Signed video data with linked hash

Abstract

To provide methods and devices for signing prediction-coded video data with a mechanism protecting against unauthorized replacement, removal or insertion of video frames.SOLUTION: A method includes: obtaining a coded video sequence including at least one I-frame (I) which contains independently decodable image data, and at least one predicted frame which contains image data decodable by reference to at least one other frame; generating a fingerprint of each I-frame; generating a fingerprint of each predicted frame by hashing a combination of data derived from the predicted frame and data derived from an I-frame to which the predicted frame refers, the fingerprint of the predicted frame being independent of any further predicted frame to which the predicted frame refers; and signing the video sequence including the generated fingerprint.SELECTED DRAWING: Figure 6

Description

[Technical Field]

[0001] The present disclosure relates to the field of security measures for protecting programs or data from fraudulent activity. In particular, a method and device for signing predictively coded video data is proposed here. [Background technology]

[0002] A video sequence is an ordered sequence of video frames, which are composed of pixels. Successive video frames will be strongly correlated if they represent a common scene. The ability to predict one video frame given past frames is the underlying assumption of predictive coding. Predictive coding can be described as a data compression technique specifically adapted for video data.

[0003] 1A illustrates a segment of a predictively coded video sequence comprised of I-frames and P-frames. I-frames are data structures that allow image data to be independently decoded into plaintext video frames (or blocks of video frames) through a predetermined associated decoding operation. P-frames, on the other hand, are data structures that reference image data from at least one other I-frame or P-frame in addition to their own image data through the associated decoding operation. Conceptually, and somewhat simplified, the image data in P-frames represents change or motion relative to the video frames encoded by the preceding I-frame or P-frame. When the decoding operation is successful, video frames decoded from P-frames and I-frames are generally indistinguishable.

[0004] Dependencies between frames are shown in FIG. 1A as arc-shaped arrows pointing in the negative time direction t. In the illustrated simple example, each P frame references the immediately preceding I-frame or P-frame. If a first P-frame references a second P-frame, the second P-frame does not necessarily reference at least one additional I-frame or P-frame. In this disclosure, it is said that a first P-frame directly references a second P-frame and indirectly references at least one additional I-frame or P-frame. Because the image data in an I-frame is independently decodable, the reference chain (arc-shaped arrow) does not continue past the I-frame. An I-frame and subsequent P-frames that directly or indirectly reference the I-frame may be referred to as a group of pictures (GOP). Four GOPs are shown in FIG. 1A: IPP, IPPPP, IPPP, and IPPP.

[0005] FIG. 1B illustrates two further developments of inter-frame prediction, where inter-frame dependencies are again indicated by arc-shaped arrows. In addition to the forward-predicted P-frame structure, bidirectionally predicted B-frames are used. The underlying bidirectional prediction operation may include interpolation between reference frames, such as smoothing. Furthermore, in FIG. 1B, it can be seen that P-frames can reference I-frames, P-frames, or B-frames, which need not be immediately preceding, but may be more than one step away. The video sequence IBBPBBIBBPBBI shown in FIG. 1B may be characterized as a GOP because it is decodable without referencing any other I-frame, P-frame, or B-frame. The International Telecommunication Union's Recommendation ITU-T H.264 (06 / 2019) "Advanced Video Coding for General Audiovisual Services" specifies a video coding standard in which both forward-predicted and bidirectionally predicted frames are used.

[0006] Digital signatures provide a layer of validation and security for digital messages sent over insecure channels. Digital signatures can ensure that the authenticity or integrity of a message can be verified and not repudiated. Video data, as a special case, can of course be digitally signed. However, rather than treating the video data as a general bit pattern, significant efficiency gains can be gained by leveraging the frame structure instead. For example, signing can be computationally more tractable, yet at the same time generally more secure, if applied to a fingerprint of the video data rather than the video data itself. A frame-by-frame fingerprint H I , H P The use of fingerprints is shown at the bottom of Figure 1A, where each fingerprint may be a hash of the image data for that video frame, or a hash of the image for that video frame combined with optional further information. The signatures S1, S2, S3, S4 for a GOP are obtained by digitally signing the combination of fingerprints in that GOP. The security of a video sequence is further enhanced by the use of fingerprints H for I-frames, as shown in Figure 1A. I is signed both as part of its own GOP and as part of the signature of the preceding GOP, it protects against unauthorized replacement, removal, or insertion of an entire GOP. Note that the arrow pointing straight down in the top row represents a hash function, and the arrow in the bottom row represents a digital signature.

