GB2369951A - Detecting and recovering embedded data using correlation and shifting - Google Patents

Detecting and recovering embedded data using correlation and shifting Download PDF

Info

Publication number
GB2369951A
GB2369951A GB0029867A GB0029867A GB2369951A GB 2369951 A GB2369951 A GB 2369951A GB 0029867 A GB0029867 A GB 0029867A GB 0029867 A GB0029867 A GB 0029867A GB 2369951 A GB2369951 A GB 2369951A
Authority
GB
United Kingdom
Prior art keywords
data
transform domain
symbols
transform
correlation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
GB0029867A
Other versions
GB0029867D0 (en
Inventor
Jason Charles Pelly
Stephen Mark Keating
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sony Europe Ltd
Original Assignee
Sony United Kingdom Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sony United Kingdom Ltd filed Critical Sony United Kingdom Ltd
Priority to GB0029867A priority Critical patent/GB2369951A/en
Publication of GB0029867D0 publication Critical patent/GB0029867D0/en
Priority to EP01310175A priority patent/EP1215908A3/en
Priority to US10/007,085 priority patent/US7085395B2/en
Priority to JP2001375051A priority patent/JP4094842B2/en
Publication of GB2369951A publication Critical patent/GB2369951A/en
Withdrawn legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00General purpose image data processing
    • G06T1/0021Image watermarking
    • G06T1/005Robust watermarking, e.g. average attack or collusion attack resistant
    • G06T1/0064Geometric transfor invariant watermarking, e.g. affine transform invariant
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/46Embedding additional information in the video signal during the compression process
    • H04N19/467Embedding additional information in the video signal during the compression process characterised by the embedded information being invisible, e.g. watermarking
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/20Servers specifically adapted for the distribution of content, e.g. VOD servers; Operations thereof
    • H04N21/23Processing of content or additional data; Elementary server operations; Server middleware
    • H04N21/238Interfacing the downstream path of the transmission network, e.g. adapting the transmission rate of a video stream to network bandwidth; Processing of multiplex streams
    • H04N21/2389Multiplex stream processing, e.g. multiplex stream encrypting
    • H04N21/23892Multiplex stream processing, e.g. multiplex stream encrypting involving embedding information at multiplex stream level, e.g. embedding a watermark at packet level
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2201/00General purpose image data processing
    • G06T2201/005Image watermarking
    • G06T2201/0052Embedding of the watermark in the frequency domain

Landscapes

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

Abstract

An apparatus for detecting and recovering data embedded in information material e.g. watermarks, the data embedded in the material in a transform domain representation by using the data to modulate a predetermined data sequence to form modulated data and combining the modulated data with the material. The apparatus comprising a transform processor operable to transform the material into the transform domain representation, and a correlation processor 334 operable to correlate transform domain data symbols bearing the modulated data from the transform domain representation with a reproduced version of the predetermined data sequence to form a correlation output signal, and to recover the embedded data from the correlation output signal. The correlation processor 334 is operable to repeat the correlation for transform domain data symbols and data symbols of the predetermined data sequence for a plurality of start position (Figure 10) in transform domain, the start positions representing the transform domain data. If the shift of transform domain data represents a loss of transform domain symbols, corresponding symbols from the predetermined data sequence are omitted and the lost transform domain data symbols and corresponding symbols of the predetermined data sequence are not included in calculating the correlation output signal.

