WO2009122385A2

WO2009122385A2 - Fingerprinting method and system

Info

Publication number: WO2009122385A2
Application number: PCT/IB2009/052737
Authority: WO
Inventors: Zhongxuan Liu; Shiguo Lian; Zhen Ren
Original assignee: France Telecom
Priority date: 2008-03-29
Filing date: 2009-03-30
Publication date: 2009-10-08
Also published as: WO2009122385A3

Abstract

A method for embedding a fingerprint into a media content, said fingerprint comprising elements belonging to an alphabet, said method comprising the acts of embedding a plurality of watermarks into the media content, said watermarks corresponding to the elements of the fingerprint and at least to one other element of the alphabet and eliminating a number of watermarks so that only watermarks corresponding to the elements of the fingerprint remain embedded in the media content.

Description

FINGERPRINTING METHOD AND SYSTEM

Field of the Invention

The present invention relates in general to a method for watermarking media content and more specifically for securely fingerprinting media data.

Background of the Invention

Media content, for example a video, is intended to be distributed, but only users paying for the content should be allowed to access it. In order to trace the users distributing the content illegally, user identification information, or fingerprint, is imperceptibly inserted or embedded into the media content. This fingerprint is unique for each user and allows the identification of the user. As far as fingerprinting is concerned, the goal of hackers is to distribute the content illegally without being traced by removing, at least partially, the fingerprint from the media content. One of the most common types of attack is called Linear Combination Collusion Attack (LCCA) and consists in combining several media copies of the same content (for instance, several copies of the same video) and extracting an averaged video so that the fingerprint is at least partially altered and hence prevent any identification. Media content such as e.g. a video is usually compressed using a compression algorithm according to a standard such as e.g. Moving Picture Experts Group 2 (MPEG2) and where said video is compressed into a plurality of frames.

A fingerprint is composed of a series of elements from an alphabet. Usually, these elements are 0 and 1 if the binary alphabet is used. Fingerprinting usually consists in embedding a plurality of watermarks into a media content, wherein each watermark corresponds to one or several elements of the fingerprint. But such a raw method for fingerprinting directly into the media content does not resist LCCA. In order to limit LCCA, a technique called time desynchronization consists in suppressing randomly different frames in each video copy. The number of suppressed frames needs to remain relatively small, usually around 10% of the total number of frames, in order to avoid degrading the quality of the video too much, i.e. in a perceptible manner for the human eye.

This time desynchronization technique is efficient, but is still not secure enough against hackers' attacks.

The method proposed in "Fingerprint Multicast in Secure Video Streaming. H.V. Zhao, and K.J. R. Liu, IEEE Trans. Image Processing, vol. 15, no. 1 , pp. 12-29, January 2006" is not suitable for DVDs because extra different data should be transmitted to every copy. Coded fingerprints need thus to be used like in "Collusion-Resistant Fingerprinting for Multimedia. M.

Wu, W. Trappe, Z.J. Wang, and K.J. R. Liu, IEEE Signal Processing Magazine, March 2004, 15-27" and "A Scalable AND-ACC Fingerprinting Scheme for

Practical Contents Distribution. J. M. Seol, and S.W. Kim, SPIE Visual

Communications and Image Processing 2005" a major drawback is that the needed codes for applications such as e.g. DVDs are too long because they depend on the number of copies. Today there is a need for a secure media content fingerprinting solution that can be easily implemented on the existing communication infrastructures.

Summary of Invention It is an object of the present system to overcome disadvantages and/or make improvement over the prior art.

To that extend, the invention proposes a method according to claim 1 . The invention proposes a solution for resisting to LCCA as the media content is fingerprinted and desynchronized in a single process. The invention also relates to a system according to claim 8.

The invention also relates to a fingerprinting device according to claim 15.

The invention also relates to a computer program according to claim 21 .

Brief Description of the Drawings

Embodiments of the present invention will now be described solely by way of example and only with reference to the accompanying drawings, where like parts are provided with corresponding reference numerals, and in which:

Figure 1 schematically illustrates a process of media lengthening and watermark embedding according to an embodiment of the present invention; Figure 2 schematically illustrates a process of media fingerprinting according to an embodiment of the present invention;

Figure 3 schematically illustrates a process of frames selection for fingerprinting media according to an embodiment of the present invention;

Figure 4 schematically illustrates a process of frames selection for fingerprinting encrypted media according to an embodiment of the present invention;

Figure 5 schematically illustrates a process of fingerprint extraction according to an embodiment of the present invention.

