FR2786583A1 - Procedure for insertion of a secret authentication mark, such as a digital watermark, into a digital data signal, where the signal is broken down into a collection of coefficients and a control bit is determined - Google Patents

Abstract

Procedure involves a calculation step (E5,E6) for at least one coefficient (Xij,Iij), of a collection of coefficients that represent a digital signal broken down into coefficients, in which a control bit (Cij) is determined as a function of the first (L-M) levels of the said coefficient, calculation (E7,E8) of the last levels (M) of the control bit and of the supplementary data value.

Description

The present invention relates to a method of inserting information

  additional such as a secret mark in a digital signal. It also relates to a method for authenticating a signal

  digital from a secret mark inserted in the signal.

  Correlatively, the present invention relates to a device for inserting additional information and a device for authenticating a digital signal respectively adapted to implement the methods.

  insertion and authentication according to the invention.

  The present invention relates generally to the authentication of digital data, and more particularly of digital images. Its purpose is to allow the detection of changes

  intervened in original digital data.

  More specifically, the authentication method according to the invention is in the technical field of marking (watermarking in English) of digital data which can be interpreted as the insertion of a seal in the digital data, allowing authenticate the content of a file

original digital data.

  Conventionally, and as described for example in European patent application EP 0 766 468 filed in the name of NEC CORPORATION, the marking techniques can be used for the protection of copyright on a digital document. In such a case, the mark inserted must, among other things, be robust to the various manipulations carried out on digital data, and in particular to the techniques

  conventional digital data compression.

  Unlike digital data marking techniques aimed at protecting the author of a digital document, the marking techniques to authenticate the data themselves must make it possible to detect any change or manipulation of digital data. The mark inserted for the authentication of original digital data must therefore not be robust to the various conventional manipulations of the image. In addition to its fragility with regard to the modifications undergone by the document, this inserted mark must be must be imperceptible, in order to preserve the quality of the digital document, and in particular the visual quality when the document to be authenticated is a digital image, and be difficult to counterfeit in order to prevent a counterfeiter from reinserting the same mark into data

  that have been changed.

  Such a digital data authentication method by inserting a secret mark in the data is described in the article entitled "Digital watermarking using multiresolution wavelet decomposition" (in French digital marking using multiresolution decomposition in wavelets) by D. KUNDUR and D. HATZINAKOS, Proc. ICASSP, pages 2969 2972, May 1998. This method thus makes it possible to insert additional information, which may contain, for example, information on the source of the data or its creation date, and to sign this data.

original.

  However, in this article, the described insertion and authentication method requires the spectral decomposition in wavelets of a digital image to be authenticated then the spectral recomposition in wavelets

  reverse of the digital image after insertion of the secret mark.

  In addition to the fact that this spectral decomposition is costly in computation time, the method described makes it possible to authenticate only an original digital image but does not make it possible to protect digital data

  already compressed by various coding methods.

  The present invention aims in particular to propose a method of inserting additional information for the authentication of digital data which makes it possible to protect digital data in

  various areas of spatial or spectral representation.

  In particular, it allows additional information to be inserted into digital data which is compressed to be stored or

  transmitted, compatible with the compression used.

  The insertion method is applied to a digital signal broken down into a set of coefficients, each coefficient being represented on L bits, and makes it possible to insert additional information represented by a set of binary values of the same size as the set of coefficients

  representative of the digital signal to be authenticated.

  According to the invention, the insertion method is characterized in that it comprises the following steps: - calculation, for at least one coefficient, of a control bit according to a predetermined operation as a function of L - M foregrounds bits of said coefficient; - Calculation of M last bit plans according to a predetermined reversible rule as a function of the control bit and of a binary value of the additional information; and - substitution of the M last bit plans of said coefficient by the M

last calculated bit plans.

  Thus, the method of inserting a secret mark in accordance with the invention applies to any digital signal as soon as it is represented by a

  set of spatial or spectral coefficients.

  The insertion process can thus be applied in particular directly

  on a compressed representation of a digital signal.

  The additional information is incorporated directly into the digital signal, by modification only of the last bit planes of the coefficients, that is to say of the least significant bits LSB (in English Least

  Significant Bits) coefficients so that it remains imperceptible.

  Furthermore, the control bit calculated from the coefficient itself only depends on the first bit planes, that is to say the most significant bit plans of the coefficient. These L - M first bit plans are not modified during the substitution step, which has the advantage during the authentication of the digital signal, of being able to recalculate for each coefficient a control bit identical to that calculated during the insertion of additional information and to facilitate the detection of possible modifications

  made to the original digital signal.

  According to a symmetrical aspect of the invention, a method of authenticating a digital signal, on the basis of additional information, such as a secret mark, inserted into the digital signal by the insertion method as described below. above, is characterized in that it comprises the following steps: - calculation, for at least one coefficient, of a control bit according to said predetermined operation as a function of L - M first bit planes of said coefficient; - Extraction of the value of the additional information inserted according to said predetermined reversible rule as a function of the control bit and of M last bit plans; - comparison of the value extracted from the additional information inserted and the binary value of the additional information; and decision whether or not to authenticate the digital signal as a function of the identity or not of said extracted value and said binary value of

  additional information inserted.

  As described above, the authentication method is all the more simple to implement when the control bit, calculated from unmodified bit plans of the coefficients, is identical to that calculated during

  the step of calculating the insertion process.

  In the extraction step, it is therefore possible to use a deterministic rule which is simple to reverse in order to extract the binary value of the additional information inserted. In addition, according to the spatial or spectral representation of the digital signal by the coefficients, the authentication method makes it possible to locate in the representation space the modifications made to the digital signal. Furthermore, as before, the authentication method makes it possible to authenticate a compressed representation of a digital signal and is therefore intrinsically robust compared to the compression algorithm used. According to a preferred characteristic of the invention, in the step of calculating a control bit of the insertion method and of the authentication method, said predetermined operation is a function of the L - M foregrounds of

  bits and a confidential key represented on L bits.

  Although optional, the use of a confidential key makes it safer and more difficult to counterfeit the insertion of information

  additional in the digital signal.

  Advantageously, said predetermined operation consists in performing a sum of binary operations on the L - M first bit plans of said coefficient and the L - M first bit plans of said confidential key and

  calculating the control bit as a function of the parity of said sum.