[0007] Patent document 1 discloses a method for digitally signing data added to an initial set of unique data items. All data items are converted into unique digests (or hashes) of equal length using a one-way function, and the hashes are stored as linked data sets in which the value of each successive hash depends on the value of the previous hash combined with a new hash resulting from hashing the most recently added data. Recent hashes in the linked data set (hash chain) are the cumulative result of all previous hashes in their original sequence, and each hash in the data set is unique to the most recently added data. Each cumulative hash is constructed by incorporating one or more pre-existing cumulative hashes into the source data before the new hash is calculated.

[0008] While the hash chain link structure according to U.S. Patent No. 6,277,633 may protect against attempts by unauthorized parties to replace or remove data items or to insert unauthentic data items, these teachings are not directly appropriate for signing predictively coded video data, where each "data item" is a video frame. In practice, it would have to be accepted that, as a result of inter-frame dependencies, one unverifiable frame fingerprint will result in the unverification of fingerprints for one or more further video frames, even if these video frames are successfully decodable and associated with their own fingerprints. All subsequent frames in the GOP will be "false positives" if an IPP...PP format is used, and the impact may be even greater in coding formats where B frames are used.

[0009] Similarly, US Patent No. 6,269,669 discloses a specific way of using hash functions to calculate linked watermarks for I-frames and two types of predicted frames in a video sequence. The frame data of each predicted frame is combined with the hash of the previous frame before hashing. Thus, the hash of a subsequent predicted frame (B) is calculated from the combination of the frame data of the predicted frame and the hash of the previous predicted frame (P).

[0010] A related approach is shown in Figure 2. The circular arrows above the video sequence indicate dependencies between frames, and the operation d represents the decoding. Using a given hash function h, we generate the following fingerprint: I =h(I), H P1 =h([I, P1]), H P2 =h([P1, P2]), H P3 =h([P2, P3] and H P4 = h([P3, P4]) is computed by the sender, where [·] indicates data concatenation. If this link setup makes it impossible for the receiver to verify the fingerprint of the nth P frame, then the nth P frame shall be deemed inauthentic, and thereby also unavailable for recomputing the fingerprint of the (n+1)th P frame. In that way, the receiver cannot verify the authenticity of the (n+1)th P frame, even assuming successful decoding. In a variant of the hash link setup shown in Figure 2, the following fingerprint: H I =h(I), H P1 =h([H I , P1]), H P2 =h([H P1 , P2]), H P3 =h([H P2 , P3]), and H P4 =h([H P3, P4]) is computed by the sender. Furthermore, if the receiver is unable to verify the fingerprint of the nth P frame, the nth P frame shall not be used to recompute the fingerprints of the (n+1)th P frame and subsequent P frames, and the receiver will not be able to verify the authenticity of the (n+1)th P frame and subsequent P frames, even if they become decodable.

[0011] These examples illustrate the shortcomings of available methods for digitally signing predictively coded video data. [Prior art documents] [Patent documents]

[0012] [Patent Document 1] U.S. Patent Application Publication No. 20190103975 [Patent Document 2] U.S. Patent No. 6,996,717 Summary of the Invention

[0013] One objective of the present disclosure is to make available a method and device for signing (digitally signing) predictively coded video data that has a mechanism for protecting against unauthorized substitution, removal, or insertion of video frames. A further objective is to propose such a method and device that allows fine-grained signature verification. It is desirable for the associated signature verification process to be, among other things, limited in its susceptibility to error propagation and false positives. A still further objective is to propose such a method and device that is suitable for video coding formats that have unidirectionally (forward) predicted frames and / or bidirectionally predicted frames. A particular objective is to propose a computationally efficient signature technique that has the above properties.

[0014] At least some of these objects are achieved by the present invention as defined by the independent claims. The dependent claims relate to advantageous embodiments of the invention.

