinguish so-called bottom-up methods and top-down methods depending on whether the iterative means of partitioning is merger or division.

Firstly, with reference to FIG. 5, a first embodiment of the invention will be described in which the partitioning is performed by iterative merging from an initial division of the set of coefficients into size regions. statistically significant to validate a detectability criterion for these regions.

Following the transformation steps E10, generation E11, modulation E12 and distortion E13 as described above, in a step E14, a first frequency sub-band is considered as a set of coefficients to be partitioned.

The initial division of this set of coefficients corresponds to a spatial division of the image I into unit size blocks as defined above. By construction itself, the pseudo-random modulation sequence is centered on each of these unit blocks.

In a quaternary tree fusion, four blocks of equal size are always grouped together, called "child blocks" to form a "father block" and it is iteratively repeated this fusion, hence the notion of recursion.

Of course, the partition is subband by subband frequency signal from the spectral decomposition.

Thus, in a step E15, each unit block is considered in a scanning order of the subband, for example from left to right and from bottom to top.

A validity criterion is applied to a first unit block that has not already been processed.

A statistical test of detection on this unit block of modulated coefficients X'i is applied in a calculation step E16.

An example of a standardized statistical test is given below.

Let Xi, 1 <_ i _ <P be the set of coefficients to be tested after insertion of the pseudo-random sequence w and application of a pre-distortion.

où M représente une valeur moyenne de corrélation entre les valeurs de chaque coefficient X; et de modulation wi, et V la variance de ces valeurs. Soit en pratique

Le calcul de cette valeur de test T requiert la connaissance de la séquence de modulation pseudo-aléatoire w qui est générée une fois pour toute à l'étape de génération E11 précédemment décrite. The test consists of calculating for this set of size P

where M represents an average value of correlation between the values of each coefficient X; and modulation wi, and V the variance of these values. In practice

The calculation of this test value T requires knowledge of the pseudo-random modulation sequence w which is generated once and for all at the generation step E11 previously described.

The detectability is given by the comparison of this test value T with a certain threshold PP, which corresponds theoretically to a probability of correct detection of p% if the test T follows a Gaussian law of variance equal to 1 for the two hypotheses tested ( presence or absence of marking signal).

A comparison step E17 makes it possible to compare this test value T with the threshold value Pp or correct detection rate.

At the end of this comparison, if the calculated value T is greater than the correct detection rate Pp, the tested block is kept in memory in a storage step E19 as an available region for inserting an information bit.

In practice, a capacitance calculation step E20 can be implemented so that at each tested coefficient block that is suitable for performing the insertion of an information bit, the value of the capacity of the data bit is increased. a unit.

On the other hand, if the calculated value T is smaller than the correct detection rate Pp, a higher rank block in the partition is selected in a merge step E18, that is to say here a parent block corresponding to the merger of four unit blocks and the calculation steps E16 of the value of the test T and comparison E17 are repeated with the correct detection rate Pp on this new block.

During this merging of child blocks to form a father block, son blocks are preferably chosen for which the detectability criterion is not validated.

For the application of this detectability criterion, it is possible to use any standardized statistical test for detection such as that described in the article entitled "A method for signature casting on digital images" by I. PITAS, in Proc. ICIP, pages 215-218, September 1996, which allows to characterize detection in terms of probability.

It is possible to choose a threshold value Pp corresponding to a fixed detection probability ratio, for example 99.95%.

Thus, for a given region defining a set of coefficients X of size P and with a fixed weighting coefficient a, it is possible to calculate a detectability amplitude and to compare it with a threshold value PP in order to validate or not the criterion of detectability in this region.

Advantageously, reference may be made to patent application FR 9904462 in the name of the Applicant for details relating to the variation of this test value T as a function of the size of the region of coefficients tested.

In general, the probability of detection decreases with the size of the regions tested, so that by merging regions to increase the size of the block of coefficients used for the insertion of a marking signal, the value of the tag is increased. T test on this new block.