  The step of calculating a control bit is thus implemented

  simple and fast way on the coefficients representative of the digital signal.

  According to a preferred characteristic of the invention, which makes it possible to make imperceptible at best the additional information inserted, the number

  M of substituted bit plans is 1 or 2.

  According to another preferred characteristic of the invention, in the step of calculating the last M bit planes of the insertion method, the value of the last bit plan is equal to the value of the control bit or to the alternative value of the

  control bit according to the state of the binary value of the additional information.

  Correlatively, at the extraction step of the authentication method, the value of the additional information is equal to a state or another state depending on whether or not the value of the last bit plane and the value of the bit

control.

  The calculation rule used during insertion is particularly simple to implement from the value of the control bit and the value

  additional information to insert.

  It also has the advantage of being simple to reverse for the authentication of the digital signal in order to find, from the value of the control bit and the value of the last bit plane, the binary value of

  the information that is supposed to have been inserted.

  According to an advantageous characteristic of the insertion method, the number of substituted bit plans is equal to 2 and, in the step of calculating the M last bit plans, the value of the penultimate bit plan is equal to the

  alternative value of the last bit plane.

  Correlatively, at the extraction step of the authentication process, the number of bit plans is equal to 2 and the value of the additional information is extracted only if the value of the penultimate bit plan

  is equal to the alternative value of the last bit plane.

  This characteristic is of interest during the authentication of the digital signal since the probability that a modified coefficient nevertheless makes it possible to extract a correct value from the information inserted is lower since the last two bit planes must have been modified while

retaining alternate values.

  According to a preferred embodiment of the invention, when the digital signal is broken down into a set of quantized spectral coefficients, the insertion method further comprises a prior step of choosing a subset of quantized spectral coefficients, in which the coefficients chosen have at least a magnitude strictly greater than zero, the steps of calculating a control bit, of calculating the M last bit plans

  and of substitution being implemented for each coefficient of said sub-

together.

  The restriction of the coefficients liable to be modified to a sub-

  set of non-zero coefficients avoids modifying the coefficients set to zero when quantizing the digital signal. We don't lose

  advantages of quantizing the digital signal for its compression.

  Advantageously, the subset of coefficients comprises the quantized spectral coefficients whose magnitude is greater than a threshold value. The threshold value can thus be fixed so that the coefficients capable of being modified have a magnitude high enough for the substitution of the last bit planes to be insignificant compared to the

coefficient value.

  In this preferred embodiment of the invention, the set of binary values representative of the additional information is advantageously generated by the repetition of an initial binary information of

  size smaller than the size of the set of quantized spectral coefficients.

  This arrangement increases the probability that all the initial binary information is inserted into the digital signal even if

  some coefficients are not subject to change.

  On the contrary, if the additional information has a resolution identical to that of the digital signal to be authenticated, certain binary values are not inserted when they correspond to spectral coefficients

  quantified magnitudes lower than the predetermined threshold value.

  Correlatively, the authentication process in this mode of

  preferred embodiment, furthermore comprises a prior step of choosing a sub-

  set of quantized spectral coefficients, in which the chosen coefficients have a magnitude greater than a threshold value, the steps of calculating a control, extraction, comparison and decision bit being implemented

  works for each coefficient of said subset.

  This characteristic makes it possible to implement the authentication method only on the set of coefficients likely to have been

  changed when inserting additional information.

  According to an advantageous characteristic of the invention, which allows a practical implementation of the present invention for the authentication of a digital signal stored in a compressed file, the insertion method further comprises a prior step of suitable entropy decoding extracting the quantized spectral coefficients and an entropy coding step

  after said substitution step.

  Likewise, the authentication method for authenticating a digital signal stored in a compressed file also comprises a prior step of entropy decoding adapted to extract the spectral coefficients

quantified.

  Correlatively, the present invention also relates to a device for inserting additional information such as a secret mark for authenticating a digital signal broken down into a set of coefficients, each coefficient being represented on L bits, and said information additional being represented by a set of binary values of the same size as said set of coefficients, characterized in that it comprises: - means for calculating a control bit according to a predetermined operation as a function of L - M first planes of bits of a coefficient; means for calculating M last bit plans according to a predetermined reversible rule as a function of the control bit and of a binary value of the additional information; and - means for substituting the M last bit plans of said

  coefficient by the M last calculated bit plans.

  Likewise, the present invention relates to a device for authenticating a digital signal from additional information such as a secret mark inserted into said digital signal by the insertion method according to the invention, said digital signal being broken down into a set of coefficients, each coefficient being represented on L bits, characterized in that it comprises: - means for calculating a control bit according to said predetermined operation as a function of L - M first bit plans d 'a coefficient; means for extracting the value of the additional information inserted according to said predetermined reversible rule as a function of the control bit and of M last bit plans of said coefficient; - means for comparing the value extracted from the additional information inserted and the binary value of the additional information; and - means for deciding whether or not to authenticate the digital signal as a function of the identity or not of said extracted value and of said value

  binary of additional information inserted.

  These insertion and authentication devices have preferential characteristics and advantages similar to those described.

  previously for the insertion process and the authentication process.

  The present invention also relates to a computer, an apparatus for processing a digital signal, a digital photographic apparatus or a digital camera comprising means suitable for carrying out the insertion method and / or means adapted for carrying out the authentication process. It also relates to a computer, a digital signal processing apparatus, a digital still camera or a digital camera comprising an insertion device and / or a device

  authentication according to the invention.

  The present invention also relates to a means of storing information, readable by a computer or by a microprocessor, integrated or not in an insertion or authentication device, possibly removable, which stores a program implementing the method of insertion or

  authentication according to the invention.

  The characteristics and advantages of this computer, digital signal processing device, digital still camera, camera

  digital or storage medium are identical to those set out above.

  Other features and advantages of the invention will become apparent

  again in the description below.