Description

APPARATUS FOR DETECTING AND RECOVERING DATA
Field of Invention The present invention relates to methods and apparatus for detecting and recovering data embedded in information material.
Information material as used herein refers to and includes one or more of video material. audio material and data material. Video material in this context may be still
... images or moving images.
Background of Invention Steganography is a technical field relating to the embedding of data into material such as video material, audio material and data material in such a way that the data is imperceptible in the material.
Watermarks are data embedded in material such as video material, audio material and data material. A watermark may be imperceptible or perceptible in the material.
A watermark may be used for various purposes. It is known to use watermarks for the purpose of protecting the material against, or trace, infringement of the intellectual property rights of the owner (s) of the material. For example a watermark may identify the owner of the material.
Watermarks may be"robust"in that they are difficult to remove from the material. Robust watermarks are useful to trace the provenance of material which is processed in some way either in an attempt to remove the mark or to effect legitimate processing such as video editing or compression for storage and/or transmission.
Watermarks may be"fragile"in that they are easily damaged by processing which is useful to detect attempts to remove the mark or process the material.
Visible watermarks are useful to allow, for example, a customer to view an image via, for example, the Internet to determine whether they wish to buy it but without allowing the customer access to the unmarked image they would buy. The watermark degrades the image and the mark is preferably not removable by the customer. Visible watermarks are also used to determine the provenance of the material into which they are embedded.
In US patent 5, 930, 369 (Cox et al) it has been proposed to embed data into material such as images to form a watermark by converting the material into the transform domain and adding the data to the image in the transform domain. For the example of images and the Discrete Wavelet Transform of these images into the transform domain. the data to be added can be combined with the wavelet coefficients of one of a plurality of sub-bands which are formed in the transform domain.
Generally, the data to be embedded is arranged to modulate a predetermined data sequence such as a Pseudo Random Bit Sequence (PRBS). For example, each bit of the data to be embedded is arranged to modulate a copy of the PRBS, and this copy is then added, for example into one of the sub-bands of the image in the transform domain. The image is then converted back to the spatial domain.
If it is desired to detect and recover the embedded data from the image, the image is converted back to the transform domain and the embedded data is recovered from the sub-band in the transform domain by correlating the transform coefficients in the sub-band with the predetermined data sequence which is known to the detecting apparatus. Data sequences such as PRBSs have properties which provide a high autocorrelation result when the sequence coincides with itself and a low or ideally zero correlation result at all other shifts of the sequence with respect to itself. As such, the embedded data can be recovered from the result of correlating the sub-band data with the predetermined data sequence.
Summarv of Invention According to the present invention there is provided an apparatus for detecting and recovering data embedded in information material, the data having been embedded in the material using a transform domain representation of at least one of the data and information material by arranging for the data to modulate a predetermined data sequence to form modulated data and combining the modulated data with the material, the apparatus comprising a transform processor operable to transform the material into the transform domain representation of said information material, and a correlation processor operable to correlate transform domain data symbols bearing the modulated data from the transform domain representation with a reproduced version of the predetermined data sequence to form a correlation output signal and to recover the
embedded data from the correlation output signal, wherein the correlation processor is operable to perform the correlation of transform domain data symbols with data symbols of the predetermined data sequence for a plurality of start positions in the transform domain, the start positions representing at least one relative possible shift of the transform domain data, and if the shift of the transform data represents a loss or corruption of transform domain data symbols, omitting corresponding symbols from the predetermined data sequence in said correlation, the lost or corrupted transform domain data symbols and the corresponding symbols of the predetermined data sequence not being included in calculating the correlation output signal.
Information material such as images are processed in order to convert them from one form to another, so that, for example the images can be compression encoded and decoded. A result of processing the image may be to introduce a shift in the image in the transform domain or the spatial domain with an effect that some of the transform domain data symbols may be lost or corrupted. Furthermore, if the data in the transform domain has shifted from its original position, it may not be possible to detect the embedded data from the correlation of the transform domain symbols with the reproduced predetermined data sequence, because the predetermined data sequence will no longer align with the modulated version added to the transform domain. The amount of the shift will be unknown.
Generally, a previous proposal has been to add the data to be embedded to the transform domain in a central region only, leaving a number of pixels at either edge where data is not added. This previous proposal overcomes the disadvantage that the data at the edges of the image may be lost or corrupted if the image is shifted. However this is at the expense of a reduction in the capacity for embedding data for each transform domain image.
Embodiments of the present invention can utilise the entire transform domain providing an increased data embedding capacity. In order to detect and recover this data, the data is added to the transform domain in combination with a predetermined data sequence, each data symbol of the data to be embedded being arranged to modulate a copy of the predetermined data sequence. In the apparatus for detecting and recovering the data, each of the embedded data symbols is recovered by correlating the data symbols in the transform domain with a reproduced version of the
predetermined data sequence. The correlation is repeated for each of a plurality of start positions within the transform domain. at least one of the start positions representing possible shifts of the transform domain data symbols. The embedded data may be recovered from the correlation output signal. in spite of a shift of the transform domain data, provided that the start positions of the correlation include a start position representing the actual shift which has been introduced into the transform domain data symbols.
Where it is possible that the shift of the data within the transform domain corresponding to a particular start position is representative of a loss or corruption of transform domain data symbols. for example at the edges of the transform domain, symbols from the predetermined data sequence corresponding to the lost or corrupted transform domain symbols are omitted from the correlation, the lost or corrupted transform domain data symbols and the corresponding symbols of the predetermined data sequence are therefore not included in calculating the correlation output signal.
The number of transform domain symbols lost is therefore also the number of the corresponding symbols discounted from the reproduced version of the predetermined sequence, thus maintaining the correspondence between the symbols of the correlating version of the predetermined sequence and the symbols of the predetermined sequence which form the modulated data.
As will be appreciated therefore, in preferred embodiments the plurality of start positions includes representations of a plurality of possible shifts of the transform domain symbols from an original position in which the modulated data was combined with the symbols.
A shift of the information material can cause a loss of symbols from one edge of the transform domain representation and also cause symbols at the opposite edge to be corrupted. Therefore, when performing a correlation which would otherwise have included these symbols, the corresponding symbols of the reproduced data symbols are excluded from the correlation sum. By discounting these symbols when calculating the correlation a potentially disrupting effect of these possibly corrupted or lost data symbols is avoided, whilst providing correspondingly a facility for utilising the entire data bearing capacity in the transform domain for embedding data.
