Classifications

• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
  • H04N1/387—Composing, repositioning or otherwise geometrically modifying originals
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T1/00—General purpose image data processing
  • G06T1/0021—Image watermarking
  • G06T1/0028—Adaptive watermarking, e.g. Human Visual System [HVS]-based watermarking
  • G06T1/0035—Output size adaptive watermarking
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N5/00—Details of television systems
  • H04N5/76—Television signal recording
  • H04N5/91—Television signal processing therefor
  • H04N5/913—Television signal processing therefor for scrambling ; for copy protection
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2201/00—General purpose image data processing
  • G06T2201/005—Image watermarking
  • G06T2201/0052—Embedding of the watermark in the frequency domain
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2201/00—General purpose image data processing
  • G06T2201/005—Image watermarking
  • G06T2201/0053—Embedding of the watermark in the coding stream, possibly without decoding; Embedding of the watermark in the compressed domain

Definitions

• the invention relates to a video watermarking method.
• the field of the invention is the watermarking of compressed video streams.
• Video On Demand Video On Demand
• the videos are, in this application, stored and sent in compressed format (e.g MPEG2 or H264).
• the watermark is inserted just before the file is downloaded by a user who has requested it. It would then be ineffective watermark the video in its pixel form: it would need to be decompressed, watermarked and recompressed. Apart from an excessive overhead in computation time, the act of decompressing/recompressing the video adversely affects its visual quality and weakens the watermark that has been inserted.
• a stream watermarking system In addition to the universal requirements for invisibility, ruggedness and the capacity to be embedded, a stream watermarking system also mostly needs to be quick to write, and not to increase the size of the video stream, and even produce a stream of a size that is strictly identical to the original stream. This last requirement is difficult to satisfy, because any change to the image data itself can result in very major changes to the video stream and therefore, possibly, modify the size thereof. In general, the act of watermarking a stream increases its size, which causes problems when the bit rate in the networks is limited.
• the aim of the present invention is to overcome at least one of these drawbacks by proposing a stream marking system that is quick and that generates a stream whose size is close to the size of the original stream.
• the invention relates to a method of watermarking a binary stream resulting from the encoding of a set of coefficients comprising at least one coefficient encoded by predictive coding, called predictively coded coefficient.
• the method comprises the steps of: a) identifying, in the binary stream, a first binary word representing the predictively coded coefficient; b) computing a modified coefficient from the predictively coded coefficient on the basis of at least one watermarking bit; and c) encoding the modified coefficient into a second binary word; and d) replacing in the binary stream the first binary word with the second binary word.
• the watermarking method is made more robust while the size of the watermarked stream is kept close to the size of the original stream.
• the coefficients being discrete cosine transformed coefficients
• the predictively coded coefficient is of null frequency
• the modified coefficient is computed as the sum of the predictively coded coefficient and a marking value derived from the at least one watermarking bit.
• the set of coefficients comprises at least two predictively coded coefficients, a first coefficient and a second coefficient.
• the method further comprises before the identifying step a), a step for selecting one coefficient among the at least two coefficients and wherein steps a) to d) are applied on the selected coefficient.
• the selected coefficient is selected according to a watermarking key.
• the selected coefficient is selected according to a predetermined visibility criteria.
• the steps a) to d) are applied on the first coefficient and wherein the method further comprises the following steps applied on the second coefficient:
• the invention also relates to a storage unit for storing a binary stream resulting from the encoding of a set of coefficients comprising at least one predictively coded coefficient, characterized in that the storage unit further comprises a marking map, the map indicating for each predictively coded coefficient an offset value with respect to the beginning of the binary stream indicating the position of a binary word, called original binary word, representing the predictively coded coefficient and associated with this offset value a modified binary word to be inserted in replacement of the original binary word.
• the modified binary word is generated by encoding a modified coefficient, itself computed from the predictively coded coefficient on the basis of a watermarking bit.
• the invention described with DC coefficients that are encoded with differential coding according to MPEG2, may be applied on others type of coefficients that are encoded with predictive coding.
• the Intra images are coded independently, whereas the P and B images are encoded by temporal prediction based on one or two reference images.
• the reference image (unique) for a P image must be the preceding Intra image.
• For a B image the reference images is (are) the preceding or next Intra or predicted image.
• the images of different types are combined in a GOP (Group Of Pictures), each GOP containing at least one Intra image.
• Each image is divided into macroblocks of 16x16 size. Each macroblock is then subdivided into four blocks of 8x8 size. Each macroblock can be of I, P or B type. An I image will contain only I blocks, whereas a P or B image can contain l-type blocks.
• the macroblocks are then combined in slices, a slice comprising a certain number of consecutive macroblocks, all located in the same vertical position, i.e. on a same line of macroblocks.
• Each block of an Intra macroblock is DCT transformed (DCT stands for Discrete Cosine Transform) before encoding: a DC component (frequency 0,0) and 63 AC coefficients are therefore obtained for each of the blocks.
• the DC coefficient of a current block is then encoded by prediction.
• the value of the difference between this current DC coefficient and the DC coefficient of the preceding block, called preceding DC coefficient, is transmitted.
• the syntax of a block is exhibited below in an array in the form of a pseudo-code with the same conventions as in the document ISO/IEC 13818-2.
• the operator ' ! ' is the "NOT" logical operator.