Thus, at the end of these adaptive partitioning steps E14 to E18, it is possible to determine an insertion capacity of the image, equal to Q bits, and an adequate partition of the image corresponding to this Q capacitor.

A marking signal can then be formed, of length equal to Q bits, and then inserted, bit by bit, in the different regions of the partition.

Of course, the actual insertion step E30 of the marking signal could be coupled to the determination of the partition, so that each region available at the end of the detectability test E16, E17, can be really marked by inserting an information bit by modulating the coefficients of this region.

The insertion step E30 makes it possible to actually insert an information bit of the marking signal, by choosing a value of coefficient b equal to 1 or -1 in the previously described modulation formula, the alternation of the bits + I-1 translating the content of the message inserted.

It is possible to use a pseudo-random sequence centered and of known distribution, different for each region of the partition, and initialized by a secret key K ', as illustrated in FIG.

Note that this secret key K may possibly be identical to the secret key K used during the generation of the pseudo-random sequence w for the determination of the partition.

In addition, one could generate a pseudo-random sequence centered w identical to that generated in the generation step E11. A partition of an image I is shown schematically by way of example in FIG. 4, at the end of steps E14 to E18, in which, for example, 25 distinct regions are obtained making it possible to insert a 25-bit marking signal.

Of course, any other partitioning technique could be used.

A second embodiment of the invention will be described with reference to FIG. 6, in which the partitioning is performed by iterative division of the set of coefficients.

As a result of the transformation steps E10, generation E11, E12 modulation and E13 distortion as described above, in a step E21 is considered a first frequency sub-band as a set of coefficients to be partitioned.

We consider here a quaternary tree division in which we always divide a block, called "father block" into four blocks of equal size, called "son blocks".

Of course, the partition is subband by subband frequency signal from the spectral decomposition.

Thus, in a step E22, an initial block corresponding to a first frequency sub-band is considered.

This initial block is stored in a list L of blocks to be processed.

Next, in a step E23, a first block B is considered to be treated from the list of blocks to be treated L.

A detectability criterion is applied to this block B in a validation step E24.

In practice, as previously described, in a calculation step E24a, the value T of a statistical test of detection on this block of modulated coefficients is calculated. This step is analogous to the calculation step E16 described with reference to FIG.

The detectability is then given by the comparison of this test value T with a certain threshold PP, which corresponds theoretically to a probability of correct detection of p% if the test T follows a Gaussian law of variance equal to 1 for the two hypotheses tested. (presence or absence of marking signal).

A comparison step E24b makes it possible to compare this test value T with the threshold value Pp or correct detection rate. This comparison step E24b is analogous to the comparison step E17 described with reference to FIG.

After this comparison, if the calculated value T is less than the correct detection rate Pp, it is checked in a test step E25 if there are blocks to be treated in the list L.

If not, the recursive partitioning algorithm is complete. Otherwise, we consider the next block B of the list L in step E23 and the validation step E24 is repeated.

If at the end of the comparison step E24b, the calculated value of the test T is greater than the correct detection rate Pp, the detectability is tested on each child block of this parent block B in a validation step E26.

In practice, the parent block B is divided into four child blocks and on each of these blocks, one verifies the detectability criterion according to a step similar to the previously described validation step E24.

An E27 test verifies whether the detectability criterion is valid on at least one child block.

If so, a division step E28 is implemented to effectively divide the block B into four child blocks and an adding step E29 allows to add these child blocks to the list of blocks to be treated L.

The steps E23 to E29 of test and validation are then repeated on each of the blocks of the list L until it is exhausted.

On the other hand, if at the end of the E27 test, the detectability criterion is not checked on any of the child blocks of the parent block B, the stored block B is stored in a storage step E19 as a region available to insert a bit of information.

In practice, a capacity update step E20 can be implemented as previously described so that for each tested block of coefficients that is suitable for carrying out the insertion of an information bit, the value value is increased. the capacity of a unit.