Although a shift of the information material by a number of data symbols may cause a shift of a corresponding number of data symbols in the transform domain, some transforms may result in a different shift being produced in the transform domain than the un-transformed domain of the information material. Therefore, a shift of the information material by a first number of data symbols, may produce a shift by a second number of transform domain symbols. For this reason, the apparatus may comprise a control processor operable to shift the information material by an amount determined in accordance with the first and second numbers to the effect that the plurality of start positions of said correlation of said predetermined data sequence with the transform domain data symbols can include an integer shift of the information material in the un-transformed domain. By pre-shifting the information material before being transformed into the transform domain, the start positions corresponding to different shifts of numbers of transform domain data symbols can be made to corespond to integer shifts of information material data symbols.
Considering the example of information material such as images which are transformed using the discrete wavelet transform into the wavelet domain, a shift by a two pixels of the image will cause a shift of one wavelet coefficient in the wavelet transform domain. As such a shift of 0, 1, 2 and 3 wavelet coefficients for a level one wavelet corresponds to a shift of 0,2, 4 and 6 pixels of the image. However, it is possible that the image in the spatial domain may be shifted by only a single pixel. As such, shifting the start position of the correlation in the transform domain by a single wavelet coefficient will not correspond to the actual shift of the transform domain image, and so the correlation output signal may not be large enough for the data to be recovered. However by pre-shifting the image by one pixel, the transform domain shifts of 0, 1, 2 and 3 wavelet coefficients will correspond to 1, 3,5 and 7 pixels, which include the actual shift introduced into the image.
In preferred embodiments the transform domain may have a plurality of subbands, the modulated data being added to at least one of the sub-bands, the correlation processor being operable to correlate the modulated data with the predetermined data sequence except the excluded data symbols.
Although it will be understood that any suitable transform may be used to transform the image into sub-bands, in preferred embodiments, the transform is the
Discrete Wavelet Transform (DWT), the data symbols in each of the sub-bands comprising wavelet coefficients.
Although embodiments of the invention find application in detecting and recovering data from any information material, a particular application of the invention is in detecting and recovering data embedded in video image or audio signals.
Various further aspects and features of the present invention are defined in the appended.
Brief Description of the Drawings Figure I is a schematic block diagram of a watermarking system ;
Figure 2 is a schematic block diagram of a watermark embedder appearing in zn Figure 1 ; Figure 3 is a schematic representation illustrating the form of the wavelet domain transform; Figure 4 is a schematic representation of an iterative filtering process which can be used to form the wavelet transform; Figure 5 is a schematic block diagram of a combiner forming part of the watermark embedder shown in Figure 2; Figure 6 provides an illustrative representation of a transform domain image in which data is added in accordance with a previous proposal ; Figure 7 provides an illustrative representation of a transform domain image in which data is added by the combiner shown in Figure 5 ; Figure 8 is a schematic block diagram of a watermark decoder appearing in Figure 1 ; Figure 9 is a schematic block diagram of a correlator forming part of the watermark decoder shown in Figure 8; Figure 10 provides an illustrative representation of a transform domain image from which embedded data is recovered by the correlator of Figure 9; Figure 11 is a schematic representation of the operation of the control processor and correlation processor forming part of the correlator of Figure 9, to detect and recover data from the transform domain image of Figure 10 in the presence of shifts of the image;
Figure 12 provides a graphical illustration of a correlation output signal with respect to start positions of the correlation ; Figure 13 provides a schematic illustration of the correlation performed by the correlation processor within the correlator of Figure 9 : and Figures 14A and 14B are schematic block diagrams of the structure of an extended and a basic UMID respectively.
Description of Preferred Embodiments An example embodiment of the present invention will be described with reference to a watermarking system in which data is embedded into a video image. Any type of data can be embedded into the image. However. advantageously the data embedded into the image may be meta data which describes the image or identifies some attributes of the content of the image itself. An example of meta data is the Universal Material Identifier (UMID). A proposed structure for the UMID is disclosed in SMPTE Journal March 2000. A more detailed explanation of the structure of the UMID will be described later.
Watermarking System Figure I illustrates a watermarking system. generally 10, for embedding a watermark into a video image 115, and recovering and removing a watermark from the watermarked image 125. The watermarking system 10 in Figure 1 comprises an image processor 100 for embedding the watermark into the video image, and a decoding image processor 102 for detecting and recovering the watermark, and for removing or 'washing'the watermark from the video image.
The image processor 100 for embedding the watermark into the video image comprises a strength adapter 180, and a watermark embedder 120. The watermark embedder 120 is arranged to embed the watermark into the video image 115, produced from the source 110, to form a watermarked image 125. The watermark to be embedded into the video image is formed from data 175 representing a UMID.
Generally, the UMID identifies the content of the video image, although it will be appreciated that other types of meta data which identify the content or other attributes of the image can be used to form the watermark. In preferred embodiments the watermark embedder 120 embeds the UMID into the video image 115 in accordance
with a particular application strength 185 from the strength adapter 180. The strength adapter 180 determines the magnitude of the watermark in relation to the video image 115. the application strength being determined such that the watermark may be recovered whilst minimising any effects which may be perceivable to a viewer of the watermarked image 125. After embedding the watermark, the image may be transmitted, stored or further processed in some way. such as for example. compression encoding the image. This subsequent processing and transmitting is represented generally in Figure 1 as line 122.
In Figure 1 the decoding image processor 102 for detecting and removing the watermark is shown as comprising a watermark decoder 140. a data store 150 and a watermark washer 130 which removes the watermark from the watermarked image 125.
The watermark decoder 140 detects the watermark from the watermarked video image and in the present example embodiment, generates a restored UMID 145 from the watermarked image 125. The watermark washer 130 generates a restored image
135, by removing as far as possible the watermark from the watermarked image 125.
17, ZD In some embodiments, the watermark washer 130 is operable to remove the watermark from the image substantially without leaving a trace. The restored image 135 may then be stored in a store 150, transmitted or routed for further processing.
The Watermark Embedder The watermark embedder will now be described in more detail with reference to Figure 2, where parts also appearing in Figure 1 have the same numerical references. In Figure 2 the watermark embedder 120 comprises a pseudo-random sequence generator 220, an error correction encoder 200, a wavelet transformer 210, an inverse wavelet transformer 250, a modulator 230 and a combiner 240.
The error correction encoder 200 receives the UMID 175 and generates an error correction encoded UMID comprising redundant data in combination with the UMID, in accordance with an error correction encoding scheme. It will be appreciated that various error correction coding schemes could be used to encode the UMID. The UMID in the example embodiment is a binary sequence of 31 bytes. Applying for example, a Bose-Chaudhuri-Hocquenghem (BCH) code, the 31-bytes are converted to