• dct dc differential encodes the value of the difference between the current DC coefficient and the preceding DC coefficient. Its size is dct dc size bits which takes its value between 0 and 11 (if it is equal to 0, dct dc differential is not transmitted and the difference from the preceding block is zero). Dct dc size bits is itself encoded by a variable length code (VLC) of a size between 2 and 9 bits for the luminance. The variable length encoding tables differ according to whether it is the luminance or one of the chrominance components that is being processed.
• VLC variable length code
• the variable length encoding tables differ according to whether it is the luminance or one of the chrominance components that is being processed.
• the DC value of the current block is obtained by adding the value of the predictor, i.e. of the preceding DC coefficient, to that of the differential dct dc differential. The predictor (for the next block) then takes the value of the current DC coefficient.
• the predictors
• the reset value of the predictor depends on the parameter intra dc precision transmitted in the structure picture coding extension .
• the watermark is inserted by modifying the value of the DC coefficients of the Intra blocks. This modification can be done directly in the stream by modifying the parameter dct dc differential. If necessary, if the new value taken by dct_dc_differential after watermarking cannot be encoded on dct dc size bits, it will also be necessary to modify the latter value. In this case, the size of the stream obtained can vary after watermarking, because dct dc size is encoded by a variable length code. If there is a wish to strictly maintain the size of the stream, the coefficients dct dc differential, the modification of which also entails modifying dct dc size, can be excluded from the watermarking.
• changes to dct dc size are authorized only if the word of the variable length code obtained is of the same size. It may also happen that a slight increase of the size of the VLC word is compensated by a decrease of the number of stuffing bits. In this latter case, the size of the stream is therefore not modified.
• One advantage of the method is its simplicity of implementation. Indeed watermarking the stream merely entails modifying a very small number of parameters. These modifications are advantageously propagated (therefore making the watermarking more robust) over each modified slice due to the differential coding of DC coefficient. The watermarking operation can therefore be extremely fast.
• the method retains the size of the original stream, strictly if the size of dct dc size is controlled, or approximately otherwise.
• the method also makes it possible, by modifying a very small portion of the stream, to insert the watermark over an important portion of the corresponding image. This is due to the differential coding of DC coefficients. The same effect may be obtain with coefficients coded with predictive coding.
• the encoding of the DC coefficients is differential: a modification to the value of dct dc differential on a block leads to an effective modification of the DC on all the subsequent blocks in the slice (that is, to the right of the current block). The robustness of the watermark is thus increased.
• the watermark can be inserted on Intra images and/or on the Intra blocks of the P and B images. In the latter case, the watermark is propagated only over the Intra blocks that follow the modified block in the slice.
• an additive modulation algorithm is used to insert a watermark.
• a set of N slices S(k) is selected using a watermarking key and/or visibility criteria.
• a substitutive modulation algorithm is used to insert the watermark. This algorithm, derived from the Zhao and
• Koch algorithm works in the following way. A pair of slices S(k1 ), S(k2) is considered. S(k1 ) and/or S(k2) will then be modified so as to obtain an order relation:
• Avg(S) denotes the average of the luminance values of all the pixels contained in the slice S. It is possible, for example, to obtain this order relation as follows:
• the slice S'(k2) remains unchanged.
• the parameter dct_dc_size is then recoded if necessary.
• the order relation of the equation (eq.1 ) may also be obtained by modifying both S(k1 ) and S(k2), or only S(k2). To reduce the visibility of the watermark, it is possible to choose pairs of slices of identical (or very similar) average. With the above method, all or a very large part of the slice is modified.
• This solution has the advantage to propagate a modification, i.e. a watermark, over a slice by only modifying selectively the stream and therefrom maintain the stream size. The higher the value of x, the more restricted the scope of the modification and therefore the visibility of the inserted watermark (but the more the robustness also risks being diminished).
• d after having added d to the DC coefficient of the block B(x o ,y) (d possibly being a positive or negative), d can be subtracted from the DC of a block B(xi,y), xi>xo.
• This method makes it possible to modify the DC coefficients of a slice several times, for example to bury different watermark bits. This solution makes it possible to reduce the visibility of the watermark.
• the visibility of the watermark is controlled. Indeed, modifying the values of the parameters dct dc differential for a block can have a major impact on quality. In practice, this modification is propagated to all the subsequent blocks in the slice. Also, dct_dc_differential can take only integer values. The minimum modification amplitude possible for the DC is therefore intra_dc_mult, which depends on intra dc precision . This quantification can provoke a visible discontinuity, either in time or in space. Over a relatively uniform area of the image, the boundary between a watermarked slice and the slices just above or just below (if they have not been watermarked) can be perceptible. Similarly, if the luminance varies very little from one image to another, the modification caused by the watermark can make an annoying flickering appear.
• the visibility can be controlled by marking only the images where intra_dc_precision is greater than a given threshold. It is also possible to reduce the modification amplitude by increasing the value of intra_dc_precision by ⁇ . It will then be necessary, to retain the same dynamic, to multiply the coded values of DC by 2 ⁇ ⁇ . To do this, it is enough to increase the values of dct_dc_size by ⁇ , and to multiply by 2 (add a 0) the values of dct_dc_differential.
• a strong temporal activity in the sequence helps to mask the watermark and make it less visible.
• a temporal change located in a static sequence is very visible. Indeed, it produces a flickering effect.
• the motion compensation used in the encoding can help to make the watermark less visible. If the sequence is very static, the motion compensation is effective and a major proportion of the visual information of the I images and blocks is propagated to the subsequent P and B images. The watermark inserted on the I blocks is then itself propagated to the P and B images, which increases its robustness and reduces its visibility (less flickering).