Figure 6 schematically illustrates the method according to an embodiment of the present invention.

Figure 7 schematically illustrates the method according to an embodiment of the present invention using frames.

Figure 8 schematically illustrates the method according to an embodiment of the present invention using frames, time-desynchronization and fingerprinting sequences.

Description of Preferred Embodiments

The method according to the invention is a media content fingerprinting method robust to collusion attack, one of the main threats to copyrights infringements.

The method according to the invention allows embedding user identification information, i.e. a fingerprint, into a media content and wherein the code length of the fingerprint is small, for application like e.g. DVDs.

The method according to the invention may be implemented for example during the media making process (DVD creation, DVD copy...), before the media opening or reading. The media content, e.g. a video, is typically divided or compressed into a series of elements such as e.g. frames. The fingerprint is composed of elements belonging to an alphabet.

In the method according to the invention: - the elements of the fingerprint alphabet corresponding to the fingerprint and at least one other element of the fingerprint alphabet are embedded into some elements of the media content using watermarks,

- a number of said elements of the media content is eliminated so that only the elements of the media content comprising the elements of the fingerprint remain.

In optional exemplary embodiments of the method according to the invention: time desynchronization and encryption/decryption are combined with the method according to the invention in order to enhance the security.

The security of the system does not decrease and no extra computation is added to the reading system.

According to one optional exemplary embodiment of the present invention, the media content may be composed of frames and the act of eliminating a number of watermarks may encompass an act of eliminating a number of frames so that only frames comprising the watermarks corresponding to the elements of the fingerprint remain embedded in the media content.

According to one optional exemplary embodiment of the present invention, a number of frames may be inserted prior to embedding watermarks. This allows lengthening the video in order to further eliminate a number of elements of the media content comprising certain watermarks.

According to one optional exemplary embodiment of the present invention, the media content may be desynchronized by eliminating randomly part(s) of it, e.g. by eliminating a number of frames. This desynchronization allows having different copies of the same media content so that it is harder to use LCCA against the video to collude the fingerprint and have a clean video. In order not to degrade the media content too much for it to be readable, only a small part of the media content is randomly removed, e.g. typically 5 to 10 %. In this manner, the missing part(s) of the media content do not affect too much the quality of the media so that each desynchronized copy looks identical for the human eye in terms of perception.

According to one optional exemplary embodiment of the present invention, desynchronization of the media content and watermarks, or frames, elimination may be achieved together in a single step in order to fingerprint the media content and enhance the security at the same time.

A watermarking device allows the watermark embedding whereas a fingerprinting device allows the elimination of a number of watermarks so that only the ones corresponding to the fingerprint remain in the media content. These two devices may be one single device. If separated and for example (e.g. broadcasting), the watermarking device may be on the emission side whereas the fingerprinting device may be on the reception side (or user side).

According to one optional exemplary embodiment of the present invention, an act of encrypting the media content subsequently to the act of watermarks embedding and an act of decryption during the act of eliminating the watermarks corresponding to the elements of the fingerprint remain embedded in the media content may be added.

According to one optional exemplary embodiment of the present invention, watermarks embedding, desynchronization, encryption, watermarks/frames elimination and decryption may ultimately be achieved altogether in the same process in order to fingerprint the media content and enhance more the security at the same time.

According to one embodiment of the present invention, Figure 1 describes the preliminary acts of the fingerprinting process. In an act 20, the original video 10 is lengthened or extended by interpolating or repeating some frames. Neighbor frame pairs may be similar or non-similar. This lengthening may consist in duplicating a number of frames equal to the number of bits in the fingerprint. Each original frame which is duplicated gives an associated duplicated frame, forming thus a pair of associated frames. The act 30 consists in selecting the pairs of associated frames comprising the original frames that have been duplicated in the lengthening process and their associated duplicated frames. The act 40 allows embedding a sequence or watermark into each frame of each pair associated frame. In one exemplary embodiment according to the invention, this watermark corresponds to either a bit 0 or a bit 1. It may be for instance a vector who is added or embedded or inserted or combined into the media frame and which represents either a 0 or a 1. This watermark could also correspond to other type of data comprising a code as e.g. hexadecimal or else... Watermarks may be embedded for example through adding/minusing random signals.