a 511-bit encoded binary sequence. Such a BCH code word can correct up to 31-bit errors, and detect a greater number of errors, which may be clusters or bursts of errors.
As shown in Figure 2 the error correction encoded UMID 205 is received at a first input to the modulator 230. The pseudo-random sequence generator 220 outputs a PRBS 225 which is received at a second input to the modulator 230. The modulator 230 is operable to modulate each copy of a PRBS. generated by the pseudo-random sequence generator 220, with each bit of the error correction encoded UMID. In
preferred embodiments the PRBS is modulated by representing the values of each bit 1 Z7 of the PRBS in bipolar form ('1'as +1, and'0'as-1) and then reversing the polarity of each bit of the PRBS, if the corresponding bit of the encoded UMID is a'0'and not reversing the polarity if the corresponding bit is a'1'. The modulated PRBS is then received at a first input of the combiner 240. The combiner 240 receives at a second input the image in which the PRBS modulated data is to be embedded. However the data is combined with the image in the transform domain.
The use of a pseudo-random sequence 225 to generate the spread spectrum signal representing the watermark data allows a reduction to be made in the strength of the data to be embedded in the image. By correlating the data in the transform domain image to which the modulated PRBS has been added, a correlation output signal is produced with a so called correlation coding gain which allows the modulated data bit to be detected and determined. As such, the strength of the data added to the image can be reduced, thereby reducing any perceivable effect on the spatial domain image.
The use of a spread spectrum signal also provides an inherent improvement in robustness of the image because the data is spread across a larger number of transform domain data symbols.
As shown in Figure 2, the wavelet transformer 210 receives the video image 115 from the source 110 and outputs a wavelet image 215 to the combiner 240. The image is thus converted from the spatial to the transform domain. The combiner 240 is operable to add the PRBS modulated data to the image in the transform domain, in accordance with the application strength, provided by the strength adapter 180. The watermarked wavelet image 245 is then transformed into the spatial domain by the inverse wavelet transformer 250 to produce the watermarked image 125. The operation of the combiner 240 will be explained in more detail shortly.
In order to explain the embodiment of the present invention a brief description of wavelet transforms will be provided in the following paragraphs with reference to Figures 3 and 4.
The Wavelet Transform Wavelets are well known and are described in for example"A Really Friendly Guide to Wavelets"by C Valens. 1999 (c. valens@mindless.com). Valens shows that the discrete wavelet transform can be implemented as an iterated filter bank, as used in sub-band coding, with scaling of the image by a factor of 2 at each iteration.
Thus, referring to Figure 3 a spatial domain image is applied to a set of high pass HP and low pass LP filters. At level 1, the first stage of filtering, the image is filtered horizontally and vertically and, in each direction, scaled down by a factor of 2.
In level 2, the low pass image from level I is filtered and scaled in the same way as in z level 1. The filtering and scaling may be repeated in subsequent levels 3 onwards.
The result is shown schematically in Figure 4. Figure 4 is a representation normal in the art. The horizontal axis H indicates increasing frequency of the subbands. At level one the image is spatially filtered into four bands; the lower horizontal and vertical band, ! HI, IV, ; the upper horizontal band hH,, IV, ; the upper vertical band It, havi ; and the upper horizontal and vertical band, hH ;, hV ;. At level 2, the lower
horizontal and vertical band, IH I, IV I is filtered and scaled into the lower horizontal and vertical band, ! Hz, IV z ; the upper horizontal band hH IV2 ; the upper vertical band IH2, hV2 ; and the upper horizontal and vertical band, hl-b, hV2 At level 3 (not shown in Figure 4), the lower horizontal and vertical band, IH2, 1V2 is further filtered and scaled.
Combiner The operation of the combiner 240 will now be explained in more detail. The combiner 240 receives the wavelet image 215 from the wavelet transformer 210, and the modulated PRBS from the modulator 230 and the application strength 185 from the
strength adapter 180. The combiner 240 embeds the watermark 235 onto the wavelet image 215, by adding, for each bit of the modulated PRBS a factor a scaled by 1, in dependence upon the value of the bit. Selected parts of the wavelet image 215 are
used to embed the watermark 235. Each coefficient of the predetermined region of the wavelet image 215 is encoded according to the following equation : t : l =. Y, + (1) Where Xi is the i-th wavelet coefficient, an is the strength for the n-th PRBS and Wi is the i-th bit of the modulated PRBS in bipolar form.
An example of the combiner and the operation of the combiner will now be described with reference to Figures 5,6 and 7. In Figure 5 the combiner 240 is shown to receive the transform domain image from the connecting channel 215 which provides the transform domain image to a frame store 236. The frame store 236 is arranged to store a frame of transform domain data. The combiner 240 is also arranged to receive the spread spectrum encoded and error correction encoded UMID after it has been spread using the PRBS (modulated PRBS data). For this example embodiment one UMID in this error correction and spread spectrum encoded form is to be embedded in the frame of image data within the frame store 236. Thus, each encoded UMID forms an item of data which is to be embedded into each frame of image data. To this end, the frame store stores a frame of data representing the image in the wavelet transform domain. The data to be embedded is received at a combining processor 237 which combines the data to be embedded into selected parts of the wavelet transform domain image stored in the frame store 236. The combiner 240 is also provided with a control processor 238 which is coupled to the combining processor 237.
In Figure 6 an illustrative representation of a first order wavelet transform is presented. This wavelet transform is representative of a frame of the image
transformed into the wavelet domain and stored in the frame store 236. The wavelet transform image WT~IMG is shown to comprise four wavelet domains representative of sub-bands into which the image has been divided. The wavelets have been labelled in correspondence with the illustration provided in Figure 3. In the example embodiment of the present invention, the data to be embedded is only written into the low vertical, high horizontal frequencies sub-band HHIIVI and the low horizontal, high vertical frequencies sub-bands labelled IHhVl.
By embedding the data in only the two sub-bands HHIIVI, IHIHVI, the likelihood of detecting the embedded data is improved whilst the effects that the embedded data will have on the resulting image are reduced. This is because the wavelet coefficients of the high horizontal, high vertical frequencies sub-bands hHihV} are more likely to be disturbed, by for example compression encoding.
Compression encoding processes such as JPEG operate to compression encode images by reducing the high frequency components of the image. Therefore, writing the data into this sub-band hHjhV} would reduce the likelihood of being able to recover the embedded data. Conversely, data is also not written into the low vertical, low horizontal frequencies sub-band IHIIVI. This is because writing data into this subband would have a more disturbing effect on the image, since the low frequency components of the image have generally lower energy than the higher frequency components. Therefore, writing the data in the low vertical, low horizontal frequencies sub-band would have a more disturbing effect on the image. As a compromise the data is added into the high horizontal, low vertical frequencies sub band hHV and the low horizontal, high vertical frequencies sub-bands IHIHVI.
As already explained, an effect of processing the image in the spatial or the transform domains is to shift the wavelet coefficients in the sub-band with an effect that coefficients at the edges of the sub-band could be lost or corrupted. Thus the subband includes a region in which the transform domain data symbols (wavelet coefficients) may not be recoverable. To illustrate this problem Figure 6 shows a previously proposed arrangement in which the embedded data unit is not written into or added with wavelet coefficients at the edges of the sub-band EXSYS. However an effect of this is to reduce the data bearing capacity of the sub-band. Conversely this can be considered as representing a reduction in the error correction capacity which can be added to the embedded data.