• the average M(t) of the pixel luminance values over the areas z of the image can be considered. These areas are preferably the slices, they can also be groups of slices, blocks, macroblocks, pixels or the entire image.
• - M'(t) is the average of the pixel luminance values of the area in the image of temporal index t after said image has been watermarked and reconstructed from the MPEG stream;
• - T1 is a set of images around the Intra image.
• T1 [ti - ⁇ ;ti -l] u [t/ + l;t/ + r] , where r is a parameter linked to the number of images to take into account for computing the motion, this number of images equals 2 * ⁇ .
• r is a parameter linked to the number of images to take into account for computing the motion, this number of images equals 2 * ⁇ .
• C1 it is also possible to consider only the images preceding the Intra, or only the images following the Intra.
• Ci is not defined if the denominator is zero; an equivalent criterion is, however, obtained by comparing the numerator to Si times the denominator. This note applies for all the criteria defined hereinafter.
• T2 being the set of images around the Intra image, including the Intra image, for example
• Another criterion consists in assessing the significance of the motion in an area based on the motion vectors of the stream, for example:
• V'(t,b) being the i th motion vector of the block N° b of the image t (non Intra) (i is equal to 1 in the P images, it is equal to 1 or 2 in the B images).
• a different threshold can be used based on the actual observed visibility.
• This operation can be performed in the space domain or in the frequency domain (DCT). It is possible, for example, in the frequency domain, to calculate C4 only on the DC coefficients. It will be noted that the watermark modifies a closely related area of the image (slice or group of slices). Visually, there is therefore a risk of seeing a transition "line" appear between marked and unmarked area. Another criterion consists in comparing the amplitude of the vertical component of the gradient of the image in the transition area between marked/unmarked area, or between two differently marked areas.
• the vertical component oy can be approached, in the discrete domain, by any filter that is high pass in the vertical direction, for example:
• y is the y axis of the transition between marked/unmarked area
• x takes the set of x-axis values that are possible over the transition area.
• the graphics cards and video playback software can affect the colours displayed on the screen.
• the colours can be quantized.
• the table of the colours used for the display has only a limited number of elements, far less than the size of the overall space that is equal to 2 cn (c being the number of components and n being the number of bits per component).
• the drivers and/or graphic software perform a vector quantization of the colorimetric space, i.e. a certain number of "near" colours are replaced in the display by a single colour.
• the quantization can be performed more or less intelligently and the space can be partitioned into areas of identical or variable size (algorithms of the state of the art: uniform quantization, popularity, median cut, octree and other algorithms).
• the colour quantization can be associated with a raising of the variable contrast. These processes can affect the visibility of the watermark: if the colours are strongly quantized, over uniform areas of the image, it may be that the watermark switches the colour from one quantization level to another. In this case, the visibility of the watermark is increased.
• the effects obtained by quantization can be simulated (by applying one or more algorithms of the state of the art), and the real impact of the watermark obtained can be measured using one or more of the preceding criteria.
• the criteria can be combined with each other to obtain a finer control of the visibility. For example, they can be combined together using a weighted sum of all the criteria. In this case an area is said to be modifiable if the weighted sum for this area is below a predefined threshold. According to a variant, they can be combined by a logical operator AND, i.e. an area being modifiable only if each of the criteria are below a predefined threshold.
• the method described above may be advantageously used for the protection of new generation DVD (Blue Ray).
• a preprocessing step is applied before the watermarking step itself.
• the preprocessing step comprises the following sub-steps:
• a marking map i.e. a map identifying for each DC coefficient to be watermarked, the position of the original binary word, i.e. the original sequence of bits, encoding said DC coefficient, by an offset value with respect to the beginning of the stream and associating to each offset value the binary word to be inserted in replacement of the original binary word, the binary word to be inserted being computed according to one of the algorithm described previously, i.e. the additive modulation algorithm or the substitutive modulation algorithm.
• the preprocessing can generate a modified stream (insertion of the synchronization watermark, modification of the coding parameters) or even maintain the original stream.
• the marking map is advantageously inserted in the DVD with the original coded stream by the DVD manufacturer.
• the watermarking step itself is advantageously achieved in the DVD player by a watermarking module. When the DVD is inserted in the DVD player, the watermarking module read the marking map from it and is able to watermark directly the original stream by replacing the original words with the modified binary words read from the marking map at the appropriate location that is itself determined by the marking map using the corresponding offset values.
• the watermarking step itself is simplified thanks to the preprocessing Depending on the message to be watermarked, it will be enough to select the replacement values from the marking table for each of the areas, and to perform the substitution in the stream.
• Another application is the fingerprinting of DVDs or screeners: a unique number is watermarked for each individual copy, making it possible to identify the recipient of the DVD. In the case of piracy, reading the number makes it possible to identify the guilty user.
• Another application is video watermarking by new generation DVD players (Blue Ray type: BDP). The watermarking is done when the video is played, on the stream that is to be decoded. The watermarked message can be unique for each appliance, or even for each model or each series of appliances. The video signal output from the DVD player will therefore be watermarked differently according to this identifier. In case of massive fraud, this makes it possible to identify the suspect models and, by adjusting complementary encryption and DRM mechanisms, to revoke the models involved.