Thus, at the end of these adaptive partitioning steps E21 to E29, it is possible to determine an insertion capacity of the image, equal to Q bits, and an adequate partition of the image corresponding to this Q capacitor.

A marking signal can then be formed, of length equal to Q bits, and then inserted, bit by bit, in the different regions of the partition.

Of course, as before, the actual insertion step E30 of the marking signal can be coupled to the determination of the partition, so that each block B available at the end of the detectability test E27 can be marked actually by inserting an information bit by modulating the coefficients of this block B.

The insertion step E30 is performed in the same way as when the partition is performed by iterative merger.

Moreover, in an improved embodiment of this iterative division partition technique, a child block division of a parent block B may be retained at the end of the detectability test E27 only if it represents a gain of capacity, that is to say in a quaternary tree division, only if the detectability criterion is validated on at least two son blocks. Otherwise, it may be preferable to keep the larger father block to ensure better invisibility of the inserted marker signal.

Of course, the partitioning techniques described above are in no way limiting.

Thus, one can possibly combine two partitioning techniques by division and fusion. This is the case when, for example, an intermediate initial partition is chosen, between the two extremes, which are on the one hand a single region and, on the other hand, a partition into regions of smaller admissible size, allowing both the operations division and merger.

The method of determining a partition can also be implemented to maximize the capacity of the digital image by using several different segmentation techniques, and then to choose the partition that provides the largest Q capacity for the digital image.

It is also possible to determine the best possible partition, that is to say providing a set of blocks of coefficients of the largest possible size, to insert a message of fixed length, less than the maximum capacity of the digital image to be marked.

In any case, the present invention makes it possible to generate and insert once a pseudo-random sequence on a set of available coefficients to achieve the insertion, and corollary to apply only one pre-distortion to the set of coefficients.

It is thus possible to quickly obtain a partition of a set of coefficients which makes it possible to insure the insertion of a marking signal with a fixed correct detection probability, for a whole range of distortion lying between "no distortion" and a "predetermined distortion" applied when determining the score. In addition, the simulation of maximum distortion before insertion is a quasi-deterministic operation that provides an advantage over a theoretical theoretical calculation.

Finally, the modulated coefficients for the insertion of a marking signal can be coefficients resulting from a spatial representation of the image or from another spatio-frequency representation than that obtained by the spectral decomposition in wavelets described above, and in particular a representation of the image in which the coefficients of the same frequency are not necessarily grouped together.

Claims (21)

