WO2005122080A1

WO2005122080A1 - Variance based variation of watermarking depth in a media signal

Info

Publication number: WO2005122080A1
Application number: PCT/IB2005/051745
Authority: WO
Inventors: Adriaan J. Van Leest
Original assignee: Koninklijke Philips Electronics N.V.
Priority date: 2004-06-08
Filing date: 2005-05-27
Publication date: 2005-12-22

Abstract

The present invention relates to a method, device and computer program product for embedding additional data in a media signal as well as a media signal processing device having such a device for embedding additional data. In the method a media signal divided into frames having blocks of a number of signal sample values is obtained, the variance (σ2) of at least one block of samples of a frame is determined, (step 34), and used for the embedding of additional data (w(i,j)) in said block, (step 36, 48).

Description

Variance based variation of watermarking depth in a media signal

TECHNICAL FIELD The present invention generally relates to the field of watermarking of media signals, preferably image signals for instance coded according to the MPEG coding scheme. More particularly the present invention is directed towards a method, device and computer program product for embedding additional data in a media signal as well as a media signal processing device having such a device for embedding additional data.

DESCRIPTION OF RELATED ART It is well known to watermark media signals in order to protect the rights of content owners against piracy and fraud. A watermark is here normally a pseudo-random noise code that is inserted in the media signal. In the watermarking process it is necessary that the watermark is not perceptible. A watermark that is embedded in for instance a video signal should then not be visible for an end user. It should however be possible to detect the watermark safely using a watermark detector. At the same time it is often desirable to be able to provide as much energy as possible into the watermarking process in order to provide better watermarks, which makes the watermark more robust to all kinds of signal processing. One known watermarking scheme for a video signal is described in WO 02/060182. Here a watermark is embedded in an MPEG video signal. An MPEG signal is received and comprises VLC (Variable-Length Coding) coded quantized DCT (Discrete Cosine Transform) samples of a video stream divided into frames, where each frame includes a number of blocks of pixel information. In this watermarking scheme the quantized DCT samples are obtained from the VLC coded stream and the watermark is directly embedded in this domain. A watermark is here embedded in the quantized DCT components of a block of size 8x8 under the use of a bit-rate controller, such that only the small DCT levels are modified with ±1 into a zero value. These values are furthermore only modified if the bit rate of the stream is not increased. It would however be advantageous if the watermarking according to this scheme could be improved. SUMMARY OF THE INVENTION It is therefore an object of the present invention to provide a watermarking scheme that is improved compared with the prior art. According to a first aspect of the present invention, this object is achieved by a method of embedding additional data in a media signal, comprising the steps of: obtaining a media signal divided into frames having blocks of a number of signal sample values, determining the variance of at least one block of samples of a frame, and using the variance for the embedding of additional data in said block. According to a second aspect of the present invention, this object is also achieved by a device for embedding additional data in a compressed media signal comprising an embedding unit arranged to: obtain a compressed media signal divided into frames having blocks of a number of signal sample values, determine the variance of at least one block of samples of a frame, and use the variance for the embedding of additional data in said block. According to a third aspect of the present invention, this object is also achieved by a media signal processing device comprising a device for embedding additional data according to the second aspect. According to a fourth aspect of the present invention, this object is also achieved by a computer program product for embedding additional data in a media signal, comprising