An illustration of the operation of the combiner 240 in accordance with an embodiment of the present invention is illustrated in Figure 7. Figure 7 corresponds to Figure 6 and so like parts are provided with corresponding reference numerals. In contrast to Figure 6, the combiner has used the entire width of the sub-band to embed the data. Thus as illustrated the entire sub-band comprising all wavelet coefficients
between the edges of the sub-band are used to convey data. To this end, the spread spectrum encoded data is added to the wavelet coefficients as illustrated by the boxes P~SEQ. However, as represented by a region EXSYS between the edge of the subband and a broken line. in common with the arrangements shown in Figure 7, the data symbols in the region designated EXSYS at the edges of the sub-band will have wavelet coefficients which may be lost or corrupted, due to processing or shifting the image. Embodiments of the present invention address the problem of recovering the embedded data in spite of the loss or corruption of the data symbols within the region EXSYS at the edges of the sub-band. The detection and recovery of the embedded data is performed by the watermark decoder 140 forming part of the decoding image processor 102 shown in Figure 1.
Decoder The operation of the watermark decoder 140 in the decoding image processor, will now be explained in more detail, with reference to Figure 8, where parts also appearing in Figure 1, bear identical reference numerals. The watermark decoder 140 receives the watermarked image 125 and outputs a restored version of the UMID 145. The watermark decoder 140 comprises a wavelet transformer 310, a pseudo-random sequence generator 320, a correlator 330, and an error correction decoder 350.
The wavelet transformer 310 converts the watermarked image 125 into the transform domain so that the watermark data can be recovered. The wavelet coefficients to which the PRBS modulated data were added by the combiner 240 are then read from the two wavelet sub-bands hHlV, IHhV These wavelet coefficients are then correlated with respect to the corresponding PRBS used in the watermark embedder. Generally, this correlation is expressed as equation (2), below, where Xn is the n-th wavelet coefficient and Rn is the n-th bit of the PRBS generated by the Pseudo Random Sequence Generator 320. However as will be explained shortly, not all the coefficients from the wavelet may be used to form the correlation sum according to equation (2).
The relative sign of the result of the correlation Cn then gives an indication of the value of the n-th bit of the embed data in correspondence with the sign used to represent this bit in the watermark embedder. The data bits recovered in this way represent the error correction encoded UMID which is subsequently decoded by the error correction decoder 350 using a decoding algorithm for the error correction code used by the encoder 200. Having recovered the UMID. the watermark can be removed from the video image by the watermark washer 130, by performing the reverse of the
operations performed by the embedder. The correlator 330 appearing in Figure 8 is I I is shown in more detail in Figure 9, wherein parts also appearing in Figure 8 bear identical numerical references.
In Figure 9 the correlator 330 is shown to comprise a frame store 331 which is arranged to receive the wavelet transformed image via the connecting channel 315.
The frame store 331 is arranged to store a frame of wavelet transformed image. A
control processor 332 is arranged to select wavelet coefficients from the wavelet domain sub-band and to feed these selected coefficients to a correlation processor 334. The control processor 332 also controls a buffer 333. The selected wavelet coefficients will be a consecutive set of coefficients to which one modulated PRBS is assumed to have been added. The buffer receives the PRBS from the PRBS generator 320 and feeds this PRBS under control of the control processor to the correlation processor 334.
The control processor 332 controls the correlation of the data symbols (wavelet coefficients) read from the sub-band of the image in the frame store 331 with the PRBS in accordance with equation (2). However as explained, the image may have suffered a shift in the spatial domain, with the effect that a corresponding shift has occurred in the wavelet domain. For the example of the wavelet transform, a shift of two pixels of the image in the spatial domain will cause a shift of one wavelet
coefficient for a level one wavelet. Therefore, a shift of 0, 1, 2, 3 wavlet coefficients is representative of a shift of 0, 2, 4, and 6 pixels of the image in the spatial domain. However, the control processor is arranged to accommodate possible shifts of the image, so that the PRBS in the correlator can be aligned with the modulated PRBS added to the wavelet coefficients. The operation of the correlator 330 in order to recover the embedded data is illustrated in Figures 10,11, 12 and 13.
In Figure 10, the wavelet transformed image WT~IMG'as produced by the wavelet transformer 310 is shown in the form as it would appear as an illustrative example in the frame store 331. The representation shown in Figure 10 corresponds to the representation shown in Figure 7. For simplicity of explanation. only the high horizontal, low vertical frequencies sub-band will considered. although it will be appreciated that the data will be recovered from the high vertical. low horizontal frequencies sub-band hH I IV I in a similar way. As illustrated the image has suffered a shift to the left SHFT, with the effect that the wavelet coefficients LST at the left hand
edge of the sub-band have been lost. z : l As can be seen in Figure 10 data symbols from the left hand edge of the subband have been lost whereas at the right hand edge of the sub-band, the data symbols are corrupted or at least unreliable as a result of the shift. This is because the right hand edge coefficients have been effectively introduced by the transform to supplement the sub-band coefficients which have been lost as a result of the processes applied to the image. However as will be appreciated the actual amount of shift of the image in the wavelet domain will be unknown at the correlator. Therefore, in order to accommodate a shift of the wavelet domain data in the sub-band so that the embedded data may still be detected and recovered from the sub-band, the correlation is performed repeatedley by the correlation processor 334 for consecutive wavelet coefficients starting from a plurality of start positions each of which is representative of a possible shift of the wavelet coefficients. For each start position, the magnitudes of the results of correlating the PRBS with the wavelet coefficients corresponding to a plurality of embedded data bits are combined to form a combined correlation result. The start position which produces the largest combined correlation result is determined to correspond to the actual shift which has been introduced into the image. The
operation of the control processor 332 in combination with the correlation processor 334 is illustrated schematically in Figures 11, 12 and 13.
In Figure 11 a section from the top of the high horizontal, low vertical frequencies sub-band hH 11V 1 shown in Figure 10 is represented in an expanded form. In Figure 11 the sections 400,402, 404,406 represent wavelet coefficients with which modulated versions of the PRBS have been added. The hatched region LST at the left hand end of two of the sections 400,406 represent wavelet coefficients which have
been lost as a result of the shift. A shaded region UNR at the right hand edge of a part of a modulated section of coefficients 404 represents unreliable coefficients. Each complete section, for example, the second section 402 represents a sequence of wavelet coefficients with which a modulated PRBS has been added. In order to recover the embedded data symbol represented by this set of wavelet coefficients, the reproduced PRBS P-REP must align with the sequence of wavelet coefficients with which the complete PRBS has been added. This is achieved by the correlation processor 334 in combination with the control processor 332 by correlating the reproduced PRBS P-REP with consecutive wavelet coefficients starting from each of a plurality of start positions STPOS. As illustrated in Figure 11 there are five start
positions corresponding to a shift in wavelet coefficients of'-2','-1'\'0','1', and'2' coefficients from a position represented as start position'0'at which the start of the sequence of wavelet coefficients to which the complete modulated PRBS was originally added by the data embedder 240. The result of the correlation is illustrated graphically in Figure 12. Since the illustrative example has been shown with a shift to the left of two wavelet coefficients, the combined correlation result for several embedded bit correlations at position'-2'produces the largest magnitude. The embedded data can be recovered from this shift. As shown in Figure 12, the correlation output signal with the highest value indicates that the shift of the image in the wavelet domain is'-2'coefficients.