[0015] In a first aspect of the present invention, there is provided a method for signing predictively coded video data, the method comprising obtaining a coded video sequence including at least one I-frame containing independently decodable image data and at least one predicted frame containing image data that can be decodized by reference to at least one other frame, and obtaining a fingerprint H for each I-frame. I and generating a fingerprint H for each predicted frame by hashing a combination of data derived from the predicted frame and data derived from I-frames that the predicted frame directly or indirectly references. P wherein the fingerprint of a predicted frame is independent of any further predicted frames to which the predicted frame directly or indirectly refers. P and providing a signature of the video sequence including the generated fingerprint.

[0016] It should be understood that in this disclosure, a “fingerprint” of a data item may be obtained by hashing the data item or by performing different operations on the data item or a subset thereof, possibly combined with other data. The data item may be image data encoding a video frame. As used herein, the term “predicted frame” further refers to unidirectional or bidirectional predicted frames, such as the P-frames and B-frames described above. The terms “I-frame” and “predicted frame” refer to image data (i.e., code) rather than the plaintext video frames that these data structures encode. As such, even strongly correlated consecutive plaintext video frames can be encoded into corresponding uncorrelated I-frames and predicted frames. In any event, the teaching that the fingerprint of a predicted frame shall be independent of any further predicted frames to which it refers should be understood to make the fingerprint of a predicted frame independent both from the image data of the further-referenced frames and from the fingerprints of the further-referenced frames. The teachings may also be understood to mean that the fingerprint of a predicted frame is generated without introducing any dependency on the image data or fingerprints of further-referenced frames. In effect, the fingerprint of a predicted frame is generated by a hashing operation whose input does not include data derived from such further-referenced frames.

[0017] An advantage of the signing method according to the first aspect of the present invention is that if the fingerprint of a predicted frame cannot be verified by the receiver as a result of manipulation or a transmission error, this only affects the predicted frame itself. Indeed, predicted frames without a verifiable fingerprint may be discarded or suspended from playback. Furthermore, because the fingerprint of a predicted frame is independent of any further predicted frames to which the predicted frame directly or indirectly refers, the receiver remains able to reconstruct other frames of the video sequence and verify their authenticity as usual. Therefore, the present invention may eliminate the propagation problem ("false positives") of the prior art without introducing new vulnerabilities. In fact, if a malicious party removes or reorders frames, the reference chain back to the I-frame will be corrupted, leading to unsuccessful decoding. Furthermore, manipulation of an I-frame will be indirectly visible through the fingerprints of all the predicted frames it references, thereby increasing the probability that such manipulation will be discovered.

[0018] In one embodiment, the data derived from the predicted frame includes image data for the predicted frame. Optionally, the data may further include a cryptographic salt. This embodiment may reduce the total number of hash operations in an implementation.

[0019] In one embodiment, the data derived from the predicted frame includes at least a hash of the image data of the predicted frame. In particular, a hash of the combination of the image data and a cryptographic salt is available. Recall that on the one hand, the hash is a much smaller data set than the image data of a typical video frame, and on the other hand, the two data items are typically concatenated by moving them into a common memory space. Based on this, this embodiment allows a fingerprint of the predicted frame to be generated after only a lightweight memory operation of concatenating the hashes of the image data of the predicted frame and the data derived from the I-frame.

[0020] In one embodiment, the data derived from the I-frame is the I-frame fingerprint H I That is, both the hash chaining operation and the video sequence signature use the same fingerprint for I-frames, which may simplify the implementation of the method and reduce the degree of computational duplication.

[0021] The fingerprints H of each of the I-frames and predicted frames I , H P can be generated using the same or two different hash functions. In one embodiment, the fingerprint H of an I-frame I is generated using a relatively more secure hash function (e.g., 1024 bits), and the predicted frame fingerprint H P is generated using a relatively simpler hash function (e.g., 256 bits) that is computationally less expensive. The fingerprint of the predicted frame, H P If V relies on fingerprints of I-frames, it can be generated in a computationally economical way using simpler hash functions without significantly reducing the overall level of security.

[0022] In one embodiment, data derived from an I-frame is cached for reuse in generating fingerprints for different predicted frames that directly or indirectly reference the same I-frame, thereby reducing the total amount of computation by avoiding duplication.