An optional act 50 allows the encryption of the lengthened media comprising the watermarked frames e.g. for DVD printing. The resulting media 60 undergoes further processes to complete the fingerprinting method according to the invention and described in the further figures.

Figure 2 describes the frames selection process for fingerprinting. The video may have been optionally encrypted (100). The video has gone through the acts 20, 30 and 40 (lengthening and watermarks embedding).

In an act 110, the video is optionally desynchronized, fingerprinted and optionally decrypted (if it were encrypted). The video is fingerprinted by eliminating a watermarked frame in each pair of watermarked frames in order to keep the watermarked frame in each pair of watermarked frames with the embedded watermark corresponding to a given bit of the fingerprint.

If the video were encrypted, the decryption process uses encryption/decryption keys.

The video is eventually decompressed to be displayed.

Figure 3 describes in more details the desynchronizing and fingerprinting acts of the method according to the invention.

A series of frames (200) resulting from a media compression has been lengthened, meaning that a number of frames have been inserted in the series. These frames may be duplicated frames. There should be at least as many as extra/inserted frames as the number of elements (e.g. bits) comprised in the fingerprint. In Figure 3, the example uses fingerprint 1001 to insert in the media. Four frames C'5, C'9, C'11 and C'15 are thus inserted in the series of frames (200). In each pair of frames (C5/C5), (C9/C9), (C11/C11) and (C15/C15), a watermark corresponding to bit 0 is embedded in one of the frame of the pair and a watermark corresponding to bit 1 is embedded in the other frame of the same pair. Thus, in each pair of frames, one of the frame is embedded with a watermark corresponding to bit 0 and the other one with watermark corresponding to bit 1.

The optional act of desynchronization consists in eliminating randomly a number of frames so that two different copies of the same media cannot be exactly superimposed and averaged without keeping the original quality of the media. It is common in such a process to eliminate up to 10% of the frames in order to avoid degrading too much the quality of the media. In Figure 3, the series 220 is called desynchronization sequence of the same length of the media (one bit in the sequence for each frame of the media) where up to 10 % of bit zeros are added, the rest being bit ones. When used, the desynchronization sequence will be combined (e.g. added) with the fingerprinting sequence 230 to eliminate a number of frames in the series 200.

The fingerprinting sequence 230 is composed of zeros, ones and twos (talking hence not of bits anymore). This allows the selection of a precise frame in the series 200 under the following criterion: frames of series 200 are only kept when the sum of each corresponding figures of the desynchronization and the fingerprinting sequences is at least equal to 2. In other words, frames for which the sum of the corresponding figures is 0 or 1 are eliminated.

Thus, for example, if watermarks corresponding to bit 0 were embedded in frames C5, C9, C11 and C15 and watermarks corresponding to bit 1 were embedded in frames C'5, C'9, C'11 and C'15, and the fingerprint to embed in the media is 1001 , then it means that only frames C'5, C9, C11 and C'15 need to be kept and frames C5, C'9, C'11 and C15 need to be eliminated from the series 200 in order to fingerprint the media with fingerprint 1001. The information about which frames are embedded with whichi watermark needs to be available in order to be able to generate the fingerprinting sequence.

In the fingerprinting sequence, a 0 is set when the corresponding frame needs to be eliminated and a 2 is set when the corresponding frame needs to be kept. A 1 is set otherwise and when combined with the corresponding bit of the desynchronization sequence (0 or 1) will either allow eliminating or keeping the corresponding frame in the series 200. Every frame corresponding to a resulting sequence (sum of desynchronization and fingerprint sequence) number equal or greater then 2 is kept, every frame corresponding to a resulting sequence number equal to 0 or 1 is eliminated.

The resulting series of frames 250 comprises a number of original frames and a given number of watermarked frames (original and/or inserted) wherein the watermarks correspond to elements/figures/bits of the fingerprint, and misses a small number of frames (original and/or inserted) due to the elimination induced by the desynchronization and the fingerprinting acts.

The desynchronization sequence and fingerprinting sequence may be generated by either the watermarking device or by the fingerprinting device or separately providing the fact said device(s) has all the information to do so.

Figure 4 describes the method according to the invention wherein encryption and decryption processes are used.