  In the accompanying drawings, given by way of nonlimiting examples: - Figure 1 is a block diagram illustrating an insertion device according to the invention; - Figure 2 is a block diagram illustrating an authentication device according to the invention; - Figure 3 is a block diagram illustrating an insertion device used to insert additional information into a compressed digital data file; - Figure 4 is a block diagram illustrating an authentication device used to authenticate a compressed file; FIG. 5 is a block diagram illustrating a computer suitable for implementing the insertion and / or authentication method in accordance with the invention; - Figure 6 is an algorithm of the insertion method according to a first embodiment of the invention; - Figure 7 is an algorithm of the authentication method according to the first embodiment; - Figure 8 is a diagram illustrating a spectral representation of an image to be authenticated; - Figure 9 is a diagram illustrating a step of the insertion method according to the first embodiment; - Figure 10 is a diagram illustrating a step of the insertion method according to a second embodiment of the invention; FIG. 11 is an algorithm of the insertion method according to the second embodiment of the invention; - Figure 12 is an algorithm of the authentication method according to the second embodiment; - Figure 13 is an algorithm of the insertion method according to a third embodiment of the invention; and - Figure 14 is an algorithm of the authentication method

  according to the third embodiment.

  We will first describe a device for inserting information

  additional in digital data with reference to Figure 1.

  In this example, and by way of nonlimiting example, the digital signal is a digital image I consisting of a series of digital samples. An original image I can be represented by a series of pixels coded for example on 8 bits or bytes. The black and white image I can thus be decomposed in the spatial domain into a set of coefficients on 256 gray levels, each coefficient value representing a pixel of the image I. Conventionally in the digital data processing domain , The image I can also be represented in a spectral domain, after spectral decomposition of the digital image, by a set of spectral coefficients, possibly quantified when the image I must be stored or transmitted in a compressed data file. These spectral coefficients are also represented on a number L of bits, by

  example L = 8 if the coefficient is represented on a byte.

  In the following description relating to the insertion devices and

  authentication, the spatial or spectral coefficients are indifferently noted Xj, with i and j varying on the dimensions of the image broken down according to

two perpendicular directions.

  The additional information to be inserted to authenticate the image I is a secret mark W. This secret mark may contain information on the origin of the digital data (identification of the author or owner of the image) or on their date of creation. It is also represented by a set of binary values wuj, equal for example to 0 or

  1, of the same size as the set of coefficients Xij.

  In general, an insertion device can be likened to an encoder 20 which encodes in an image 1, in compressed or uncompressed format, a

  mark W. An image marked 1 ′ is provided at the output of the encoder 20.

  According to the invention, the insertion device comprises: - means 21 for calculating a control bit Cij according to a predetermined operation as a function of L - M first bit plans of a coefficient Xij; - Means 22 for calculating M last bit planes according to a predetermined reversible rule as a function of the control bit Cij and of a binary value wij of the additional information W; and - means 23 for replacing the M last bit plans of the coefficient Xq by the M last calculated bit plans. The M least significant bit plans of the coefficient Xij are thus modified as a function of the other L - M bit plans of the coefficient Xij and of

the information to insert wij.

  Preferably, the number of modified bit planes is equal to 1 or 2 so as to limit the modification of the coefficients to very low bit plans and guarantee the imperceptibility of the mark inserted in the image. We

  preserves the visual quality of the document.

  The image I ′ can undergo various distortions (compression and decompression, filtering, digital / analog transformation to be viewed, etc.) so that an image J is transmitted at the input of an authentication device which can be assimilated in part to a decoder 30 as shown in FIG. 2. The purpose of the authentication device is to indicate whether

  image J is indeed image 1 'which has been transmitted.

  For this, the authentication device according to the invention comprises: - means 31 for calculating a control bit Cij according to the same predetermined operation, used for coding, as a function of L - M first bit plans of a coefficient X5j; means of extraction 32 of the value w'ij of the additional information inserted according to the predetermined reversible rule as a function of the control bit Cij and of M last bit plans of the coefficient Xij; The decoder 30 thus makes it possible to find the additional information w'1j inserted in the noisy image J from the last 1 or 2

bit plans of the coefficient Xij.

  The authentication device also includes: - comparison means 33 of the value extracted w'ij from the additional information inserted and the binary value w1j of the additional information W; and - decision means 34 to authenticate or not the digital signal J as a function of the identity or not of the extracted value w'1j and the binary value w of the additional information inserted W. The calculation means 21 and 31 of a control bit Cj for each coefficient Xj are identical in the insertion device and in the authentication device. Preferably, the predetermined operation implemented by these calculation means 21, 31 is a function of the L - M first bit planes of the

  coefficient Xj and a confidential key K also represented on L bits.

  The use of a confidential key is useful for securing the information entered. For example, you can have a key per image I to authenticate, or a key for a set of images 1, or a unique key for each type of device that creates or stores images. , for example a

digital camera.

  In a preferred embodiment of the invention, which is particularly advantageous when it is desired to authenticate images which are compressed to be stored in a compressed file, the insertion device as illustrated in FIG. 3 also comprises means of entropy decoding 25 adapted to extract the quantized spectral coefficients Xij. In particular, the insertion device can thus process a file compressed according to the JPEG standard (in English Joint Photographic Expert Group), in which the digital signal is decomposed by a spectral transformation, then quantified and finally coded by an entropy coding method. such as coding

  arithmetic or Huffman coding.

  The digital signal being decomposed into a set of quantized spectral coefficients at the output of the entropy decoding means 25, selection means 26 of a subset of quantized spectral coefficients are adapted to choose coefficients Xij having at least one

  magnitude strictly greater than zero.

  We can set a strictly positive threshold value and choose only quantized spectral coefficients Xij whose magnitude is greater than this

threshold value.

  In this embodiment, the insertion device also comprises entropy coding means 27 which make it possible to compress

  again the digital file including the secret mark. -

  As illustrated in FIG. 4, and in a completely similar manner, the authentication device also includes decoding means 35, similar to the entropy decoding means 25, adapted to extract from a

  compressed file of quantized spectral coefficients.

  Choice means 36, identical to choice means 26 of the associated insertion device, make it possible to find the subset of

  Xij coefficients into which the additional information has been inserted.

  The insertion device and the authentication device can be incorporated into a microprocessor 100 of a computer 10 as illustrated in the

figure 5. A read only memory 102 or ROM (in English Read Only Memory)

  includes a program for inserting additional information into a digital signal and a program for authenticating the digital signal from the inserted additional information, and a random access memory or RAM (in English Random Access Memory) comprises registers adapted to record variables changed during program execution. In particular, the random access memory comprises a register for recording all the coefficients Xij representative of the digital signal such as an image I to be authenticated, a register for recording the control bit Cij and a

  register to save the M last modified bit plans.