[0023] In one embodiment, the fingerprint of a predicted frame is generated by hashing a combination that further includes data derived from additional I-frames that the predicted frame references directly or indirectly. The I-frame and the additional I-frames may belong to the same or two different GOPs. For example, the predicted frame may be a bidirectionally predicted frame that references two different I-frames; in this embodiment, data derived from both of these I-frames may be included in the combination that is hashed to generate the fingerprint of the predicted frame. Note that linking the fingerprint of the predicted frame to both of these I-frames does not introduce any new error scenarios; in fact, if the fingerprint of either I-frame cannot be verified at the receiver, decoding of the predicted frame will fail anyway because the reference chain back to that I-frame is broken.

[0024] In some embodiments, the signature of a video sequence consists of multiple countersignatures provided for different segments of the video sequence. This allows for continuous signature verification during playback and supports signing of video data encoding live video streams, which is particularly useful in video surveillance applications. The countersignatures may be included in the metadata of the video sequence.

[0025] In one embodiment, separate countersignatures are provided for separate GOPs. A countersignature for a GOP may include fingerprints of frames in that GOP as well as fingerprints of frames (e.g., I-frames) in adjacent (e.g., subsequent) GOPs. This protects the signed video sequence from unauthorized substitution, removal, or insertion of GOPs.

[0026] In a second aspect of the present invention, there is provided a device arranged to carry out the above method. Broadly speaking, the second aspect of the present invention shares the effects and advantages of the first aspect and can be implemented with a corresponding degree of technical versatility.

[0027] The present invention further relates to a computer program containing instructions for causing a computer to carry out the above-described method. The computer program may be stored or distributed on a data carrier. As used herein, a "data carrier" may be a transitory data carrier, such as a modulated electromagnetic wave or light wave, or a non-transitory data carrier. Non-transitory data carriers include volatile and non-volatile memories, such as permanent and non-permanent storage media of the magnetic, optical, or solid-state type. Furthermore, within the scope of "data carrier," such memories may be fixedly mounted or portable.

[0028] In general, all terms used in the claims should be interpreted according to their ordinary meaning in the art unless expressly defined otherwise. All references to "a / an / the element, apparatus, component, means, step, etc." should be openly interpreted as referring to at least one instance of the element, apparatus, component, means, step, etc., unless expressly stated otherwise. The steps of any method disclosed herein do not have to be performed in the exact order described, unless expressly stated otherwise.

[0029] Aspects and embodiments will now be described, by way of example, with reference to the accompanying drawings. [Brief explanation of the drawings]

[0030] [Figure 1A] FIG. 2 illustrates a segment of a predictively coded video sequence, along with frame-by-frame fingerprints and signatures generated from them. [Figure 1B] FIG. 2 shows a segment of a predictively coded video sequence further including bidirectionally predicted frames. [Figure 2] FIG. 1 illustrates the data flow in the hash-based computation of a concatenated fingerprint of a predictively coded video frame (h), and the decoding operation (d). [Figure 3]FIG. 2 illustrates data flow in a particular method for signing predictively coded video data, according to one embodiment of the present invention. [Figure 4] FIG. 2 illustrates data flow in a particular method for signing predictively coded video data, according to one embodiment of the present invention. [Figure 5] FIG. 1 illustrates a device arranged to perform signing of predictively coded video data, according to an embodiment of the present invention; [Figure 6] 3 is a flowchart of a method for signing predictively coded video data according to an embodiment of the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0031] Aspects of the present disclosure will now be described more fully hereinafter with reference to the accompanying drawings, in which certain embodiments of the invention are shown. These aspects may, however, be embodied in many different forms and should not be construed as limiting; rather, these embodiments are provided as examples so that this disclosure will be thorough and complete, and will fully convey the scope of all aspects of the invention to those skilled in the art. Like numbers refer to like elements throughout the specification.