In order to detect and recover the embedded data from the sub-band after the image has shifted, the correlation processor 334 operates to exclude the PRBS bits from the correlation sum which correspond to wavelet coefficients which are either lost or corrupted. The number of omitted coefficients corresponds to the shift which is assumed for each of the start positions. Therefore for example if the start position is'2', then it is assumed that two wavelet coefficients at either of the edges of the sub band are unreliable or lost and therefore the correlation with the corresponding bits of the PRBS are not included in the correlation sum. This is represented in Figure 13.
As illustrated in Figures 11 and 13, the correlation processor 334 operates to read the wavelet coefficients from the sub-band as illustrated by the read arrows RI.
Figure 13 provides a schematic representation of the operation of the data processor 334, where the wavelet coefficients WTS corresponding to one modulated PRBS are
shown aligned with the reproduced PRBS P REP. As shown in Figure 13. the recovered wavelet transform coefficients which were read from the region at the right hand edge of the sub-band, and the assumed lost data symbols are shown as a hashed region ER~SYS. Accordingly, the corresponding symbols in the reproduced PRBS PREP are marked as an excluded region EXREG.
The correlation sum performed by the data processor 334 is represented as the combining function COMB and the adder ADD which receives the result of the combination of the wavelet coefficient with the bit from the reproduced P-REP. This correlation sum is performed for all symbols between the wavelet transform coefficients WTC and the bits of the replicated PRBS P-REP except for those symbols within the hashed regions ERSYS and the excluded regions EX~REG. The result of each combination is added by the adder ADD thereby forming a representation of the
correlation of the wavelet coefficients and the reproduced PRBS PREP. The correlation result is presented at an output channel CORROUT. Thus it will be appreciated that the correlation sum according to equation (2) is adjusted accordingly.
Predetermined data sequences such as Pseudo Random Bit Sequences have an
auto correlation property that when the PRBS is aligned with itself a high correlation 9 t : l output is produced but for any shifts of the PRBS with respect to itself a zero output or at least a low correlation output is produced. As a result by excluding the wavelet coefficients from the excluded region EXSYS from the correlation sum the correlation output may be reduced. However this reduction may be relatively small if the PRBS is sufficiently long that the number of excluded correlation symbols does not have a significant reduction on the correlation output signal. Thus, by discounting the corresponding wavelet transform coefficients and the corresponding bits of the PRBS but otherwise maintaining the relative alignment of the symbols, the exclusion of the symbols does not have a significant effect on the performance of the correlator. As a result the entire region of the sub-band can be used for embedding data.
Further Embodiments As explained above the control processor detects an amount by which the information material has shifted in accordance with which of the start positions provides a combined correlation result with the largest magnitude. Having established this shift the control processor may be arranged in other embodiments to adapt and maintain the start position for performing subsequent correlations corresponding to
this established shift thereby avoiding the need to repeat the correlation at other start Z=7 positions.
As already explained, a shift of a single wavelet coefficient of a level one wavelet transform corresponds to a shift of two pixels in the spatial domain of the image. However if the image in the spatial domain shifts by only a single pixel, there will be a reduction in the correlation output when the reproduced PRBS is correlated with the wavelet coefficients at their original position. Otherwise, the position of the wavelet coefficients to which the modulated PRBS has been added would approximately correspond with the position of the wavelet coefficients in which the modulated PRBS was originally added. However the reduction of the combined correlation signal which results may prevent the correct start position and shift value from being detected.
In order to accommodate shifts of the image in the spatial domain by a single or any odd number of pixels, the watermark decoding processor operates to generate a second version of the image frame which is pre-shifted by one pixel before being transformed into the wavelet domain. Accordingly the watermark decoder 140 shown in Figure 8 operates to feed the watermarked image 125 for a second time to the wavelet transformer 310, after shifting the image by a single pixel value. This generation of a shifted version of the second transform domain image can be effected by the control processor 332 using feedback signals which are not shown in Figure 9 or Figure 8 although persons skilled in the art will understand how this may be implemented.
The second wavelet transformed image is also received on the connecting channel 315 by the frame store 331. The control processor 332 therefore forms the correlation of the reproduced PRBS with the wavelet coefficients for each version of the transform domain image, that is the image with the shift in the spatial domain and without the shift in spatial domain. As a result, the shifts in the start position corresponding to an integer number of wavelet coefficients will, for the second shifted version of the transform domain image, represent odd numbers of pixel shifts in the spatial domain, and the correlation with the unshifted version of the image will
represent even numbers of pixel shifts of the image in the spatial domain. Therefore it will be appreciated that the generation of a pre-shifted version of the image provides a facility for correlating the reproduced PRBS with all integer shifts of the image in the spatial domain and from this, the data may be recovered from the correlation output.
As will be appreciated, although the example embodiments have been described with the data added to the wavelet domain sub-band in the horizontal direction, in other embodiments the data may be added in the vertical direction.
Accordingly, it will be appreciated that corresponding adjustment would be introduced to the data embedder and the correlation and control processors to the effect of combining and recovering the data in the vertical direction. This is because shifts of the image may also occur in the vertical direction, so that the start positions for the correlation may represent possible vertical shifts.
Although in the example embodiment the same PRBS has been used to represent each bit of the data to be embedded in alternative embodiments a different PRBS may be used for different bits of the data to be embedded.
The Universal Material Identifier (UMID) A brief explanation will now be given of the structure of the UMID, with reference to Figure 14A and 14B. The UMID is described in SMPTE Journal March 2000. Referring to Figures 14A an extended UMID is shown to comprise a first set of 32 bytes of a basic UMID, shown in Figure 14B and a second set of 32 bytes referred to as signature meta data. Thus the first set of 32 bytes of the extended UMID is the basic UMID. The components are: . A 12-byte Universal Label to identify this as a SMPTE UMID. It defines the type of material which the UMID identifies and also defines the methods by which the globally unique Material and locally unique Instance numbers are created.
*A 1-byte length value to define the length of the remaining part of the UMID.
. A 3-byte Instance number which is used to distinguish between different 'instances'of material with the same Material number.
A 16-byte Material number which is used to identify each clip. Each Material number is the same for related instances of the same material.
The second set of 32 bytes of the signature metadata as a set of packed metadata items used to create an extended UMID. The extended UMID comprises the basic UMID followed immediately by signature metadata which comprises: - An 8-byte time/date code identifying the time and date of the Content Unit creation.
- A 12-byte value which defines the spatial co-ordinates at the time of Content Unit creation.
*3 groups of 4-byte codes which register the country, organisation and user codes.
More explanation of the UMID structure is provided in co-pending UK patent application number 0008432. 7.
Various modifications may be made to the embodiments herein before described without departing from the scope of the present invention. Although in this example embodiment, the data to be embedded is added to the image in the transform domain, in alternative embodiments the data could be represented in the transform domain, inverse transformed into the spatial domain. and added to the data in the transform domain.