A method for determining a partition in distinct regions of a set of representative coefficients of at least a part of an image (I) for inserting a marking signal, characterized in that it comprises the following steps - generating (E11) a centered pseudo-random sequence (w) of size equal to the cardinal (N) of said set of coefficients, consisting of pseudo-random centered subsequences (wk); modulating (E12) said set of coefficients by said pseudo-random centered sequence (w) to insert a same information bit on said set of coefficients; and - validating (E16, E17, E24, E26) a detectability criterion on each region of the information bit inserted by modulation. 2. Method for determining a partition according to claim 1, characterized in that it comprises a single step of applying a distortion (E13) to said set of modulated coefficients, before the validation step (E16, E17, E24, E26) of a detectability criterion. 3. Method for determining a partition according to claim 2, characterized in that said set of coefficients is a set of spectral coefficients (X) resulting from a spectral transformation of a digital image, and in that said step application (E13) of a distortion comprises the following sub-steps - inverse spectral transformation <B> (El </ B> 3a) of said set of modulated coefficients to obtain a marked image; - application <B> (El 3b) </ B> of a distortion on said marked image; and spectral transformation (E13c) of the marked image to obtain a set of modulated coefficients after distortion. 4. Method for determining a partition according to one of claims 1 to 3, characterized in that the pseudo-random sequence centered (w) is formed of pseudo-random subsequences (wk) of identical size. 5. Method for determining a partition according to one of claims 1 to 4, characterized in that the pseudo-random subsequences (wk) have a size greater than or equal to a minimum number of coefficients forming a region of said statistically significant size partition to validate said detectability criterion on said region. 6. Method for determining a partition according to one of claims 1 to 5, characterized in that in the generation step (E11), each pseudo-random centered sub-sequence (wk) is created by generating half of the pseudo-random samples of said subsequence by a pseudo-random number generator and by symmetrizing said generated pseudo-random samples. A method of determining a partition according to claim 6, characterized in that in the generating step (E11), each pseudo-random centered subsequence (wk) is created by further performing on said samples. pseudo-random a permutation using a secret key. 8. Method for determining a partition according to one of claims 1 to 5, characterized in that in the generation step (E11), each pseudo-random centered subsequence (wk) is created by generating pseudo-random samples by a pseudorandom number generator and redistributing on each sample the sum of said pseudo-random samples generated. 9. A method for determining a partition according to one of claims 1 to 8, characterized in that it performs an iterative fusion partition from an initial division of said set of coefficients into regions of statistically significant size for validating said detectability criterion on said regions and in that, for each region of said partition for which said detectability criterion is not validated, said region is fused with another region of said partition. 10. A method of determining a partition according to one of claims 1 to 8, characterized in that it performs a partition by iterative division of said set of coefficients and in that, for each region of said partition, it is valid a sub-region partition of said region if and only if there is at least one sub-region for which said detectability criterion is validated. 11. A method of insertion in a digital image of a marking signal comprising one or more information bits, characterized in that it comprises a method for determining a partition according to one of claims 1 to 10. , and an insertion step (E30) of the information bit or bits by modulating the coefficients of one or more regions of said partition. 12. Device for determining a partition in distinct regions of a set of coefficients representative of at least a part of an image (I) for inserting a marking signal, characterized in that it comprises: generating (100, 102, 103) a centered pseudo-random sequence (w) of size equal to the cardinal (N) of said set of coefficients, consisting of centered pseudo-random subsequences (wk); modulating means (100, 102, 103) of said set of coefficients by said centered pseudo-random sequence (w) for inserting a same information bit on said set of coefficients; and means for validating (100, 102, 103) a detectability criterion on each region of the information bit inserted by modulation. 13. Device for determining a partition according to claim 12, characterized in that it comprises means for applying (100, 102, 103) a distortion to said set of modulated coefficients. 14. Device for determining a partition according to claim 13, characterized in that said set of coefficients is a set of spectral coefficients (X;) resulting from a spectral transformation of a digital image, and in that said means for applying a distortion comprise - means of inverse spectral transformation (100, 102, 103) of said set of modulated coefficients to obtain a marked image; means for applying (100, 102, 103) a distortion on said marked image; and spectral transformation means (100, 102, 103) of the marked image to obtain a set of modulated coefficients after distortion. 15. Device for determining a partition according to one of claims 12 to 14, characterized in that the pseudo-random sequence centered (w) is formed of pseudo-random subsequences (wk) of identical size. 16. Apparatus for determining a partition according to one of claims 12 to 15, characterized in that the pseudo-random subsequences (wk) have a size greater than or equal to a minimum number of coefficients forming a region of said statistically significant size partition to validate said detectability criterion on said region. 