The sequence 200 of raw frames from the media content and inserted frames is encrypted into a sequence 210 using encryption keys. The combination of the desynchronization sequence 220 and fingerprinting sequence 230 allows the elimination of a number of frames so that the resulting encrypted sequence 240 is together desynchronized and fingerprinted (i.e. containing a number of frames with embedded watermarks which correspond to the elements of the fingerprint). It is to be noticed that the desynchronization/elimination process may obviously be applied either on the encrypted frames or on the encryption keys. In the latter case, a key corresponding to a frame to eliminate is removed or tagged so that the corresponding frame is eliminated. Figure 5 describes the fingerprint extraction.

The fingerprinted is extracted by firstly searching the first position(s) where the fingerprint can be extracted. The correlation between the media content (e.g. video clip) (310) and the random sequence of fingerprint (320) is computed. If the correlation is larger than a predetermined threshold (340), fingerprint is extracted (350) for tracking illegal users (399); else correlation between next frame and the random sequence is computed.

Figure 6 describes the method according to the invention wherein watermarks corresponding to elements of the fingerprint to insert into the media content and at least another element of the fingerprint alphabet are embedded in an act 600. The embedded watermarks that do not correspond to elements of the fingerprint are eliminated in a subsequent act 610 so that only watermarks corresponding to elements of the fingerprint remain into the media content, i.e. meaning the fingerprint is embedded or inserted into the media content for further authentication.

Figure 7 describes the method according to the invention wherein the media content comprises frames. The watermarks are thus embedded into frames (act 700) and the frames corresponding to watermarks to rule out are selected (act 710) and eliminated (act 720).

Figure 8 describes the method according to the invention wherein the media content comprises frames and wherein prior to frame selection, a desynchronization sequence and a fingerprinting sequence are generated.

The watermarks are embedded into frames (act 800), then a desynchronization sequence is generated (act 810) like previously explained in Figure 3 and 4 in order to time-desynchronized the copied of a given media content. Then a fingerprinting sequence is generated (act 820) in order to indicate (e.g. here with a 2) the frames comprising the watermarks corresponding to the elements of the fingerprint and the frames(e.g. here with a 0) to eliminate. The desynchronization and fingerprint sequences are then summed in order to further select and eliminate the frames that are weighted with a number smaller than 2 (0 or 1). The frames corresponding to watermarks to rule out are selected (act 830) and eliminated (act 840). The frames with no embedded watermarks are weighted with a 1 in the fingerprinting sequence, the frames to keep are weighted with a 2 and the frames with embedded watermarks to eliminate are weighted with a 0. The desynchronization sequence is typically a random sequence with 90% of ones and 10% of zeros so that around 10% of the frames (different frames in each copy of media content) are eliminated during the elimination process to time-desynchronized the media content copied.

Figure 9 describes the system according to the invention wherein the original raw media content 900 is inputted into the watermarking device 910 for watermark embedding. The resulted watermarked media content is then communicated via the communication link 920 to the fingerprinting device 930 along with the needed information to select the parts of the watermarked media content to eliminate in order to obtain a resulting media content fingerprinted copy 940. The watermarking device 910 and fingerprinting device 930 comprise each a input unit and a output unit to receive and send data.

Claims

1 . A method for embedding a fingerprint into a media content, said fingerprint comprising elements belonging to an alphabet, said method comprising the acts of:

- embedding a plurality of watermarks into the media content, said watermarks corresponding to the elements of the fingerprint and at least to one other element of the alphabet;

- eliminating a number of watermarks so that only watermarks corresponding to the elements of the fingerprint remain embedded in the media content.

2. A method according to claim 1 , wherein the media content is composed of frames, the act of embedding comprises the act of embedding each watermark into a different frame of the media content and the act of eliminating comprises the act of eliminating a number of frames so that only frames comprising the watermarks corresponding to the elements of the fingerprint remain into the media content.

3. A method according to claim 2, wherein the act of eliminating further comprises the acts of:

- generating a desynchronization sequence, said desynchronization sequence indicating potential frames to eliminate,

- generating a fingerprinting sequence, said fingerprinting sequence indicating the frames with the watermarks corresponding to the elements of the fingerprint and the frames with the watermarks that do not correspond to the elements of the fingerprint,

- eliminating the frames using a combination of the desynchronization sequence and the fingerprinting sequence.

4. A method according to claim 3, wherein the alphabet of the fingerprint elements is a binary alphabet and wherein the act of embedding comprises the act of embedding for each element of the fingerprint watermarks corresponding to the whole alphabet into a pair of distinct frames and wherein the act of eliminating comprises the act of eliminating the one of the pair of the distinct frames the watermark of which does not correspond to an element of the fingerprint.