computer program code, to make a computer do, when said program is loaded in the computer: obtain a media signal divided into frames having blocks of a number of signal sample values, determine the variance of at least one block of samples of a frame, and use the variance for the embedding of additional data in said block. Claim 2 is directed towards modifying the stream with additional data for the block according to a modifying scheme, where the energy of the additional data is varied at least partly in dependence of the variance. This feature enables the lowering of the perceptibility of additional data in the frame as well as the possibility to use additional data with more energy content. According to claim 3 the block is an intracoded block. These types of blocks are the blocks where the advantages mentioned in relation to claim 2 are most evident. According to claim 4 the variance is determined in the DCT domain. This is advantageous since here the number of calculations needed to be performed are relatively few. According to claim 5 a whole block is selected for embedding of additional data or not in dependence of the determined variance. This measure allows the additional data to be harder to perceive in the media signal, since the variance is a good measure of the properties of the signal in the spatial domain. Claim 6 is directed towards comparing the variance with a first threshold and selecting the block for modification if the variance is above said first threshold and refraining from modifying the block if the variance is below said threshold. According to this feature too low a variance, which indicates that additional data will be perceptible, means that no additional data is embedded in the block. Claim 7 is directed towards comparing the variance with a second threshold, and lowering the energy of the additional data in the scheme if the variance exceeds said second threshold. According to this measure sharp edges that appear in the block are considered such that the perceptibility of the additional data is lowered where these edges are provided. Claim 8 is directed towards dividing a squared non-zero sample value in the block with the variance in order to obtain a depth measure and varying the embedding depth for the sample in dependence of the size of depth measure. This feature allows the use of the variance to select embedding depth of the additional data. This allows the energy of the additional data to be varied on a coefficient-by-coefficient basis in the block. Claim 9 is directed towards comparing the depth measure with at least one third threshold and providing a lower or zero depth if the third threshold is exceeded and a higher depth if the depth measure is lower than the third threshold. This feature allows smaller coefficients to receive relatively more additional data energy than bigger coefficients. This is advantageous since larger components mask small coefficients, which small coefficients can therefore be changed more and thus the additional data receive more energy in relation to the media signal. The additional data therefore becomes more robust to different forms of attacks. Claim 10 is directed towards comparing the depth measure with a fourth higher threshold, where a depth measure under the third and fourth thresholds provides a high depth in the embedding, a depth measure above the third but below the fourth threshold provides a low depth in the embedding, and a depth measure above the third and fourth thresholds provides a depth of zero in the embedding. This feature allows a finer determination of different embedding depths in coefficients of a block. The present invention has the advantage of either lowering the perceptibility of the additional data in a media signal or the optimizing of the energy of the additional data in the signal or a combination of both. The essential idea of the invention is that the variance of the pixels in the spatial domain can be used for an indication that there is energy in the spatial domain. This information can therefore be used for controlling the embedding depth of additional data, like a watermark, such that the perceptibility of the additional data in the signal is lowered and/or such that the energy of the additional data in the signal can be optimized while still having a low perceptibility. These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