Landscapes

• Engineering & Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Theoretical Computer Science (AREA)
• Multimedia (AREA)
• Signal Processing (AREA)
• Editing Of Facsimile Originals (AREA)
• Compression Or Coding Systems Of Tv Signals (AREA)
• Image Processing (AREA)
• Television Signal Processing For Recording (AREA)

Priority Applications (5)

Applications Claiming Priority (2)

Publications (1)

Family

ID=38293199

Family Applications (1)

Country Status (6)

Cited By (3)

Families Citing this family (5)

Family Cites Families (12)

• 2007
  • 2007-04-23 JP JP2009507051A patent/JP5638239B2/ja not_active Expired - Fee Related
  • 2007-04-23 CN CN2007800151758A patent/CN101432774B/zh not_active Expired - Fee Related
  • 2007-04-23 US US12/226,585 patent/US8270485B2/en not_active Expired - Fee Related
  • 2007-04-23 WO PCT/EP2007/053950 patent/WO2007122216A1/en not_active Ceased
  • 2007-04-23 KR KR1020087024202A patent/KR101424751B1/ko not_active Expired - Fee Related
  • 2007-04-23 EP EP07728407.3A patent/EP2011075B1/en not_active Not-in-force

Non-Patent Citations (4)

Also Published As

Similar Documents

Legal Events