5. A method according to any of the preceding claims, further comprising, prior to the act of embedding a plurality of watermarks, the act of inserting a number of frames.

6. A method according to claim 5, wherein each pair of distinct frames comprises one frame from the media content and one inserted frame.

7. A method according to any of the precedent claims, further comprising an additional act of encrypting the media content subsequently to the act of embedding.

8. A system for embedding a fingerprint into a media content, said fingerprint comprising elements belonging to an alphabet, said system comprising: - a watermarking device operable to embed a plurality of watermarks into the media content, said watermarks corresponding to the elements of the fingerprint and at least to one other element of the alphabet;

- a fingerprinting device operable to eliminate a number of watermarks so that only watermarks corresponding to the elements of the fingerprint remain embedded in the media content;

- a communication link for transmitting the watermarked media content from the watermarking device to the fingerprinting device.

9. A system according to claim 8, wherein the media content is composed of frames and wherein the watermarking device is further operable to embed each watermark into a different frame of the media content and wherein the fingerprinting device is further operable to eliminate a number of frames so that only frames comprising the watermarks corresponding to the elements of the fingerprint remain into the media content.

10. A system according to claim 9, wherein any of the watermarking device and fingerprinting device is further operable to:

- generate a desynchronization sequence, said desynchronization sequence indicating potential frames to eliminate, - generate a fingerprinting sequence, said fingerprinting sequence indicating the frames with the watermarks corresponding to the elements of the fingerprint and the frames with the watermarks that do not correspond to the elements of the fingerprint; the fingerprinting device being further operable to:

- eliminate the frames using a combination of the desynchronization sequence and the fingerprinting sequence.

1 1 . A system according to claim 10, wherein the alphabet of the fingerprint elements is a binary alphabet and wherein the watermarking device is further operable to embed for each element of the fingerprint watermarks corresponding to the whole alphabet into a pair of distinct frames and wherein the fingerprinting device is further operable to eliminate the one of the pair of the distinct frames the watermark of which does not correspond to an element of the fingerprint.

12. A system according to any of the preceding claims 8 to 1 1 , wherein the watermarking device is further operable to, prior to embedding a plurality of watermarks, insert a number of frames.

13. A system according to claim 12, wherein each pair of distinct frames comprises one frame from the media content and one inserted frame.

14. A system according to any of the precedent claims 8 to 13, wherein the watermarking device and fingerprinting device are one single device.

15. A fingerprinting device for embedding a fingerprint into a media content, said fingerprint comprising elements belonging to an alphabet, said fingerprinting device comprising a receiving unit for receiving said media content, said fingerprinting device being operable to: - embed a plurality of watermarks into the media content, said watermarks corresponding to the elements of the fingerprint and at least to one other element of the alphabet;

- eliminate a number of watermarks so that only watermarks corresponding to the elements of the fingerprint remain embedded in the media content.

16. A fingerprinting device according to claim 15, wherein the media content is composed of frames, said fingerprinting device being further operable to eliminate a number of frames so that only frames comprising the watermarks corresponding to the elements of the fingerprint remain into the media content.

17. A fingerprinting device according to claim 16, said fingerprinting device being further operable to: - generate a desynchronization sequence, said desynchronization sequence indicating potential frames to eliminate,

- generate a fingerprinting sequence, said fingerprinting sequence indicating the frames with the watermarks corresponding to the elements of the fingerprint and the frames with the watermarks that do not correspond to the elements of the fingerprint,

18. A fingerprinting device according to claim 17, wherein the alphabet of the fingerprint elements is a binary alphabet, said fingerprinting device being further operable to:

- embed for each element of the fingerprint watermarks corresponding to the whole alphabet into a pair of distinct frames,

- eliminate the one of the pair of the distinct frames the watermark of which does not correspond to an element of the fingerprint.

19. A fingerprinting device according to any of the preceding claims 15 to

18, said fingerprinting device being further operable to, prior to embedding a plurality of watermarks, insert a number of frames.

20. A fingerprinting device according to claim 19, wherein each pair of distinct frames comprises one frame from the media content and one inserted frame.

21 . A computer program providing computer executable instructions stored on a computer readable medium, which when loaded on to a data processor causes the data processor to perform a method for embedding a fingerprint into a media content according to claims 1 to 7.