BRIEF DESCRIPTION OF THE DRAWINGS The present invention will now be explained in more detail in relation to the enclosed drawings, where Fig. 1 schematically shows a number of frames of video information in a media signal, Fig. 2 schematically shows one such frame of video information where a watermark has been provided, where the frame is divided into a number blocks, Fig. 3 shows an example of a number of luminance levels in the spatial domain for one block, Fig. 4 shows DCT levels corresponding to the luminance levels in Fig. 3 for the block, Fig. 5 shows the default intra quantizer matrix for the block in Figs. 3 and 4, Fig. 6 shows the scanning of quantized DCT coefficients for obtaining a VLC coded video signal, Fig. 7 shows a device for embedding additional data according to the present invention, Fig. 8 shows a flow chart of a method of embedding additional data according to a first embodiment of the present invention, Fig. 9 shows a flow chart of a method of embedding additional data according to a second embodiment of the present invention, and Fig. 10 schematically shows a computer program product comprising computer program code for performing the method according to the invention.

DETAILED DESCRIPTION OF EMBODIMENTS The invention is directed towards the embedding of additional data in a media signal. Such additional data is preferably a watermark. However the invention is not limited to watermarks but can be applied for other types of additional data and also to other fields of use such as transcoding of the media signal. The media signal will in the following be described in relation to a video signal and then an MPEG coded video signal. It should be realized that the invention is not limited to MPEG coding, but other types of coding can just as well be contemplated. It should furthermore be realized that the media signal is not limited to video signals, but can be applied for other types of media signals as well like still images. A video signal or stream X according to the MPEG standard is schematically shown in Fig. 1. An MPEG stream X comprises a number of transmitted frames or pictures denoted I, B and P. Fig. 1 shows a number of such frames shown one after the other. Under the frames a first line of numbers is shown, where these numbers indicate the display order, i.e. the order in which the information relating to the frames is to be displayed. Below the first line of numbers, there is shown a second line of numbers indicating the transmission and decoding order, i.e. the order in which the frames are received and decoded in order to display a video sequence. Above the frames there are shown arrows that indicate how the frames refer to each other. It should be realized that the stream also includes other information such as overhead information. The different types of frames are divided into I-, B- and P-pictures, where one such picture that is an I-picture is indicated with reference numeral 10 and is an intraframe coded picture. This picture is coded independently of other pictures and thus contains all the information necessary for displaying an image. P- and B- pictures are so called interframe coded pictures that exploit the temporal redundancy between consecutive pictures and they use motion compensation to minimize the prediction error. Only the prediction error and some overhead, like the motion vectors for pixels are here coded. P-pictures refer to one picture in the past, which previous picture can be an I-picture or a P-picture. B-pictures refers to two pictures one in the past and one in the future, where the picture referred to can be an I or a P picture. Because of this the B-picture has to be transmitted after the pictures it refers to, which leads to the transmission order being different than the display order. Each frame contains a number of pixels, where the luminance and chrominance are provided for each pixel. In the following, focus will be made on the luminance, since watermarks are embedded into this property of a pixel. Each such frame is further divided into 8x8 pixel blocks of luminance values. One such I-frame 10 is shown in Fig. 2, which shows an object 12. As an example, there is here provided twelve 8x8 pixel blocks of luminance values, where there are four such blocks in the horizontal direction and three in the vertical. Some of the blocks in the Fig.ure are furthermore watermarked, which is here indicated with the letter w in some of the blocks in question in order to show that a watermark is embedded in these blocks. It should be known that watermarks are in general not visible. One of the blocks 14 is highlighted and will be used in relation to the description of the following invention, which will be further described shortly. Fig. 3 shows an example of some luminance values y for the block indicated in Fig. 2. In the process of performing the coding of intracoded blocks a DCT (Discrete Cosine Transform) operation is performed on the blocks resulting in 8x8 blocks of DCT coefficients. Fig. 4 shows such a DCT coefficient block for the block in Fig. 3. The coefficients contain information on the horizontal and vertical spatial frequencies of the input block. The coefficient corresponding to zero horizontal and vertical frequency is called a DC component, which is the coefficient in the upper left corner of Fig. 4. Typically for natural images these coefficients are not evenly distributed, but the transformation tends to concentrate the energy to the low frequency coefficients, which are in the upper left corner of Fig. 4. Thereafter the AC coefficients in the intracoded block are quantized by applying a quantization step q * Q _ntra(m, n)/16 and in inter-coded blocks by applying the quantization step q * Q_non-i_ntra(m, n)/16. Fig. 5 shows the default quantization values Q_mt_ra used here. The quantization step q can be set differently from block to block and can vary between 1 and 112. After this quantization the coefficients in the blocks are serialized into a one- dimensional array of 64 coefficients. This serialization scheme is here a zigzag scheme as shown in Fig. 6, where the first coefficient is the DC component and the last entry represents the highest spatial frequencies in the lower corner on the right side. From the DC component to this latest component the coefficients are connected to each other in a zigzag pattern. The one-dimensional array is then compressed or entropy coded using a VLC (variable length code). This is done through providing a limited number of code words based on the array. Each code word denotes a run of zero values, i.e. the number of zero valued coefficients preceding a quantized DCT coefficient followed by a non zero coefficient of a particular level. This leads to the creation of the following line of code words for the values in Fig. 6:

