Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T1/00—General purpose image data processing
  • G06T1/0021—Image watermarking
  • G06T1/005—Robust watermarking, e.g. average attack or collusion attack resistant
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2201/00—General purpose image data processing
  • G06T2201/005—Image watermarking
  • G06T2201/0065—Extraction of an embedded watermark; Reliable detection

Definitions

• the invention is in the field of digital video tattooing
• the invention relates to a tattoo reading method.
• the invention has many applications, such as, for example, tattoo playback on a video file (of the MPEG 2 type for example). It allows for example the protection against the piracy of films by screen shots ("DVD screeners"), that is to say by illegal recording by filming a cinema screen.
• DVD screeners screen shots
• the present invention can be applied in all cases where the data to be processed can be stored to a certain depth, so as to be able to go back during the processing of these data.
• the invention can be applied to any type of support data (video, image, audio, text, 3D data, etc.).
• the invention applies regardless of the format of the support data: in the form of a stream (for example a video signal captured on a television channel), in the form of a file, in the form of a recording (cassette , DVD ...), etc.
• a stream for example a video signal captured on a television channel
• a file for example a file
• a recording for example a recording of a recording
• the context of the invention is explained in more detail below by taking as an example the case of a sequence of video type data.
• the invention can however be applied to any type of data set.
• the tattoo's reading can be made difficult by the alterations suffered by the video tattoo support.
• a tattoo system is an information transmission system. As illustrated in FIG. 1. it thus comprises (at least) a transmitter 1 (the "tattoo artist" itself), which modifies a video support for inserting the tattoo message, and
• a receiver (at least) a receiver ("tattoo player") 2, which upon receipt of any video determines whether it is tattooed and, if so, what is the message inserted.
• the transmitter (tattoo artist) 1 takes as input a set of "support” data (typically video data) and a certain number of parameters.
• the parameters include generally the tattoo message (also called “tattoo carrier"), which is represented by a sequence of M bits (M ⁇ l).
• M ⁇ l the tattoo message
• the tattoo artist modifies the video to produce a video tattooed.
• the receiver (tattoo player) 2 takes a video input as well as possibly a certain number of parameters (for example a key).
• the original video which can be used to facilitate the determination of the channel parameters, by comparison between the original video and the video tested (video on which we seek to read the tattoo).
• the original video can also be used to facilitate the reading of the tattoo message: for example, subtracting the original from the tested video can reduce the impact of the noise affecting the tattoo, especially the noise produced by the original video. - even. From these data he usually performs a detection step, which determines whether the video is tattooed or not.
• the detection If the detection is positive, it carries out a step of determining (or decoding) the tattoo message or messages (one can have several tattoo messages if the support data comes from a compositing operation, for example mounting multiple video sequences, averaging images, making a montage or mosaic using several support data, etc. ).
• the tattoo messages detected are possibly accompanied by auxiliary information: for example a reliability index (estimation of the probability that the estimated tattoo message is wrong), tattoo location information (temporal and / or spatial location, in the case of compositing). Note that all the steps included in the reading process can be performed either directly on the video data itself, or on a transformed representation of the data (Fourier transform, discrete cosine transform - DCT-, wavelet transform, etc.). .
• tattooing is generally of interest only being associated with supporting data, for example to identify a broadcast extract.
• tattoo channel Between the tattoo artist 1 and the reader 2, the support data (video) generally undergo a whole series of transformations. These transformations are linked to the technical contingencies of the image's life (transmission, storage, exploitation ...), but also to possible intentional attacks aimed at making the tattoo unusable.
• the term "tattoo channel” is used to describe all the modifications that the tattooing signal undergoes as a result of changes in the support data. These modifications can cause an alteration (“noise”) of the tattoo signal and thus make the reading more difficult and / or subject to errors.
• the tattoo signal (or some of its characteristics) is intrinsically invariant after passage through the tattooing channel (choice of modulation technique, redundant coding, etc.); when reading the tattoo, estimate the value of the alterations generated by the tattooing channel (channel parameters) in advance so as possible to "reverse" the transformation performed by the channel.
• This second strategy is widely used to estimate the parameters of the geometric transformations caused by the tattoo channel. She sometimes uses an auxiliary tattoo signal (also called "cotatouage").
• the reading process generally takes place in three main steps: a step (41) of determining some of the channel parameters (typically geometric deformations), possibly using a cotatouage signal; a step (42) for resetting the video: from the previously estimated channel parameters, the transformation corresponding to the tattooing channel is estimated, then this estimated transformation is reversed (if this is possible) and finally the "estimated transformation” is applied reverse "on the video test; a step (43) of determining the tattoo message on the failed test video.