Claims (24)

1. An apparatus for detecting and recovering data embedded in information material, the data having been embedded in the material using a transform domain representation of at least one of said data and said information material by arranging for the data to modulate a predetermined data sequence to form modulated data and combining said modulated data with said material. the apparatus comprising a transform processor operable to transform the material into a transform domain representation of said information material, and a correlation processor operable to correlate transform domain data symbols bearing said modulated data from said transform domain representation with a reproduced version of said predetermined data sequence to form a correlation output signal and to recover said embedded data from said correlation output signal. wherein said correlation processor is operable to perform said correlation of transform domain data symbols with data symbols of the predetermined data sequence, for a plurality of start positions in said transform domain, said start positions representing at least one relative possible shift of said transform domain data, and if said shift of said transform data represents a loss or corruption of transform domain data symbols, omitting corresponding symbols from said predetermined data sequence in said correlation, said lost or corrupted transform domain data symbols and said corresponding symbols of said predetermined data sequence not being included in calculating the correlation output signal.
2. An apparatus as claimed in Claim 1, wherein said plurality of start positions includes representations of a plurality of possible shifts of said transform domain symbols from an original position in which said modulated data was combined with said symbols.
3. An apparatus as claimed in Claim 2, wherein said possible shifts correspond to shifts of an integer number of transform domain data symbols.
4. An apparatus as claimed in any of Claims 1. 2 or 3, comprising a ZD control processor operable to detect an amount by which said information material has shifted in accordance with which of said start positions provides a correlation output signal with the largest magnitude, said start position being subsequently used to detect and recover said embedded data in accordance with said correlation output signal.
5. An apparatus as claimed in any preceding Claim, wherein a shift of data symbols representing said information material by a first number of information material data symbols, produces a shift by a second number of transform domain symbols of said representation of said information material in said transform domain. said apparatus comprising a control processor operable to shift said information material by an amount determined in accordance with said first and second numbers to the effect that said plurality of start positions of said correlation of said predetermined data sequence with said transform domain data symbols represents a corresponding plurality of integer shifts of said information material data symbols.
6. An apparatus as claimed in Claim 5, wherein said correlation at each of said plurality of start positions with said transform domain data symbols is performed with and without said shift.
7. An apparatus as claimed in Claims 5 or 6, wherein said first and second numbers are two and one respectively, said information material being shifted by one data symbol, each of said start positions in said transform domain corresponding to odd numbers of shifts of said information material data symbols.
8. An apparatus as claimed in any of Claims 6 or 7, wherein said control processor is operable to detect said amount by which said information material has shifted in accordance with which of said start positions and which of said shift and not shifted version of said information material provides the correlation output signal with the highest value.
9. An apparatus as claimed in any preceding Claim. wherein said predetermined data sequence is a Pseudo Random Bit Sequence, said data symbols being bits.
10. An apparatus as claimed in any preceding Claim, wherein said transform domain has a plurality of sub-bands. said modulated data being added to at least one of said sub-bands, said correlation processor being operable to correlate the transform domain data symbols from the sub-band with said predetermined data sequence except said excluded data symbols.
11. An apparatus as claimed in Claim 10. wherein said lost transform domain data symbols are a predetermined number of symbols at an edge of said subband.
12. An apparatus as claimed in any preceding Claim, wherein said z transform is the discrete wavelet transform, said transform data symbols being wavelet coefficients, each symbol of said modulated data being added to the wavelet coefficients.
13. An apparatus as claimed in any preceding Claim, wherein said information material in which said data is embedded is one of video images, audio signals, video and/or audio signals.
14. An apparatus for embedding data into information material, which data can be detected and recovered by the apparatus as claimed in any preceding Claim, said apparatus comprising a combining processor operable to modulate a predetermined data sequence with said data, to form modulated data and to combine said modulated data with said material in one of a transform domain representation or an inverse transform domain representation of said material, wherein said combining processor is operable to form said modulated data into a transform domain representation, by introducing said data into at least one of a plurality of transform domain sub-bands, said modulated data
being added to data symbols within said sub-band, including transform data symbols within extremes of said sub-band.
15. An apparatus as claimed in Claim 14. wherein said at least one subband represents in said transform domain low spatial frequencies in one direction and high spatial frequencies in another direction.
16. An apparatus as claimed in any of Claims 14 or 15, wherein said predetermined data sequence is a Pseudo Random Bit Sequence (PRBS) each bit of said PRBS being represented in bipolar form, said data to be embedded modulating the bits of said PRBS by reversing the sign of each bit, said modulated Pseudo Random Bit Sequences being added to respective transform domain data symbols of said subband.
17. An apparatus as claimed in any of Claims 14,15 or 16, wherein said transform is the Discrete Wavelet Transform, said modulated data being added to said sub-band at each of said wavelet coefficients between the edges of said sub-band.
18. A method of detecting and recovering data embedded in information material, the data having been embedded in the material using a transform domain representation of at least one of said information material and said data by arranging for the data to modulate a predetermined data sequence to form modulated data and combining said modulated data with said material, the method comprising forming a transform domain representation of said information material, correlating transform domain data symbols bearing said modulated data from said transform domain representation with a reproduced version of said predetermined data sequence, to form a correlation output signal, and recovering the embedded data from said correlation output signal, wherein said correlating said predetermined data sequence with said transform domain data symbols, comprises correlating transform domain data symbols and data symbols of the predetermined data sequence
repeating said correlation for at least one other start position in said transform domain, said start position representing at least one relative possible shift of said transform domain data. and if said shift of said transform domain data represents a loss or corruption of transform domain data symbols. omitting corresponding symbols from said predetermined data sequence, said lost or corrupted transform domain data symbols and said corresponding symbols of said predetermined data sequence not being included in calculating the correlation output signal.
19. A signal representing information material in which data has been embedded by an image processing apparatus according to any of Claims 14 to 17.
20. A computer program providing computer executable instructions, which when loaded on to a data processor configures said data processor to operate as an apparatus according to any of Claims 1 to 17.
21. A computer program having computer executable instructions, which when loaded on to a data processor causes the data processor to perform the method according to Claim 18.
22. A computer program product having a computer readable medium having recorded thereon information signals representative of the computer program claimed in any of Claims 20 or 21.
23. An apparatus as herein before described with reference to the accompanying drawings.
24. A method of embedding data in an image as herein before described with reference to the accompanying drawings.
GB0029867A 2000-12-07 2000-12-07 Detecting and recovering embedded data using correlation and shifting Withdrawn GB2369951A (en)