(0,4),(0,7),(1,-1),(0,1),(0,-1),(0,1),(0,2),(0,1),(2,1),(0,1),(0,-1), (0,-1),(2,1),(3,1), (10,1),EOB

where EOB indicates the end of the block. These so-called run/level pairs are then converted to digital values using a suitable coding table. In this way the luminance information has been highly reduced. Intercoded blocks are treated in the same way. The difference is that for these blocks the coefficients represent prediction errors and not luminance values. As is indicated above additional information in the form of a watermark is embedded in the different blocks. A typical algorithm is the so-called run-merge algorithm described in WO-02/060182, which is herein incorporated by reference. According to this document, a watermark w, in the form of a pseudo-random noise sequence, is embedded in the blocks of a frame. A watermark is here provided as number of tiles provided over the whole image and where one tile can have the size of 128x128 pixels. The watermark tile is divided into blocks corresponding to the size of the DCT blocks and transformed into the DCT domain and these DCT blocks are then stored in a watermark buffer. In the algorithm the watermark is embedded in the quantized DCT coefficients under the control of a bit-rate controller. The watermark is embedded by adding ±1 to the smallest quantized DCT level. However since many of the signal coefficients are zero an addition of ±1 may lead to an increased bit rate, which is disadvantageous. There is furthermore a risk that the watermark will be visible. Therefore the watermark is embedded such that no modification of the signal is performed if a modification would lead to an increased bit-rate. Only the smallest quantized DCT levels ±1 are turned into a zero according to the watermark. This can be seen as:

10 if £s_n (i, j) -r w (i , ) = 0 and the budge allows if, I £In(Zj J ) otherwise , where , is the quantized input DCT level w is the watermark and /_out is the resulting watermarked quantized DCT level. When performing this type of watermarking there is however a need for a watermarking process that provides watermarks that are less visible. It is also advantageous if the watermarking energy can be varied more than previously, especially for intracoded blocks. A media processing device according to the invention is shown in a block schematic in Fig. 7. The media processing device includes a parsing unit 18, a device for embedding additional data 20 and an output stage 22. The parsing unit is connected to the device 20 as well as to the output stage 22, also the device 20 is connected to the output stage 22. The device 20 includes a first processing unit 26, connected to an embedding unit 28 and a second processing unit 30. A watermark buffer 24 is connected to the embedding unit 28. In operation the parsing unit 18 receives a media signal X in the form of a number of video images or frames including blocks with VLC coded code words. The parsing unit separates the VLC coded code words from other types of information and sends the VLC coded code words to the first processing unit 26 of device 20, which processes the stream X in order to recreate the run-level pairs of each block. The run- level pairs, i.e. the quantized DCT coefficient matrix, are then sent to the embedding unit 28, which in this way obtains this matrix. The embedding unit 28 embeds a watermark stored in the watermark buffer, which will be described in more detail later, provides the watermarked DCT matrix to the second processing unit 30, that VLC codes it and provides it to the combining unit 22 for combination with the other MPEG codes. From the combining unit 22 the watermarked signal X' is then provided. The present invention is here described in relation to intracoded blocks, because that is where the principles of the invention are applied. Intercoded blocks are normally handled as outlined in WO 02/060182, but possibly allowing higher/lower levels than ±1 of the watermark. The embedding unit 28 receives overhead information from the parsing unit 18 indicating that a block processed is an intracoded block and therefore performs the method of the invention on such a block. A method according to a first embodiment of the present invention when being applied on an I- frame, where, all the blocks are intracoded blocks, is shown in a flow chart in Fig. 8. When the embedding unit 28 is notified that an I-frame is to be received it assumes that all the blocks of the frame are to be processed according to the principles of the present invention. It then starts with setting a block counter MB to one, step 32. Thereafter the variance of the block in question is calculated, step 34. The variance is here calculated according to the formula: σ² = Ey² - (Eyf = - { yf - l(Vy){Q, 0} )