• the cototouage signal also called reference signal or tattoo, synchronization signal, auxiliary signal, learning signal, service signal
• the cototouage signal is inserted by the tattoo artist 1. It is known to the reader 2 and makes it possible to estimate certain alterations ( channel settings) experienced by the video.
• a first known technique of digital tattooing is described in the following article: S.Baudry, JFDeligle, B.Sankur, B.Macq, H.Maitre, "Analyzes of error correction strategies for typical communication channels in watermarking", pp 1239 - 1250 in Signal Processing, Vol 81, No. 6, June 2001.
• This article proposes a method for encoding and decoding the carrier tattoo signal in an image or a video.
• a correction code is used, formed by concatenation of a BCH code and a repetition code.
• a flexible decoding method is proposed, which increases the detection performance compared to a conventional firm decoding. However, the decoding is carried out frame by frame and does not take into account the whole of the video.
• the tattooed message contains only 1 bit: the reading is simply to determine if the tattoo is present or absent. If we insert a message with a large number of bits, the method is no longer suitable because it requires a number too many operations (see below for exhaustive decoding considerations).
• the interest of this method is to be robust to the geometrical transformations operating on the images. On the other hand, it allows to bury only a tattoo of low information density (a few bits per minute).
• the proposed detection method does not allow to decode the tattoo correctly.
• This US patent 6516079 discloses a tattoo detection method in two or more steps. Each step consists of the measurement of probabilities of presence ("detection value”) calculated in different ways. In particular, the authors propose two types of measures: the absolute one, consists in the comparison of a detection score
• the tattoo message is generally encoded using an error correcting code. From k message bits, such a code provides a word of n bits (n> k).
• the set of n-bit words obtained from the coding of all the possible words of k bits is called the code (the set of code words). Since n> k, all words of n bits do not belong to the code.
• a n-bit word is received. This one is not necessarily a code word (some bits could be modified because of the errors of channel). It can be shown that for most "realistic" channels, the optimal decoding (which produces the probability of minimal decoding error) is MAP (Maximum A Posteriori) decoding. MAP decoding searches for, among the set of code words, the one that is closest in the sense of the Hamming distance of the received word. It is recalled that the Hamming distance between 2 words ml and m2 is the number of different bits between these words ml and m2. The reliability of the decoding can then be given as a function of the Hamming distance between received word and decoded word.
• MAP decoding can be implemented simply by performing an exhaustive decoding. For that, one generates the set of the words of code (2 k possible words), then, for each one of the code words, one calculates its Hamming distance to the received word. The decoded word is then the code word whose distance to the received word is minimal.
• Figure 5 is a geometric representation of an incomplete tattoo decoding.
• the decoding regions are spheres of radius d min , each centered on a separate code word C 1 .
• the received word R 2 is located in a decoding sphere, it can be decoded at C 3 .
• the received word R 1 is located outside the decoding regions and therefore can not be decoded.
• the density of the decoding spheres with respect to the set of possible words is very low. This means that for a random word, decoding is very unlikely to succeed. If the channel is very noisy, the decoding rate of the tattoo will also be very low.
• the invention particularly aims to overcome these disadvantages of the state of the art.
• the invention also aims, in at least one embodiment, to provide such a technique to easily obtain a measure of reliability of the decoding performed.
• Another objective of the invention in at least one embodiment, is to provide such a technique making it possible to obtain, in addition to the value of the tattooing message or messages, additional information which may be indispensable in certain applications: typically evaluation the reliability of the tattoo read, location of this tattoo in the support data.
• a complementary objective of the invention in at least one embodiment, is to provide such a technique to save the memory space necessary for its implementation in the tattoo reader. 4.
• a tattoo reading process of the type comprising a step of determining at least one tattoo message contained in a set of data, possibly preceded by a step of determining at least one set of parameters of a tattooing channel undergone by said set of data, each set of channel parameters comprising at least one channel parameter.
• Said step of determining at least one tattoo message and / or said prior step of determining at least one set of channel parameters comprises (nent): an estimation phase, of processing at least one batch of data comprising data of said data set, so as to obtain N estimated elements, with N>0; and a verification step of validating or invalidating, in view of all or at least a subset of said set of data, each estimated item; where each estimated element is an estimated tattoo message, in the case of said step of determining at least one tattoo message, and an estimated set of channel parameters, in the case of said prior step of determining at least one tattoo message, a set of channel parameters.
• the invention makes it possible to improve the performance of at least one of the two determination steps conventionally included in a tattoo reading process, namely (see FIG. 4): the step (41) of determination of at least one channel parameter and the step (43) of determining at least one tattoo message.