[0032] Referring to FIG. 6, a method 600 for signing predictively coded video data will be described. The method 600 for providing a signature based on a coded video sequence may be performed by a suitably arranged general programmable computer (e.g., with associated input and output interfaces), in particular by the device 500 shown in block diagram form in FIG. 5. The device 500 comprises a processing circuit 510, a memory 520, and an input / output interface 530 adapted for bidirectional communication with an external memory 590 that, at some operational stages, stores the coded video sequence to be signed. The device 500 and the external memory 590 may be owned and operated by different entities, such as an external service or when the signatures are provided by a common entity. The (internal) memory 520 of the device 500 may be suitable for storing a program 521 having software instructions for performing the method 600, cryptographic information (e.g., private keys) for generating the signature, as well as logs, configuration files, and data supporting various internal housekeeping procedures. Device 500 may be provided as a local computer or server, or may be implemented in a distributed manner based on networked (cloud) processing resources, and may, inter alia, be integrated in a digital video camera, such as a video camera adapted for surveillance applications.

[0033] In a first step 610 of method 600, an encoded video sequence is obtained. The video sequence includes at least one I-frame containing independently decodable image data and at least one predicted frame containing image data decodable by reference to at least one other frame. I-frames and predicted frames should not be confused with the plaintext video frames that these data structures encode. It should be understood that predicted frames may include unidirectionally predicted frames, bidirectionally predicted frames, or a mixture of both types. Obtaining the video data may include accessing a memory in which the video data is stored (e.g., external memory 590 in FIG. 5), downloading the video data, and / or receiving a transmission with the video data. The video sequence may be encoded in any suitable format, such as ITU-T H.265, AV1, and the previously mentioned ITU-T H.264.

[0034] In a second step 612, the fingerprint H of each I-frame is calculated. I is generated. The fingerprint of each I-frame, H I can be, for example, a hash of the image data of an I-frame, or a hash of a subset of this. I may be temporarily stored in anticipation of the fourth step 616 of the method 600.

[0035] In a third step 614, the fingerprint H of each predicted frame is calculated. Pis generated by hashing a combination of data derived from a predicted frame and data derived from I-frames to which the predicted frame directly or indirectly references. However, the fingerprint of a predicted frame is independent of any further predicted frames to which it directly or indirectly references. This independence allows a recipient of the signed video data to decode predicted frames and verify their authenticity even when the recipient is unable to verify the fingerprints of adjacent predicted frames (for simplicity of this presentation, we will assume that the only exception is the inability to verify fingerprints). The fingerprint calculation is H P =h({λ, π}) where h is a hash function, λ is data derived from the I-frame, and π is data derived from the predicted frame. The hash function (or one-way function) may be a cryptographic hash function that provides a level of security deemed appropriate, taking into account the sensitivity of the video data being signed and / or the value that would be at stake if the video data were manipulated by an unauthorized party. Three examples are SHA-256, SHA3-512, and RSA-1024. The hash function shall be predetermined (e.g., reproducible) so that the fingerprint can be reproduced when verified. The curly bracket notation {} refers to a general data combination operation, which may include concatenating data linearly (juxtaposed) or in various staggered arrangements. The combination operation may further include arithmetic operations on the data, such as bitwise OR, XOR, multiplication, division, or modulo operations.

[0036] Examples of the selection of λ and π are now described. In one implementation of the third step 614, the data derived from the predicted frame is the image data of the predicted frame. In other words, the fingerprint of the nth predicted frame is H Pn =h({λ, Pn}) where Pn denotes the image data of the nth predicted frame. The combination may further include additional information σ. H Pn =h({λ, Pn, σ}) where σ may be a cryptographic salt of the type used in the prior art to make duplicate or common security codes (e.g., passwords, signatures) harder to identify. Among other things, the presence of a salt may stop an unauthorized party from guessing what hash function is being used. Potentially useful salts are known to include active internal counters, random numbers, and signing time and place values. Furthermore, the fingerprint of the nth predicted frame may be based on a subset of the image data of the first predicted frame, where the subset is extracted in a pre-agreed manner that may be repeatable at the receiver side. In practice, the image data P encoding the complete nth predicted frame may be n Instead of σ, image data may be used that encodes only certain image bands or certain macroblocks of a video frame according to a predetermined segmentation that applies across the entire frame and that may be invariant over the duration of the video sequence. Instead of or in addition to the encryption salt, the additional information σ may include the sequence number n of the frame, which allows the removal of corrupted frames and frame reordering to be detected.