where E corresponds to the expectation, y are the luminance values of the pixels in the block in the spatial domain, <D is the unitary two-dimensional DCT transformation, c are the DCT coefficients and where Parseval's energy theorem ||y|H|Dy||is used. The variance is thus equal to the sum of all 64 squared DCT coefficients minus the squared DC component [c²(0,0)] divided by 64. Since many DCT coefficients are zero in a block due to the compression, the computation of the variance in the DCT domain is actually faster than the computation in the spatial domain. After the variance σ² has been calculated it is compared with a first user selected threshold Tl, step 36. The threshold could here have a level of for example 25. If it is below this threshold, step 36, the output DCT matrix C is set as the input DCT matrix C, step 38, i.e. no watermark is added to this block and the method proceeds with investigating if the block was the last block of the frame, step 62. If however the variance σ² was above the threshold a watermark w is added to the coefficients. In this embodiment more watermarking levels than ±1, as was described in WO-02/060182, are allowed. However it is preferred that the watermark does not raise the signal level and that a watermark is not added to zero level coefficients. It is furthermore preferred to first dequantize the coefficients and then add the watermark to the dequantized DCT components, also here so that the bit rate is not increased. In the process of actually watermarking a row counter j is first set to one, step 40, and a column counter i is set to one, step 42, where the position of row one and column one point at the DC component. This component has to be treated specially. Therefore if both the column and row counters have the value of one, step 44, a DC component handling is performed, step 46. This is handled by a DC component handling unit (not shown) in a known way. When this has been done, the embedding unit 28 continues to investigate the column counter i to see if it has reached its maximum value, step 54. If the row and columns values were not equal to one, step 44, it is investigated if the variance σ² is larger than a second user selected second threshold T2, step 48, which is much higher than Tl, like for instance around 1000. If the variance σ² was higher than the threshold T2, the watermark energy for the watermark coefficient in question is halved, step 50, whereupon watermarking of the coefficient follows, step 52. If the variance was below the second threshold T2, step 48, watermarking is performed, step 52. Watermarking is here performed in the quantized DCT domain, which means that the energy level of the DCT component c(i,j) is changed with the watermark energy level w(ij). The watermark w(i,j) for the coefficient c(i j) is taken from the watermark buffer 24, where it is stored in the DCT domain. The watermark coefficient here has a value which defines the amount and direction (i.e. the sign) that the corresponding dequantized coefficient c(i,j) is allowed to change. When the coefficient in question has been watermarked, step 52, it is investigated if the column counter i has reached its maximum value i_ma , step 54. If it has not the counter is incremented by one, step 56, and the method returns to step 44, for checking if the coefficient is the DC coefficient or not. If the counter had reached its maximum value, step 54, it is now investigated if the row counter j has reached its maximum value j_max, step 58. If it has not, step 58, the counter is incremented by one, step 60, and the embedding unit 28 returns to step 42 and sets the column counter i to one, step 42. If the row counter has reached its maximum value, step 58, it is now investigated if the block is the last block of the frame, step 62. If it was not, the block counter is incremented by one, step 64, and the method returns to step 34 and calculates the variance σ² for the next block. If it was the last block, step 62, the method is ended, step 66. This method is then repeated for further intracoded frames of the media signal. The above described method is as was mentioned earlier not limited to I- frames, but can be applied for intracoded blocks appearing also in P- and B-frames. When this is done, the steps described above related to the block counter are omitted, instead the method is started when the embedding unit is informed that the next block is an intracoded block from the parsing unit. The use of the variance and the first threshold for deciding if a block is to receive a watermark or not is of advantage because then the visibility of the watermark in the image is lowered. By using an appropriately set threshold, it is guaranteed that the energy content of the image is enough for concealing a watermark. Visibility tests have revealed that a watermark can be visible in blocks with a high variance, like a sharp edge from a dark to a bright background. By using an appropriately set second threshold, the watermark energy is reduced if the variance is high and the watermark energy is decreased. This also makes the watermark less visible in this area. It should be realized that it is possible to only compare the variance with the first threshold, in which case it is possible to combine this feature with watermarking according to the principles described in WO 02/060182, i.e. by adding ±1 to the lowest non-zero coefficients if that does not raise the bit rate. In the method described above the watermarking energy of all watermarking coefficients were reduced if the second threshold was exceeded that is the amount of change of the dequantized DCT level is lowered