  The microprocessor 100 integrated into the computer 10 can be connected to various peripherals, for example, a digital camera 107 or a microphone 111 via an input / output card 106 in order to

  receive and store documents (images, audio signal, ...).

  This computer 10 includes a communication interface 112 connected to the communication network 113 also suitable for transferring

  documents to be authenticated on the computer 10.

  The computer 10 further comprises document storage means, such as a hard disk 108. It can also be adapted to cooperate by means of a floppy drive 109 for example with document storage means

  removable such as floppy disks 110.

  These fixed or removable storage means may also include the code of the insertion or authentication method according to the invention, which, once read by the microprocessor 100, will be stored in the

hard drive 108.

  As a variant, the program allowing the insertion or authentication device to implement the invention may be

stored in ROM 102.

  In the second variant, the program can be received to be stored as described above via the network of

communication 113.

  The computer 10 also has a screen 104 allowing for example to serve as an interface with an operator using the keyboard 114 or a mouse or any other means. The operator can thus modify certain parameters of the insertion or authentication device, such as for example the value

  threshold used to choose a subset of coefficients.

  The central unit 100 will execute the instructions relating to the implementation of the invention. During power-up, the programs and methods relating to the invention stored in a non-volatile memory, for example read-only memory 102, are transferred to 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.

  The 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 capable of communicating instructions to any sub-element directly

  or through another sub-element.

  In practice, the insertion device and the authentication device can be incorporated in the same computer 10 or else in two separate computers 10 connected for example by the communication network 113. In general, the insertion device and the authentication device are incorporated together or separately in a digital signal processing apparatus, such as a digital still camera or a digital camera which create and store a digital image, under

compressed form.

  We will now describe a first embodiment of the insertion method and the associated authentication method with reference to FIGS. 6 to 9. In this example, we consider that the digital signal is an image I stored in a compressed file. This image I was compressed using a compression algorithm based on the JPEG standard as commonly implemented in various image processing devices, in particular in a digital camera. The image is thus decomposed in a spectral domain by a transformation into a discrete cosine DCT (in English Discrete Cosine Transform), then quantified for example by a scalar quantification technique. All of the

  quantized spectral coefficients, noted Xij in the remainder of this description, is

  then coded by an entropy coding.

  The insertion method according to the invention firstly comprises a prior step of entropy decoding E1 adapted to extract the

  quantized spectral coefficients Xij.

  We thus obtain a set of coefficients Xij structured as illustrated in FIG. 8 for an image I of dimension N1 on N2 with N1 and N2 divisible by 8. The image I is thus decomposed into a set of blocks of 64 coefficients Xij, a first coefficient being the DC continuous coefficient (low frequency coefficient in the DCT spectral decomposition) followed by 63 others

  AC coefficients (high frequency coefficients).

  In this example, it is assumed that each coefficient Xij is represented on 16 bits and that there is a confidential key K also defined on 16 bits. The secret mark is in this example a binary image, representing a logo for example, generated by the repetition of an initial binary information of size smaller than the size N1 on N2 of the set of

  quantized spectral coefficients Xij.

  The initial binary information can thus have a resolution 8 times smaller than the image I to be processed. Each bit of the initial binary information is repeated 8 times in each direction so as to have a binary image to be inserted W of the same size as the image I to be processed and such that each block 8 of 8 of

  the binary image W contains the same bit of information.

  The binary image W is thus formed of a set of coefficients wij in number equal to that of the quantized spectral coefficients Xij representing

the image I to process.

  The advantage of using such a binary image W for the insertion of the

  secret mark will be better understood in the following description of the process

of insertion.

  First of all, in a reading step E2, the first coefficient Xij is considered in a predetermined scanning order of the set of

  coefficients Xj, where i varies between 1 and N1 and j between 1 and N2.

  In a choice step E3, the magnitude of the coefficient is compared to a threshold value T. If the magnitude of the coefficient is less than this threshold value T, this coefficient Xj is not modified and the following coefficient is read directly in

a step E4.

  The following steps of the insertion process are thus only implemented for the coefficients Xij whose magnitude is greater than this threshold value T. At a minimum, only the coefficients Xj which have a magnitude strictly greater than zero will be chosen. so as not to modify the spectral coefficients which are set to zero by the quantification process used during the compression of the image I. This precaution makes it possible to keep the zero coefficients of the quantized spectral decomposition of the image and thus to not not significantly increase the size of the compressed file by inserting a

additional information.

  Preferably, a strictly positive threshold value T is chosen, for example equal to 3, so that the modification of the weakest bit plans is

  unimportant given the magnitude of the coefficient.

  In addition, the selected threshold value must make it possible to find, during the authentication of image 1, all the coefficients which could have been modified by the insertion of the secret mark. This threshold value T must therefore be chosen so that all the coefficients, even after modification of the last M bit plans, always have a magnitude strictly greater than this

threshold value.

  In general, this threshold value T meets the rule: T = k.2M - 1 o M is the number of bit planes to be modified in the coefficients

  Xi and k any strictly positive integer.

  The value chosen in this example, equal to 3, is perfectly suitable

  for the modification of one or two last bit plans.

  It can also be decreed that all the low frequency DC coefficients will be chosen for the insertion of the additional information,

regardless of their magnitude.

  It is thus understood that at least one coefficient Xij of each block 8 of 8 of the DCT spectral decomposition of the image is modified so that the binary information contained in each block 8 of 8 of the binary image W to be inserted can be inserted. Thus, thanks to the construction of a binary image W as described above, containing the same information bit in each block 8 of 8, it is ensured that this information bit will be inserted at least

once in frame I to process.

  According to the invention, for each chosen coefficient Xij, a control bit Cij is calculated in the calculation steps E5 and E6, according to a predetermined operation as a function of L - M first bit plans of the coefficient Xij, and in this example, also as a function of the confidential key K. This predetermined operation consists for example in carrying out in a first step E5 a sum S of binary operations on the L - M first bit plans of the coefficient Xij and the L - M first bit plans of the confidential key K and to calculate in a second step E6 the control bit Cij as a function of your parity of this sum S. In this example of embodiment, a number M equal to 1 of least significant bits is chosen to modify so that the first 15 shots are used

  of bits of the coefficient Xj to calculate the control bit Cij.