• the general principle of the invention consists in replacing in one and / or the other of these two determination steps (41, 43), a conventional processing mechanism, realized in a single pass and relating to the all data, through a new and inventive treatment mechanism, realized in two successive phases: firstly, an estimation phase relating only to one or more batches of data of the data set, then a verification phase relating to all or one or more subsets of the data in the set of data.
• the invention highlights the similarities between the determination of channel parameters and the reading of tattoo messages. Indeed, when reading the tattoo message or messages, we seek to estimate the "parameters" of the tattoo system ("" transmitter "), that are the tattoo messages themselves.
• the invention can also be seen as an adaptation of a tattoo decoding method and / or "run-of-the-river" channel parameter estimation (i.e. adapted to a real-time reading), to a context not real time
• each batch of data comprises only a part of said set of data.
• the tattoo reading method according to the invention is even faster than the size of each batch of data is low.
• said estimation phase consists in processing at least two data batches, so as to obtain Ni element (s) estimated for each data batch, with Ni> 0.
• said set of data is a sequence of video type data comprising a plurality of successive images, and in that each batch of data belongs to the group comprising: batches each consisting of an image portion; batches each consisting of an image; batches each consisting of a group of images.
• said verification phase comprises the following steps, for each estimated element: division of all or part of said set of data into one or more subsets of data; - for each of the subsets:
• the posterior probability of an estimated element constitutes a measure of reliability of the decoding performed (that is to say of the estimate of this element, set of channel parameters or tattoo message).
• said step of calculating the posterior probability of said element conditionally estimated to a given subset comprises the following steps: - intermediate calculation of the posterior probability of said element, conditionally to at least two sub-portions forming together said subset together ; combining the results obtained at the end of said intermediate calculation step, for said at least two sub-portions, so as to obtain said posterior probability of said element estimated conditionally to the whole of said subset.
• said verification phase makes it possible to locate each estimated element on each subset for which said estimated element is validated.
• said set of data is a video-type data sequence comprising a plurality of successive images, and in that each subsequence comprises one or more images.
• said set of data belongs to the group comprising: video data sets; image data sets; audio data sets; - text data sets; data sets of the 3D data type.
• said estimation phase and / or said verification phase also makes it possible to obtain a reliability measure associated with at least one estimated element.
• the method further comprises a step of resetting said set of data into a set of recaled data, said resetting step being executed for at least certain set (s) of parameters. determined channel (s) during said step of determining at least one set of channel parameters.
• said step of determining at least one tattoo message is implemented with each set of recaled data resulting from a separate execution of said resetting step.
• said resetting step is executed only for the set (s) of estimated channel parameters whose associated reliability measurement is greater than a second predetermined threshold.
• the invention also relates to a tattoo reading device, of the type comprising a module for determining at least one tattoo message contained in a set of data, and possibly a module for determining at least one set of parameters of a tattoo. a tattooing channel undergone by said set of data, each set of channel parameters including at least one channel parameter.
• Said module for determining at least one tattoo message and / or said module for determining at least one set of channel parameters comprises: estimation means, carrying out a treatment of at least one set of parameters; data comprising data of said data set, so as to obtain N estimated elements, with N ⁇ 0; and verification means, performing a validation or invalidation, in view of all or at least a subset of said set of data, of each estimated element; where each estimated element is an estimated tattoo message, in the case of the module for determining at least one tattoo message, and an estimated set of channel parameters, in the case of the module for determining at least one set of tattoo channel settings.
• the invention also relates to a computer program product, comprising program code instructions for executing the steps of the method according to the invention mentioned above, when said program is executed on a computer.
• the invention also relates to a storage medium, possibly completely or partially removable, readable by a computer, storing a set of instructions executable by said computer to implement the method according to the invention mentioned above. 5. List of figures
• FIG. 1 shows a generic diagram of a tattoo system
• - Figure 2 shows a generic scheme of operation of a tattooist appearing in Figure 1
• Figure 3 shows a generic scheme of operation of a tattoo reader shown in Figure 1
• Figure 4 shows a generic diagram of a tattoo reading process
• - Figure 5 is a geometric representation of an incomplete tattoo decoding
• FIGS. 1 to 5 illustrate the principle of the generic mechanism of two-phase processing according to the invention (an estimation phase and then a verification phase), which can be implemented in a tattoo reading process, and more specifically in the step of determining at least one channel parameter and / or in the step of determining at least one tattoo message;