for all the coefficients of the block. If the variance is below this second threshold the amount is kept as originally intended. It is furthermore possible to have more such thresholds and make the embedding depth depend on the variance, where the depth can vary from zero, i.e. no watermark, to a maximum embedding strength. The embedding depth on a block-by-block basis is thus made dependent on the variance. Another variation of the present invention is shown in a flow chart in Fig. 9, which uses the variance for controlling the watermarking energy on a coefficient-by- coefficient basis in a block. The method allows optimizing the embedding strength on a coefficient-by-coefficient basis. The method here has a number of steps 68 - 94 that in this embodiment replace the steps 40 - 60 in the method of Fig. 8. The other steps provided before and after steps 40 -58 in Fig. 8 are also performed here, i.e. a block counter is set and checked and the variance for the block is calculated. The method is first started at step 68. Then row j and column i counters are set, steps 70 and 72, DC-levels are investigated and handled, steps 74, 76, followed by watermarking steps 77 - 88 to be described in more detail below, counters incremented and checked for maximum values, steps 90, 92, 94 and 96, and the ending of the method, step 98. Steps 70 - 76 and 90 - 96 are performed in the same way as was described in relation to Fig. 8, consequently they will not be further described here. In the method according to Fig. 9 it is now investigated if the AC component c(i,j) is zero or not, step 77. If it is the method continues to step 88, where no watermark is added. If it is not zero a depth measure R(i,j) is calculated for each non-zero AC component such that R(i,j) = c²(i,j)/σ², step 78. This depth measure is then compared with a third threshold T3, and if the depth measure is below this third threshold T3, step 80, the DCT coefficient c(ij) is watermarked with the watermark w(i,j) in normal fashion, step 82. If however the depth measure is larger than the third threshold T3, step 80, it is then compared with a fourth larger threshold T4, step 84. If the depth measure is here below the fourth threshold T4 but still above the third threshold T3, step 84, the energy of the watermark is decreased, and in this example it is set such that the level change is halved, step 86. If, however, the depth measure R(i,j) was above also the fourth threshold T4, step 84, no watermark is added and thus the level change is equal to zero, step 88. With this method it is possible to vary the watermarking energy on a coefficient-by-coefficient basis depending on the size of the coefficient and the variance. This is advantageous since relatively large coefficients actually determine the basic shape of the pixels in the spatial domain and thus mask the relatively small coefficients. This fact is here used such that the small coefficients receive a high degree of watermarking energy, medium ranged coefficients receive a somewhat weighted degree of watermarking and large coefficients receive no watermarks. The watermarking energy is thus high for relatively small coefficients and small compared to the coefficient for bigger coefficients. In this way relatively more energy is allowed to be provided in the watermark, which is beneficial since then the watermark is more robust to different forms of signal processing attacks. This strategy also works well for pictures having white letters on a black background. It should be realized that the second embodiment described above is not limited to two thresholds, but can in its simplest form have only one threshold, where a first amount of watermarking energy is provided below said threshold and a second lower energy, which can be zero, is provided above said threshold. It should also be realized that more thresholds could be provided for providing different intervals where the watermarking energy is varied by a certain amount. Another possible variation is that the watermarking energy is increased by a certain amount below a threshold and kept at another amount above said threshold. In the previous description the watermarking was described for two different embodiments using calculations of the variance for varying watermarking on a block-by- block basis as well as by a coefficient-by-coefficient basis within a block. It should furthermore be realized that these two embodiments can be combined in one and the same embedding unit. The watermarking was above described in relation to a block in a video frame. It should be realized that it can just as well be used for still images. The present invention has been described in relation to a watermark embedding unit. This embedding unit is preferably provided in the form of one of more processors containing program code for performing the method according to the present invention. This program code can also be provided on a computer program medium, like a CD ROM 100, which is generally shown in Fig. 10. The method according to the invention is then performed when the CD ROM is loaded in a computer. The program code can furthermore be downloaded from a server, for example via the Internet. It should be emphasized that the term "comprises/comprising" when used in this specification is taken to specify the presence of stated features, integers, steps or components, but does not preclude the presence or addition of one or more other features, integers, steps, components or groups thereof. It should furthermore be realized that reference signs appearing in the claims should in no way be construed as limiting the scope of the present invention.