Priority Applications (4)

Application Number Priority Date Filing Date Title
GB0029867A GB2369951A (en) 2000-12-07 2000-12-07 Detecting and recovering embedded data using correlation and shifting
EP01310175A EP1215908A3 (en) 2000-12-07 2001-12-05 Apparatus for detecting and recovering embedded data
US10/007,085 US7085395B2 (en) 2000-12-07 2001-12-06 Apparatus for detecting and recovering data
JP2001375051A JP4094842B2 (en) 2000-12-07 2001-12-07 Data detection and playback device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
GB0029867A GB2369951A (en) 2000-12-07 2000-12-07 Detecting and recovering embedded data using correlation and shifting

Publications (2)

Publication Number Publication Date
GB0029867D0 GB0029867D0 (en) 2001-01-24
GB2369951A true GB2369951A (en) 2002-06-12

Family

ID=9904641

Family Applications (1)

Application Number Title Priority Date Filing Date
GB0029867A Withdrawn GB2369951A (en) 2000-12-07 2000-12-07 Detecting and recovering embedded data using correlation and shifting

Country Status (1)

Country Link
GB (1) GB2369951A (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0951175A2 (en) * 1998-04-15 1999-10-20 Xerox Corporation Automatic detection and retrieval of embedded invisible digital watermarks from halftone images
WO2000022772A1 (en) * 1998-10-14 2000-04-20 Liquid Audio, Inc. Robust watermark method and apparatus for digital signals
WO2000030291A1 (en) * 1998-10-14 2000-05-25 Liquid Audio, Inc. Secure watermark method and apparatus for digital signals

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP0951175A2 (en) * 1998-04-15 1999-10-20 Xerox Corporation Automatic detection and retrieval of embedded invisible digital watermarks from halftone images
WO2000022772A1 (en) * 1998-10-14 2000-04-20 Liquid Audio, Inc. Robust watermark method and apparatus for digital signals
WO2000030291A1 (en) * 1998-10-14 2000-05-25 Liquid Audio, Inc. Secure watermark method and apparatus for digital signals

Also Published As

Publication number Publication date
GB0029867D0 (en) 2001-01-24

Similar Documents

Publication Publication Date Title
US6683966B1 (en) Watermarking recursive hashes into frequency domain regions
US7181042B2 (en) Digital authentication with digital and analog documents
US8170273B2 (en) Encoding and decoding auxiliary signals
US7203336B2 (en) Embedding data in material
US7321666B2 (en) Multilayered digital watermarking system
US7336802B2 (en) Digital watermarking system using scrambling method
Singh Digital watermarking: a tutorial
JP2008252891A (en) Methods and apparatus for embedding data and for detecting and recovering embedded data
WO2003032254A1 (en) Digital watermarking method robust against local and global geometrical distortions and projective transforms
US7085395B2 (en) Apparatus for detecting and recovering data
US7127079B2 (en) Image processing apparatus
EP2024927A1 (en) Encoding and detecting apparatus
GB2438904A (en) Generation of code words for image watermarking
US7171017B2 (en) Apparatus for embedding data into transform and spatial domains of an image
WO2002019269A2 (en) A method for encoding and decoding image dependent watermarks
GB2369951A (en) Detecting and recovering embedded data using correlation and shifting
Mora-Jimenez et al. A new spread spectrum watermarking method with self-synchronization capabilities
GB2369948A (en) Apparatus for recovering embedded watermark data via correlation sequences
GB2369946A (en) Apparatus for embedding and recovering data depending on error levels
GB2369945A (en) Method of embedding watermark information according to relative importance
Huang et al. Host-Interference Rejecting Spread-Spectrum Watermarking: Implementation and Implications
Murumbekar et al. A New Approach to Audio Wave Watermarking

Legal Events

Date Code Title Description
WAP Application withdrawn, taken to be withdrawn or refused ** after publication under section 16(1)