  With reference to FIG. 9, and noting Xkj the bit plans of the coefficient Xij, with k varying from 0 to 15 from the weakest bit plane to the strongest bit plane, we calculate for example the bitwise sum between the first 15 bit plans of Xij and the first 15 bit plans of K. We denote by S the result: S = EX, XORKk k = l In this example, the operator XOR is the exclusive binary "OR" operation (the result is equal to 0 if the bits are different and to 1 if the bits are identical). Of course, other conventional operators could be used, such as for example the "AND" operator (the result is equal to 1 if the two bits are equal to 1 and to 0 otherwise) or the non-exclusive operator "OR" (the result is equal to 0 if the two bits are equal to 0 and to 1 otherwise). Any binary operation

  reproducible at the decoder can be used.

  The control bit Cij is then calculated as being the parity of the sum S: Cij = 0 if S is even and Cij = 1 if S is odd. We can write

also Cj = S (modulo) 2.

  The last bit plane X jj is then calculated in a calculation and substitution step E7 according to a predetermined reversible rule as a function of the control bit C1j and the binary value wij of the additional information W, having the same location in the binary image representation matrix W as the coefficient Xij in the spectral representation matrix of image I and the least significant bit of the coefficient Xi is replaced by the last calculated bit plane X jj. In this example, the value of the last bit plane.XX j is equal to the value of the control bit Cij or to the alternative value 1 - Cij of the control bit depending on the state of the binary value wij of the additional information. W. We test for example in a step E8 if the binary value wij is

  equal to 0. If so, Xjj = Cij.

  Otherwise, i.e. if w1j is equal to 1, X jj = 1 - Cij.

  Any reversible rule, then making it possible to know the value of wij from the values of the last bit plane X ej and the control bit Cj can

be used.

  A test step E9 makes it possible to check whether all the coefficients Xij have been considered. Otherwise, the following coefficient is considered in the scanning order in the reading step E4 and the steps E3 to E9 of the insertion process are repeated. When all the quantized spectral coefficients Xij have been scanned, entropy coding is again carried out in a coding step E10 of all the coefficients so as to reconstruct a compressed file

containing the marked image.

  Thanks to the insertion method according to the invention, the additional information inserted W is imperceptible and does not modify the visual perception of the image. This secret mark can also convey with the image

additional information.

  Although only the last bit plan carries the information, since the last bit plan substitution rule takes into account

  several parameters, the distribution of these modified bits will be of the random type.

  A fraudulent modification of these last bit plans, to mask a modification of the image, will be very difficult to achieve since it is impossible to deduce the link existing between an inserted information bit, equal to 0 or 1, and the or

the modified bit (s).

  It will also be noted that the operations implemented directly on the quantized spectral coefficients present a low algorithmic complexity, hence a rapid insertion of the information.

  additional in a compressed file.

  It directly authenticates an image in its compressed format and it is intrinsically robust to compression methods

based on the JPEG standard.

  This inserted secret mark W is moreover fragile to all the modifications which the image I can undergo, intentionally or not, in its compressed form so that it can then serve to authenticate the image or not. For this, and as illustrated in FIG. 7, the authentication method associated with the insertion method in this embodiment firstly comprises an entropy decoding step Ell in order to extract all of the quantized spectral coefficients Y1 of the compressed file supplied as input to the decoder. The image J to be authenticated must be in exactly the same format as the original image I. As before, the first coefficient Yj is considered in a predetermined scanning order in the reading step E12 and compared in a step E13 test the magnitude of this coefficient Y1j at a threshold value T identical to that used in the insertion process. If the magnitude of this coefficient Yj is less than the threshold value T, we consider in a step of

reading E14 the following coefficient.

  The subgroup of coefficients Yjj into which the additional information W has been inserted is thus reconstituted. In the calculation steps E15 and E16, the control bit Cij is then calculated according to the same predetermined operation as previously as a function of the 15 first bit planes of the coefficient Yij and

  of the first 15 bit plans of the confidential key.

  As previously, the predetermined operation consists in performing a sum S of binary operations, by the exclusive binary "OR" operator, on the first 15 bit planes of the coefficient and the first 15 bit plans of the confidential key K and at calculating the control bit Cj as a function of the parity of this sum S. It will be noted that thanks to the insertion method according to the invention, the control bit Cij is calculated from the bit plans of the coefficient which have no not changed to insert additional information. We therefore find in the calculation step E16 a control bit Cij identical to the control bit Cj calculated during the insertion process if the image to be authenticated J is identical to the original image I. The authentication process comprises then an extraction step E1 7 of the value of the additional information inserted w'ij according to the predetermined reversible rule used during the insertion, as a function of the bit of

  check Cij and the value of the last bit plane Yoj of the coefficient Yjj.

  The value of the additional information w'ij is equal to a state or another state, for example equal to 0 or 1, depending on whether or not the value is equal

  of the last bit plane Y ij and of the value of the control bit Cij.

  With the rule used previously, we have: - If YOj = Cij, then w'ij = 0; and

  - if Y 1ij = 1 - Cij, then w'ij = 1.

  We can also note in summary w'1j = Cij XOR Yjj.

  We can thus extract all the values w'ij from the additional information W which are supposed to have been inserted by modulation of the

  last bit plane Y0j of the coefficients Y1j.

  The authentication process then includes a step of comparison E18 of the value extracted w'ij from the additional information.

  inserted and binary value wuj for additional information.

  If these two values w'j and w1j are different, we increment in

  a step E19 an error counter.

  It is then checked in a test step E20 if all the coefficients Yij have been considered and if not, the following coefficient is considered in the scanning order and the steps E13 to E20 of the authentication process are repeated. The latter finally comprises a decision step E21 which makes it possible to authenticate or not the image J as being the original image I as a function of the number of errors recorded in the error counter. If the total number of errors is zero, on. can reliably conclude that the processed image J is indeed the original image I. If the number of errors is different from zero, we deduce that the image has been manipulated. We can possibly adopt a majority decision rule and decide that the image has been manipulated only if the majority of the values extracted w'jj are different from the values wij of the information

  additional W supposed to have been inserted.

  An advantage of the insertion and authentication method according to the invention lies in the possibility of locating in frequency (as a function of the spatial support of the coefficient X1j in the spectral representation of the image) the possible modifications which have been carried out on the picture and know

* possibly the type of distortion applied to the image.

  We will now describe with reference to FIGS. 10 to 12 a

  second embodiment of the invention.