Claims

CLAIMS:

1. Method of embedding additional data in a media signal, comprising the steps of: obtaining a media signal (X) divided into frames (10) having blocks (14) of a number of signal sample values, determining the variance (σ²) of at least one block of samples of a frame, (step 34), and using the variance for the embedding of additional data (w(i,j)) in said block, (step 36, 48; 78).

2. Method according to claim 1, wherein the step of using comprises modifying the stream with additional data for said block according to a modifying scheme, where the energy of the additional data is varied at least partly in dependence of the variance.

3. Method according to claim 1, wherein the block is an intracoded block.

4. Method according to claim 1, wherein the variance is determined in the DCT domain.

5. Method according to claim 1, wherein the step of using comprises selecting to modify the signal samples of the block with the additional data in dependence of the determined variance, (step 36), and modifying the stream with additional data for the block according to a modifying scheme, (step 52).

6. Method according to claim 5, further comprising the step of comparing the variance with a first threshold (Tl), (step 36), selecting the block for modification if the variance is above said first threshold, (step 52), and refraining from modifying the block if the variance is below said threshold, (step 38).

7. Method according to claim 6, further comprising the step of comparing the variance with a second threshold (T2), (step 48), and lowering the energy of the additional data in the scheme (step 50) if the variance exceeds said second threshold.

8. Method according to claim 1, further comprising the step of dividing a squared non-zero sample value in said block by the variance, (step 78), in order to obtain a depth measure (R(i,j)), and the step of modifying comprises varying the embedding depth for the sample (c(i,j)) in dependence of the size of depth measure, (step 80,82, 84, 86, 88).

9. Method according to claim 8, further comprising the step of comparing the depth measure with at least one third threshold (T3), (step 80), and providing a lower (step 86) or zero depth (step 88) if the third threshold is exceeded and a higher depth (step 82) if the depth measure is lower than the third threshold.

10. Method according to claim 9, wherein the depth measure is further compared with a fourth higher threshold (T4), (step 84), where a depth measure under the third and fourth thresholds provides a high depth in the embedding, (step 82), a depth measure above the third but below the fourth threshold, (step 86), provides, a low depth in the embedding, and a depth measure above the third and fourth thresholds provides a depth of zero in the embedding, (step 88).

11. Device for embedding additional data in a media signal, comprising an embedding unit (28) arranged to: obtain a media signal (X) divided into frames (10) having blocks (14) of a number of sample values, determine the variance (σ²) of at least one block of samples (14) of a frame, and use the variance for the embedding of additional data in said block.

12. Device according to claim 11, further comprising a watermark buffer (24) comprising a watermark (w) to be embedded in the media signal.

13. Media signal processing device comprising a device for embedding additional data according to claim 11.

14. Computer program product (100) for embedding additional data in a media signal, comprising computer program code, to make a computer do, when said program is loaded in the computer: obtain a media signal divided into frames having blocks of a number of signal sample values, determine the variance of at least one block of samples of a frame, and use the variance for the embedding of additional data in said block.