• FIG. 7 illustrates the implementation of the generic mechanism of FIG. 6 in the step of determining at least one tattoo message;
• FIG. 8 illustrates the implementation of the generic mechanism of FIG. 6 in the step of determining at least one channel parameter;
• FIG. 9 illustrates a particular embodiment of the invention, in which the generic mechanism of FIG. 6 is implemented in the step of determining at least one channel parameter and also in the determining step at least one tattoo message. 6.
• FIGS. 1 to 5 relate to the field and the technical context of the present invention. They have already been described above and are therefore not described again.
• FIGS 6 to 9 are specific to the present invention. They are described in detail below. In the rest of the description, for example, the case of a set of video type data is taken as an example. Remember that the invention can however be applied to any type of data set (image, audio, text, 3D data, etc.).
• the generic two-phase processing mechanism according to the invention which comprises a phase 61 for estimating the best candidates, followed by a verification phase 62 of the best estimated candidates, is now presented in relation with FIG.
• This mechanism according to the invention can be implemented in a tattoo reading process.
• the invention applies to the "reader" part of a tattooing system, which makes it more efficient.
• the mechanism according to the invention can be implemented: in the step of determining at least one channel parameter (step performed by the co-tattoo reader (synchronizing auxiliary signal) or by means implementing another technique not using an auxiliary synchronization signal); and / or in the step of determining at least one tattoo message (step performed by the tattoo reader carrier).
• the desired parameters also called elements
• these parameters are the tattoo messages (tattooist parameters).
• the channel parameters themselves (for example the parameters of the geometrical transformation).
• the verification phase 62 (second phase), from the video, all the a priori parameters from the previous phase 61 are consolidated and validated. During this verification phase 62, information can also be calculated. additional, such as an estimate of the reliability of the estimated parameters, or seek to locate more precisely the different parameters (for example, find which precise images are associated with a given tattoo message value). New subsets of images are then determined (possibly different from the batches of the first phase); each image subset corresponding to a tattoo value, or to a value of a set of channel parameters.
• the mechanism according to the invention optionally comprises a resetting phase 63 (third phase), only in the case of the determination of the channel parameters.
• the channel parameter or parameters determined during the second phase 62 are applied to the entire sequence or to each set of images (see in the paragraph above the notion of subset of images).
• the video registration consists in estimating the transformation corresponding to the tattooing channel from the estimated channel parameters, and then to inverting this estimated transformation (if this is possible) and finally to apply the "inverse estimated transformation" on the video.
• This results in a recalibrated video supposed to have the same spatial and geometric parameters as before the modification by the tattoo channel. It is this recalibrated video which is in this case used by the carrier tattoo reader to perform the estimation and verification phases 61 included in the step of determining at least one tattoo message.
• Figure 7. an example of implementation of the generic mechanism of Figure 6 in the step of determining at least one tattoo message (also called tattoo carrier).
• the mechanism according to the invention comprises a phase 71 for estimating the tattooing carrier values and a phase 72 for verifying the tattooing carrier values estimated during the preceding phase 71.
• a succession of decodings of the carrier tattoo is carried out, each decoding taking place on a batch of images of the video.
• a batch of images consists of either an image, an image portion, or a group of images (the number of images per group may be fixed or variable).
• the different batches of images may be disjoint or overlapping (in which case a portion of the signal may be part of two separate batches).
• the decoding may be incomplete (for example if the image batches are large) or exhaustive (for example if the image batches are of small size and / or if the number of possible values for the tattoo message is small ).
• decoding can either use only the tested video, or use the tested video and the original video. For example, the original data can be subtracted from the tested data, in order to reduce the noise due to the interference of the original data with the tattoo.
• the number of estimates may be zero (we failed to decode the lot), equal to 1, or greater than 1 (we have several possible estimates).
• Each estimate can be accompanied by a measure of reliability (for example: estimated probability of error, proportion of the lot on which it is estimated that the tattoo signal is present ).
• a measure of reliability for example: estimated probability of error, proportion of the lot on which it is estimated that the tattoo signal is present .
• the posterior probability can be calculated for example by a measurement of correlation between the two signals (m, - and r), which itself consists for example of a measurement of the Hamming distance between the two signals, possibly weighted by a measurement. reliability for each bit of the value m / estimated.
• the posterior probability can be computed either globally over the entire video, or over subsequences of the video (each comprising one or more images).
• the posterior probability can be computed either globally over the given subsequence, or over portions of the given subsequence (each comprising one or more several images).