17. Apparatus for determining a partition according to one of claims 12 to 16, characterized in that the generating means (100, 102, 103) are adapted to create each pseudo-random sub-sequence centered by generating half pseudo-random samples of said subsequence by a pseudorandom number generator and by symmetrizing said generated pseudo-random samples. 18. Device for determining a partition according to claim 17, characterized in that the generating means (100, 102, 103) are adapted to create each pseudo-random centered sub-sequence by performing further on said pseudo samples. - random a permutation using a secret key. 19. Apparatus for determining a partition according to one of claims 12 to 16, characterized in that the generating means (100, 102, 103) are adapted to create each pseudo-random sub-sequence centered generating pseudo-random samples by a pseudorandom number generator and redistributing on each sample the sum of said pseudo-random samples generated. 20. Apparatus for determining a partition according to one of claims 12 to 19, characterized in that it comprises means of partitioning (100, 102, 103) by iterative fusion, from an initial division of said a set of coefficients in regions of statistically significant size to validate said detectability criterion on said regions, the iterative fusion partitioning means (100, 102, 103) being adapted, for each region of said partition for which said detectability criterion n ' is not validated, to merge said region with another region of said partition. 21. Apparatus for determining a partition according to one of claims 12 to 19, characterized in that it comprises partitioning means (100, 102, 103) by iterative division of said set of coefficients and in that, for each region of said partition, the validation means (100, 102, 103) are adapted to validate a partition in subregions of said region if and only if there exists at least one sub-region for which said criterion of detectability is validated 22.- Device for determining a partition according to one of claims 12 to 21, characterized in that it is incorporated in a microprocessor (100), a read-only memory (102) being adapted to memorize a program for determining a partition for inserting a modulation marking signal, and a random access memory (103) comprising registers adapted to store modified variables during the execution of said program. 23. Device for insertion in a digital image of a marking signal comprising one or more information bits, characterized in that it comprises a device for determining a partition (11) according to one of the claims. 12 to 22, and insertion means (12) of the information bit or bits by modulating the coefficients of one or more regions of said partition. 24. An insertion device according to claim 23, characterized in that it is incorporated in a microprocessor (100), a read-only memory (102) being adapted to store a program for inserting a modulation marking signal. , and a random access memory (103) comprising registers adapted to store modified variables during the execution of said program. 25. Computer, characterized in that it comprises means adapted to implement a method of determining a partition according to one of claims 1 to 10. 26. Computer, characterized in that it comprises means adapted to implement the method of insertion of a marking signal according to claim 11. 27. Computer, characterized in that it incorporates a partition determining device according to one of claims 12 to 22. Computer, characterized in that it incorporates a device for inserting a marking signal according to one of claims 23 or 24. 29. Apparatus for processing a digital image, characterized in that it comprises means adapted to implement a method of determining a partition according to one of claims 1 to 10. 30. Apparatus for processing a digital image, characterized in that it comprises adapted means at mettr it implements the method of insertion of a marking signal according to claim 11. 31. Apparatus for processing a digital image, characterized in that it incorporates a device for determining a partition according to the one of claims 12 to 22. 32. Apparatus for processing a digital image, characterized in that it incorporates a device for inserting a marking signal according to one of claims 23 or 24. 33. Printer digital device, characterized in that it comprises means adapted to implement a method for determining a partition according to one of claims 1 to 10. 34. A digital printer, characterized in that it comprises adapted means to implement the method of insertion of a marking signal according to claim 11. 35. Digital printer, characterized in that it incorporates a device for determining a partition in accordance with one of the 36. Digital printer, characterized in that it incorporates a device for inserting a marking signal according to one of claims 23 or 24. 37. A digital photographic camera, characterized in that it comprises means adapted to implement a method of determining a partition according to one of claims 1 to 10. 38. Digital photographic camera, characterized in that it comprises means adapted to implement the method insertion of a marking signal according to claim 11. 39. Digital camera, characterized in that it incorporates a partition determining device according to one of claims 12 to 22. 40. Apparatus digital photographic device, characterized in that it incorporates a device for inserting a marking signal according to one of claims 23 or 24. 41. Digital camera, characterized in that it comprises means adapted to implement a method of determining a partition according to one of claims 1 to 10. 42. A digital camera, characterized in that it comprises means adapted to implement the Method for inserting a marking signal according to Claim 11. 43. Digital camera, characterized in that it incorporates a partition determining device according to one of Claims 12 to 22. 44. Camera digital device, characterized in that it incorporates a device for inserting a marking signal according to one of claims 23 or 24.