[0037] In another implementation of the third step 614, the data derived from the predicted frame includes at least a hash of the image data of the predicted frame: H Pn =h({λ, h(Pn)}) In variations of this implementation, the inner and outer hash operations may be performed using different hash functions. In computing systems where large memory operations are expensive, this implementation has the advantage that the hash h(Pn) is a data set of manageable size that contributes negligibly to the cost of forming the combination {λ, h(Pn)}, for example, by transferring it to a memory space where processing circuitry 510 can look up input data to the arithmetic operations.

[0038] In yet a further implementation of the third step 614, which can be freely combined with the two previous options, the data derived from the I-frame is a fingerprint of the I-frame. H P =h({H I , π}) An embodiment in which this choice of λ is combined with π=Pn is shown in Figure 3. According to a variant, the data derived from the I-frame is a hash of the image data I of the I-frame, but this hash is a hash of the fingerprint H of the I-frame generated in the second step 612. I , which will be used later to provide the signature. On the other hand, the fingerprint of the I-frame, H I can be generated using a different, possibly more complex, hash function g. Figure 4 shows an example of this setup, where π = h(Pn). On the other hand, alternatively or in addition, the fingerprint H of the I-frame I may be generated by hashing a combination of the image data of the I-frame and additional information, such as a cryptographic salt. A still further option is to generate the fingerprint H by hashing a combination of the hash of the image data of the I-frame and additional information. I The fingerprinting may use the entire image data of an I-frame, as described above for predicted frames, or an extracted subset of the image data that encodes a particular image band or block.

[0039] Common to all of the above options for implementing the third step 614 of method 600, it is advantageous to temporarily cache 614.1 the data λ derived from the I-frame for later reuse when the third step 614 is executed again to generate fingerprints of different predicted frames that directly or indirectly reference the same I-frame.

[0040] Further development of the third step 614 specifically handles the case where a predicted frame directly or indirectly references two I-frames. A fingerprint for the predicted frame is then generated based on a combination of data λ, λ derived from both of these I-frames in addition to data π derived from the predicted frame itself. H P =h({λ1, λ2, π})

[0041] Execution of method 600 continues in a fourth step 616 with providing a signature of the video sequence including the generated fingerprint. The signature may be formed by collecting the generated fingerprint in a so-called document (a text file or another data structure) and signing the document. The document signature may be generated, for example, by asymmetric cryptography, i.e., using a private key from a key pair whose public key is pre-shared with the recipient so that the recipient can verify the signature. Thus, the video sequence signature may consist of the document and the document signature. A recipient in possession of the video data thus signed can verify its authenticity by the following procedure. 1. Attempt to verify the document signature 2. If the document signature verifies successfully, try to verify the fingerprint on the document. 3. If all fingerprints in the document are successfully verified, conclude that the video data is authentic (validation). Typically, verification in step 2 involves replicating the fingerprinting operation supposedly performed by the sender, i.e., recalculating the fingerprint, while verification in step 1 typically involves an asymmetric signature setup where signing and verifying are different cryptographic operations corresponding to private / public keys. Other combinations of symmetric and / or asymmetric verification operations are possible without departing from the scope of the invention.

[0042] The fourth step 616 may include associating the signature with the video sequence. For example, if the video data has a predetermined data format that includes metadata fields in addition to image data, the signature may be associated with the video data by being included in such metadata fields. In the ITU-T H.264 format, the signature may be included in a Supplementary Enhancement Information (SEI) message in the video bitstream. In the AV1 standard, the signature may be included in a metadata Open Bitstream Unit (OBU). Alternatively, the fourth step 616 may include establishing the association between the video data and the signature outside of the data format, for example, by forming a data structure (container) that includes both the video data and the signature and that will be transmitted to the intended recipient. A still further option is to separate the video data from the signature so that the (smaller) signature can be transmitted over a secure channel, while the (larger) video data can be transmitted over any channel. These alternatives to including the signature in a metadata field are advantageous in that they do not require the ability to modify the video data (e.g., permission granted by the video data owner) and can therefore be performed by entities that do not have write access to the video data.