• the second phase 72 will therefore better locate the portions of the sequence on which the tattoo message m is present.
• the second phase 72 also has the advantage of reducing the risks of decoding error. If indeed decoding an erroneous value m e tattoo during the first phase 71, it is unlikely that this value is found "by chance" in the rest of the video. We will therefore have a posterior probability of very low e on most of the video. The erroneous message will therefore be eliminated at the end of the second phase 72.
• FIG. 8 shows an example of implementation of the generic mechanism of FIG. 6 in the step of determining at least one tattoo channel parameter.
• the mechanism according to the invention comprises a phase 81 for estimating at least one set of tattoo channel parameters and a phase 82 for checking the sets of channel parameters estimated during the previous phase 81.
• the phases 81 and 82 for the calculation of the channel parameters will be described more simply than the phases 71 and 72 for the calculation of the tattoo messages (see FIG. 7), since they are largely similar to the latter.
• phase 81 for estimating the values of the channel parameters each batch of images of the video is processed, and a certain number of estimated sets of parameters are obtained, each possibly accompanied by a measure of reliability.
• the channel parameters can be estimated either from a cotatouage signal or by other techniques not using a synchronization auxiliary signal.
• the original video in addition to the video tested, so by comparison between the two to determine the parameters of the channel.
• this translation can be determined by calculating F intercorrelation between the images of the original video and the images of the video under test. The position of the maximum intercorrelation will give the value of the translation carried out by the channel.
• the phase 81 for estimating the channel parameters can be automatic, or it can be done manually, for example by recalibrating the original sequence and the sequence on which one seeks to read the tattoo again.
• the verification phase 82 of the estimated channel parameter sets is similar to the phase 72 (FIG. 7) of verification of the estimated tattoo values.
• further consideration can be given in this second phase
• the reliability measurement obtained for each set of parameters in the second phase 82 can be used during the next step of estimation / verification of the carrier tattoo (phases referenced 71 and 72 in Figure 7). It is thus possible to readjust the video (resetting phase 83, equivalent to the resetting phase 63 of FIG. 6), by "trying" each of the different sets of estimated parameters (provided that their reliability is considered sufficient). In other words, we then have several videos recaled and is carried out with each of them the next step of estimation 71 / verification 72 of the tattoo carrier. A tolerance is thus introduced, which makes it possible to reduce the impact of a poor estimation of the channel parameters on the estimation of the carrying tattoo.
• FIG. 9 illustrates a particular embodiment of the invention (already mentioned above), in which the generic mechanism of FIG. 6 is implemented both in the step of determining at least one parameter of channel and in the step of determining at least one tattoo message.
• the tattoo reading method then comprises: a step 91 of determining at least one channel parameter, of the type presented above in relation with FIG. -even :
• phase 81 for estimating at least one set of tattoo channel parameters
• the present invention has many advantages over the prior art. Indeed, it combines the advantages of a complete decoding and those of an incomplete decoding, while overcoming their respective drawbacks:
• the invention is much faster (especially compared to the second known technique described above);
• the invention makes it possible to obtain higher detection rates and to reduce error rates (especially with respect to first and fourth known techniques described above). It also makes it easier to obtain a measure of reliability of the decoding performed.
• the invention makes it possible to estimate the tattoo messages even if the video has been reassembled or composited. It is more general than the third technique and can be applied to any tattoo technique.
• the invention also makes it possible to obtain, in addition to the value of the tattooed message or messages, additional information which may be indispensable in certain applications: typically evaluation of the reliability of the tattoo read, location of this tattoo in the support data.
• the invention also makes it possible to save memory space, if only batches of images are considered and not the entire video during the first phase 71, 81. second phase 72, 82 (list of estimated parameters) is of reduced size.
• the present invention can be used in all tattoo applications requiring a high reliability of the read value and / or the calculation of the tattooing channel parameters (in particular the spatial deformation parameters):

Landscapes

• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Editing Of Facsimile Originals (AREA)
• Image Processing (AREA)
• Signal Processing For Digital Recording And Reproducing (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=34953056

Family Applications (1)

Country Status (6)

Families Citing this family (3)

Family Cites Families (8)

• 2004
  • 2004-10-07 FR FR0410605A patent/FR2876531A1/fr active Pending
• 2005
  • 2005-10-07 JP JP2007535199A patent/JP4829891B2/ja not_active Expired - Fee Related
  • 2005-10-07 EP EP05809103A patent/EP1797532A1/fr not_active Withdrawn
  • 2005-10-07 US US11/663,796 patent/US7804978B2/en not_active Expired - Fee Related
  • 2005-10-07 CN CN2005800339058A patent/CN101065769B/zh not_active Expired - Fee Related
  • 2005-10-07 WO PCT/FR2005/002483 patent/WO2006037899A1/fr active Application Filing

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events