  Only the modified steps of the insertion and authentication methods, compared to the first embodiment described above, will be described in detail above, the identical steps

  bearing identical reference numbers in the figures.

  In this second embodiment, the insertion of the additional information W is also carried out in a compressed file using a coding method based on the JPEG standard, but unlike the previous example, the number M of planes of bits substituted during insertion

is equal to 2.

  The secret image to be inserted W is created as in the previous example from an image with a resolution 8 times smaller than the image to be processed I. As before, an entropy decoding step E1 makes it possible to extract the coefficients quantized spectral Xij and a test step E3 makes it possible to choose a subset of coefficients Xu whose magnitude is greater than a threshold value T. It is further decreed in this example that all the coefficients

  low frequency DC will be kept unchanged.

  As before, the coefficients Xij are represented on 16

  bits and a confidential key K, also represented on 16 bits, is used.

  As illustrated in Figure 10, we calculate a control bit Cuj aux

calculation steps E5 and E6.

  In step E5, a bit-by-bit sum S is calculated between the first 14 bit plans of the coefficient Xj and the first 14 bit plans of the confidential key K: S = EX, XORKk k = 2 Then the bit of Cij control according to the parity of the

  sum S, that is to say that Cij = 0 if S is even and Cij = 1 if S is odd.

  In the step of calculating and substituting the last two bit planes xjj and Xlij, the following determinism rule is used: - if Wij = 0, Xiij = Cij and X q = 1 - Xlij; and - if wij = 1, Xlij = 1 - Cij and X qj = 1 - X1 j Thus, in this embodiment in which the number M of substituted bit plans is equal to 2, the value of the penultimate bit plan xlj is

  equal to the alternative value of the last bit plane X jj.

  It is verified in a test step E9 that all the coefficients Xij have been processed and then an entropy coding step E10 makes it possible to compress by

  new image marked 1 'in a file.

  As illustrated in FIG. 12, we then proceed, to authenticate an image J stored in a compressed file, again to an entropy decoding El 1 of the file to extract the quantized spectral coefficients Yij from the file and we again choose a subset coefficients whose magnitude is greater than the threshold value T used when inserting the additional information W. Then, as before, the control bit Cij is calculated from the first 14 bit plans of a coefficient Yij and the First 14 bit plans of the confidential key K. We then reverse the deterministic rule used when substituting the last two bit plans to extract the value w'j from the information

binary inserted.

  - if Ylij = Cij, and Y ij = 1 - Ylij, then w'j = 0 - if Ylij = 1 - Cij and Y ij = 1 - Ylij, then w'j = 1 - if Y11j = Y ij, then we cannot decode the information w'j Thus, in the extraction step E17, when the number M of substituted bit plans is equal to 2, the value w'1j of the additional information is extracted only if the value of the penultimate bit plane Y1j is equal to the

  alternative value of the last bit plane Y j.

  The steps of comparison E18 of the values of the information extracted w'j and of the binary information inserted wij, of incrementation E19 of an error counter and of decision E21 are identical to those described in the first

embodiment.

  This embodiment is especially of interest when a majority decision is taken on the results obtained from the stages

  E17 extraction and E18 comparison.

  By thus using the last two bit planes of each coefficient to insert the modulation, the probability that a spectral coefficient which did not exist in the original image I (therefore coming from a modification of image 1) allows anyway to extract a correct value

  of the information inserted w'ij is weaker.

  In this embodiment, it is possible to envisage the distinction between large errors due to real modifications of image I and small errors due to small variations in luminance (originating from

  example of a slight overcompression of the file).

  A third embodiment will now be described in

reference to figs 13 and 14.

  Only the modified steps of the insertion and authentication methods, compared to the embodiments described above, will be described in detail above, the identical steps

  bearing identical reference numbers in the figures. -

  In this embodiment, it is considered that the original image I is represented in its spatial domain, in raw data, by a set of

  pixel lIj whose value varies on 256 gray levels for example.

  The image I of size N1 on N2 comprises a set of pixels Ii,

  located on the ith row and jth column, and coded in this example on 8 bits.

  A confidential key K, also represented on 8 bits, is used. In this example, an insert mark W is used formed of a binary image of the same size and same resolution as the original image I. Thus, the mark W comprises a set of information bits wij,

  located on the ith row and jth column, i andj varying from 1 to N1 and from 1 to N2.

  As described below, each bit of information w1j of the secret mark will be modulated in the original image I and may therefore be

  extracted for authentication of image J at the input of a decoder.

  As illustrated in FIG. 13, to insert the additional information, in a first step E2, a first pixel lIj is considered, read in a predetermined scanning order of the image I. In this example, we choose to modulate only the last plan of

bits l j of each pixel lIj.

  In a calculation step E5, a sum S of binary operations performed on the first 7 bit planes of a pixel Iij and on the first 7 bit planes of the confidential key K is calculated. The sum SS = Ei is thus obtained I'ORKk k = l

  o "OR" represents the binary "OR".

  Then, in a calculation step E6, the control bit Cj is calculated as before, as a function of the parity of the sum S. Then, in test step E8, the value of the binary information to be inserted wij is read and used the following rule for calculating and substituting the least significant bit plan in the substitution step E7: -if wj = 0, then li = Cij, and - if wj = 1, then l 0j = 1 - Cij A step of test E9 makes it possible to check whether all the pixels Ij have been considered and if not, the next pixel is considered in the reading step

  E4 and the steps E5 to E9 of the insertion process are repeated.

  This gives an image marked 1 'which protects the image

  original I in its initial spatial representation.

  To authenticate an image J, as illustrated in FIG. 14, the reverse operations are carried out on each pixel Ji of the image J to extract the

  value w '; of the information bit inserted.

  The first pixel Jj is considered in step E12 in the predetermined scanning order and the control bit Cj is calculated, as before, in the calculation steps E15 and E16 from the first 7 bit planes of the pixel Jj1 and of the first 7 bit planes of the confidential key K. We reverse the rule used during the substitution of the last bit plan 1 j of each coefficient Iu to find the value w'j of the information bit inserted: - if J ij = Cij, then w'i = 0; and

- if J0j = 1 - Cij, then w'j = 1.

  The value extracted w'j in the extraction step E17 is then compared with the value wij which has actually been inserted in a comparison step E18. An incrementing step E19 makes it possible for each pixel Ju to increment an error counter when the values extracted w '; and inserted

  wij of the information W are different.