[0043] In some embodiments, a fourth step 616 provides a signature comprised of multiple countersignatures to enable continuous verification of the countersignatures while decoding the video data. For example, the video data can be provided 616 with one countersignature for each GOP. Thus, a countersignature may include fingerprints of frames in the GOP and, optionally, fingerprints of frames in preceding or following GOPs. This type of countersignature can consist of a document collecting the stated fingerprints and a signature of the document. To enable detection of unauthorized removals and reordering within a GOP, a sender and receiver can agree that the order in which the fingerprints are listed in the document corresponds to the authentic frame order. A still further option is to generate a single fingerprint of the fingerprints of all frames in a GOP (along with optional fingerprints of frames in adjacent GOPs, as shown in FIG. 1A), sign the single fingerprint, and provide a signature for the video data consisting of the single fingerprint and this signature. Here, a single fingerprint can be considered to be a document, or a single fingerprint can be considered to complement a document that is itself empty. See Figure 1A and the accompanying explanation as to why this setup protects against unauthorized replacement, removal, or insertion of GOPs.

[0044] To perform GOP-level signatures, the fourth step 616 may include sub-step 616.1 of identifying multiple GOPs and sub-step 616.2 of providing a counter-signature for each identified GOP. Specifically, the GOP-level counter-signature may be included in the metadata of the first or last frame of the GOP, or the counter-signature may be inserted into the video bitstream near these frames.

[0045] Aspects of the present disclosure have been primarily described above with reference to a few embodiments, however, as those skilled in the art will readily appreciate, embodiments other than those disclosed above are equally possible within the scope of the invention as defined by the appended claims. [Explanation of symbols]

[0046] 500 devices 510 Processing Circuit 520 memory 530 Input / Output Interface 590 External Memory

Claims

1. A method (600) for signing predictively coded video data, comprising: obtaining (610) a coded video sequence including at least one I-frame (I) containing independently decodable image data and at least one predicted frame (P1, P2, P3, P4) containing image data that can be decodized by reference to at least one other frame; The fingerprint of each I-frame (H Ｉ ) (612); A fingerprint (H) of each predicted frame is generated by hashing a combination of data derived from the predicted frame and data derived from an I-frame that the predicted frame directly or indirectly references. Ｐ ) (614); providing (616) a signature of the encoded video sequence including the generated fingerprint; A method (600) wherein the fingerprint of the predicted frame is generated by a hashing operation that does not include as input data derived from any further predicted frames that the predicted frame directly or indirectly references.

2. The method of claim 1 , wherein the data derived from the predicted frame includes its image data.

3. The method of claim 1 , wherein the data derived from the predicted frame includes at least a hash of image data for the predicted frame.

4. The method of claim 1 , wherein the data derived from the I-frame is the fingerprint of the I-frame.

5. The method of claim 4 , wherein the fingerprints of each of the I-frame and the predicted frame are generated using different hash functions ( 612 , 614 ).

6. 6. The method of claim 1, further comprising caching (614.1) the data derived from the I-frame for reuse in generating fingerprints of different predicted frames that directly or indirectly reference the same I-frame.

7. 2. The method of claim 1, wherein the fingerprint of the predicted frame is generated by hashing (614) a combination that further includes data derived from additional I-frames that the predicted frame references directly or indirectly.

8. The signature of the encoded video sequence comprises a plurality of countersignatures, and providing the signature (616) comprises: Identifying (616.1) a plurality of groups of pictures (GOPs), each group being independently decodable; and providing a countersignature for each identified GOP (616.2).

9. The method of claim 8 , wherein the counter-signature comprises the fingerprint of the frame in the GOP and fingerprints of frames in adjacent GOPs, in particular fingerprints of frames in subsequent GOPs.

10. The method of claim 1 , wherein the signature is included in metadata of the encoded video sequence.

11. The method of claim 1 , wherein the signature of the encoded video sequence is cryptographically signed.

12. The method of claim 1 , comprising generating (614) a fingerprint of the unidirectional predicted frame.

13. The method of claim 1 , comprising generating a fingerprint (614) of a bi-predictive frame.

14. A device (500) comprising processing circuitry (510) arranged to perform the method of any one of claims 1 to 13.

15. A non-transitory computer-readable storage medium having stored thereon instructions for performing the method of any one of claims 1 to 13 when executed on a device having processing capabilities.

Images