  In the test step E20, it is checked whether all the pixels Jij of the image J have been processed, and if not, the next pixel is considered in a reading step E14. We repeat steps E15 to E20 of the authentication process on

all pixels Ij. A strict decision step E21 authenticates the image

  when the number of errors is 0.

  As in the previous examples, any local manipulation of

  the initial image I can be located at the level of each pixel. -

  The present invention thus makes it possible to authenticate data

  digital in both compressed and uncompressed form.

  It applies in particular to cameras

  digital images that store digital images in a compressed file.

  Of course, many modifications could be

  made to the examples described above without departing from the scope of the invention.

  Thus, the insertion of the secret mark can also be implemented on spectral coefficients obtained by a multi-resolution spectral decomposition of the digital signal, such as an image, for example by a

  spectral decomposition in wavelets.

  One can also, in order to better hide the information inserted, use a more complex calculation rule to calculate the control bit as a function of the confidential key, since this calculation rule can be

reproduced at the decoder.

Claims (46)

  1. Method for inserting additional information (W) such as a secret mark for authenticating a digital signal (1) broken down into a set of coefficients (Xij, Ij), each coefficient being represented on L bits, and said additional information (W) being represented by a set of binary values (wij) of the same size as said set of coefficients (Xij, luj), characterized in that it comprises the following steps: - calculation (E5, E6), for at least one coefficient (Xuj, Iij), of a control bit (Cij) according to a predetermined operation as a function of L - M first bit planes of said coefficient (Xj, Ij); - calculation (E7, E8) of M last bit plans (X dd, Xlij, 1 dd) according to a predetermined reversible rule as a function of the control bit (Cij) and of a binary value (wij) of the additional information (W); and - substitution (E7) of the M last bit plans of said coefficient (Xij,   Ij) by the M last calculated bit plans (X qi, X1j, I lj).   2. insertion method according to claim 1, characterized in that in the step of calculating (E5, E6) a control bit (Cij), said predetermined operation is a function of the L - M foregrounds bits and a key   confidential (K) represented on L bits.   3. Insertion method according to claim 2, characterized in that said predetermined operation consists in carrying out a sum (S) of binary operations on the L - M first bit planes of said coefficient (X1j, Iij) and the L - M first bit planes of said confidential key (K) and to be calculated   the control bit (Cij) as a function of the parity of said sum (S).   4. insertion method according to one of claims 1 to 3,   characterized in that the number M of substituted bit plans is equal to 1 or 2.   5. insertion method according to one of claims 1 to 4,   characterized in that in the step of calculating (E7, E8) of the M last bit planes, the value of the last bit plan (Xj, lj) is equal to the value of the control bit (Cij) or to the alternative value (1 - C) of the control bit according to the status of the value   binary (wij) additional information.   6. insertion method according to one of claims 1 to 5,   characterized in that the number M of substituted bit plans is equal to 2 and in that in the step of calculating (E7, E8) of the M last bit plans, the value of the penultimate bit plan (X1j) is equal to the alternative value (1 - X jq) of last bit plane (X ij).   7. Insertion method according to one of claims 1 to 6, the   digital signal (I) being decomposed into a set of quantized spectral coefficients (Xij), characterized in that it further comprises a prior step of choosing (E3) a subset of quantized spectral coefficients (Xij), in which the chosen coefficients (Xu) have at least a magnitude strictly greater than zero, the steps of calculating (E5, E6) of a control bit, of calculating (E7, E8) of the M last bit plans and of substitution ( E7) being   implemented for each coefficient (Xij) of said subset.   8. Insertion method according to claim 7, characterized in that in the selection step (E3), the subset of coefficients comprises the quantized spectral coefficients (Xij) whose magnitude is greater than one threshold value (T).   9. insertion method according to one of claims 7 or 8,   characterized in that the set of binary values (wij) representative of the additional information (W) is generated by the repetition of an initial binary information of size smaller than the size of the set of coefficients quantized spectral (Xij).   10. Insertion method according to one of claims 7 to 9,   for the authentication of a digital signal (1) stored in a compressed file, characterized in that it further comprises a preliminary step of entropy decoding (El) adapted to extract the quantized spectral coefficients (Xij) and a step of entropy coding (E10) after said substitution step (E7).   11. Method for authenticating a digital signal (J) from additional information (W) such as a secret mark inserted in said   digital signal by the insertion method according to one of the claims   1 to 10, said digital signal (J) being broken down into a set of coefficients (Yj, Jj), each coefficient being represented on L bits, characterized in that it comprises the following steps: - calculation (E15, E16), for at least one coefficient (Yj, J1j), of a control bit (Cij) according to said predetermined operation as a function of L - M first bit plans of said coefficient; - Extraction (E17) of the value (w'ij) of the additional information inserted according to said predetermined reversible rule as a function of the control bit (Cij) and of M last bit plans (y0jj, yjj, J0j%); - comparison (E18) of the value extracted (w'ii) from the additional information inserted and the binary value (w1j) of the additional information (W); and - decision (E21) to authenticate or not the digital signal (J) according to the identity or not of said extracted value (w'ij) and said value   binary (wij) of the additional information inserted.   12. Authentication method according to claim 11, characterized in that in the step of calculating (E15, E16) of a control bit (Cij), said predetermined operation is a function of the L - M foregrounds of   bits and a confidential key (K) represented on L bits.   13. An authentication method according to claim 12, characterized in that said predetermined operation consists in carrying out a sum (S) of binary operations on the L - M first bit plans of said coefficient (Y1j, J1j) and the L - M first bit planes of said confidential key (K) and in calculating the control bit (Cij) as a function of the parity of said sum (S).   14. Authentication method according to one of the claims   11 to 13, characterized in that the number M of bit plans in step   extraction (E17) is 1 or 2.   15. Authentication method according to one of claims   11 to 14, characterized in that in the extraction step (E17), the value (w'ij) of the additional information is equal to a state or another state depending on whether or not the value is equal of the last bit plane (Y0j, J0ij) and the value of the bit of control (Cij).   16. Authentication method according to one of the claims   11 to 15, characterized in that in the extraction step (E1 7), the number M of bit plans is equal to 2 and in that the value of the additional information (w'1j) is extracted only if the value of the penultimate bit plane (Y1j) is equal to   the alternative value (1 - Yojj) of the last bit plane (Y ij).   17. Authentication method according to one of the claims   11 to 16, the digital signal (J) being broken down into a set of quantized spectral coefficients (Jj), characterized in that it also comprises a prior step of choosing (E13) a subset of quantized spectral coefficients , in which the coefficients chosen (J1j) have a magnitude greater than a threshold value (T), the steps of calculation (E15, E16) of a control bit, of extraction (E17), of comparison (E18) and of decision   (E21) being implemented for each coefficient (Jjj) of said subset.   18. An authentication method according to claim 17, for the authentication of a digital signal stored in a compressed file, characterized in that it further comprises a prior decoding step.   entropic (El 1) adapted to extract the quantized spectral coefficients (Jij).   19. Device for inserting additional information (W) such as a secret mark for authenticating a digital signal (I) broken down into a set of coefficients (Xij, Iij), each coefficient being represented on L bits, and said additional information (W) being represented by a set of binary values (wij) of the same size as said set of coefficients, characterized in that it comprises: means for calculating (21) a control bit (Cij) according to a predetermined operation as a function of L - M first bit plans of a coefficient (Xij, Iij); - Means for calculating (22) M last bit plans (X ij, X1ij, law) according to a predetermined rule reversible as a function of the control bit and of a binary value of the additional information; and substitution means (23) of the M last bit plans of said   coefficient (X1j, Iij) by the M last calculated bit plans (X jj, X1lij, l ij).   20. Insertion device according to claim 19, characterized in that said predetermined operation of the calculation means (21) is a function of the L - M first bit plans and of a confidential key (K) represented on L bits.   21. Insertion device according to one of claims 19 or   , characterized in that the number M of substituted bit plans is equal to 1 or 2.   22. Insertion device according to one of claims 19 to 21,   the digital signal (I) being decomposed into a set of quantized spectral coefficients (Xij), characterized in that it further comprises means for choosing (26) a subset of quantized spectral coefficients adapted to choose coefficients (Xij) having at least one magnitude strictly greater than zero.   23. Insertion device according to claim 22, characterized in that the selection means (26) are adapted to choose quantized spectral coefficients (Xj) whose magnitude is greater than a threshold value (T).   24. Insertion device according to one of claims 19 to 23,   for authenticating a digital signal (1) stored in a compressed file, characterized in that it further comprises entropy decoding means (25) adapted to extract the quantized spectral coefficients and entropy coding means (27).   25. Insertion device according to one of claims 19 to 24,   characterized in that it is incorporated in a microprocessor (100), a read only memory (102) comprising a program for inserting additional information (W) in a digital signal (I) and a random access memory (103) comprising adapted registers to save modified variables at during program execution.   26. Device for authenticating a digital signal (J) from additional information (W) such as a secret mark inserted into said digital signal by the insertion process in accordance with one of the   claims 1 to 10, said digital signal (J) being broken down into a   set of coefficients (Yj, Juj), each coefficient being represented on L bits, characterized in that it comprises: - means for calculating (31) a control bit (C1j) according to said predetermined operation as a function of L - M first bit plans of a coefficient (Yij, Jij); means for extracting (32) the value (w'ij) of the additional information inserted according to said predetermined reversible rule as a function of the control bit (Cij) and of M last bit plans (Y0j, ylj, J0j) said coefficient; - comparison means (33) of the value extracted (w'u) from the additional information inserted and the binary value (wqj) of the additional information; and - decision means (34) whether or not to authenticate the digital signal as a function of the identity or not of said extracted value (w'ij) and of   said binary value (wij) of the additional information inserted (W).   27. Authentication device according to claim 26, characterized in that said predetermined operation of the calculation means (31) is a function of the L - M first bit plans and of a confidential key (K) represented on L bits.   28. Authentication device according to one of claims   26 or 27, characterized in that the number M of bit planes used by the   extraction means (32) is equal to 1 or 2.   29. Authentication device according to one of the claims   26 to 28, the digital signal (J) being broken down into a set of quantized spectral coefficients (Yij), characterized in that it also comprises means for choosing (36) a subset of adapted quantized spectral coefficients to choose coefficients (Yu) having a magnitude   greater than a threshold value (T).   30. Authentication device according to claim 29, for the authentication of a digital signal (J) stored in a compressed file, characterized in that it further comprises entropy decoding means   (35) adapted to extract quantified spectral coefficients (Yij).   31. Authentication device according to one of the claims   26 to 30, characterized in that it is incorporated in a microprocessor (100), a read-only memory (102) comprising a program for authenticating a digital signal (J) and a random access memory (103) comprising registers adapted to   save variables changed during program execution.   32. Computer, characterized in that it includes suitable means   implementing the insertion method according to one of claims 1 to 10.   33. Computer, characterized in that it includes means suitable for implementing the authentication method in accordance with one of the claims 11 to 18.   34. Computer, characterized in that it comprises a device   insertion according to one of claims 19 to 25.   35. Computer, characterized in that it comprises a device   authentication according to one of claims 26 to 31.   36. Apparatus for processing a digital signal, characterized in that it includes means suitable for implementing the insertion method   according to one of claims 1 to 10.   37. Apparatus for processing a digital signal, characterized in that it includes means suitable for implementing the method   authentication according to one of claims 11 to 18.   38. Apparatus for processing a digital signal, characterized in that   that it comprises an insertion device according to one of claims 19 to   25. 39. Apparatus for processing a digital signal, characterized in that it comprises an authentication device conforming to one of the claims 26 to 31.   40. Digital camera, characterized in that it includes means suitable for implementing the insertion process   according to one of claims 1 to 10.   41. Digital camera, characterized in that it includes means adapted to implement the authentication process   according to one of claims 11 to 18.   42. Digital camera, characterized in that it   comprises an insertion device according to one of claims 19 to 25.   43. Digital camera, characterized in that it   comprises an authentication device according to one of claims 26   to 31. 44. Digital camera, characterized in that it includes means suitable for implementing the insertion process in accordance with one of the claims 1 to 10.   45. Digital camera, characterized in that it includes means suitable for implementing the authentication method in accordance with one of claims 11 to 18.   46. Digital camera, characterized in that it includes a   insertion device according to one of claims 19 to 25.   47. Digital camera, characterized in that it includes a   authentication device according to one of claims 26 to 31.