BRPI1003850A2 - method to authenticate video - Google Patents

Abstract

METHOD FOR AUTHENTICATING VIDEO. As the cost of equipment for filming, transmitting and storing videos decreases, the use of video becomes more and more popular in different applications. Given the ease with which videos can be manipulated, a video may have no value as legal proof, if there is no way to authenticate it, that is, to prove that the video was not edited fraudulently and was generated by a specific camcorder . Thus, authentication of video content becomes extremely important. We propose a video authentication system that calculates the video's perceptual hashing function, encrypts it and inserts the resulting bits as a series of dots in a small spatial region of the video. The insertion of an authentication mark can be done on a conventional computer or on a device specially dedicated to the task. The proposed technique has the following properties: (A) The authentication code is stored in the video itself. (B) The video does not suffer any degradation in quality, except in the small spatial region destined to receive the authentication code. (C) The insertion of the authentication code can be done in embedded systems, as it requires little processing and memory. (D) The system can detect insertion, removal and replacement of frames, and spatially locate the modified region of the video. (E) Authenticated video resists different levels of compression, analog-to-digital conversion, adjustment of brightness, contrast and coloring, and can be compressed using a wide variety of codecs. (F) Depending on the encryption algorithm used, the generation of falsely authenticated videos by unauthorized people can become extremely difficult.

Description

METHOD FOR AUTHENTICATING VIDEO

FIELD OF INVENTION

The present invention relates generally to a method for authenticating video content and in particular to a method for generating and storing the authentication code.

BACKGROUND OF THE INVENTION

There are currently two basic ways to authenticate video content:

(A) An authentication code stored independently of the video.

(B) Authentication watermark, where the authentication code is stored within the video itself.

(A) Authentication Code

Video content can be authenticated using a cryptographic authentication code (such as message authentication code or digital signature). Only in this case any video distortion would be detected as a change. To distinguish "harmless" distortions from "malicious" distortions, functions called perceptual hashing (also called robust hashing or robust digital signature) have been proposed.

For example, Lin et al. [C.-Y. Lin, S.-F. Chang, "Generating Robust Digital Signer for Image / Video Authentication," Multimedia and Security Workshop at ACM Multimedia'98, Bristol, UK, September 1998] propose a compression-resistant perceptual hashing function suitable for compressed images and videos using DCT (discrete cosine transform). The idea is that the relationship between two DCT coefficients of the same block should remain unchanged after the quantization process. For example, if Fp (η)> F (η), where Fp (n) is the coefficient DCT at position η of block ρ and Fq (n) is the coefficient at position η of block q, then after compression ratio F'p (n)> F ^ {n) is guaranteed, where F'p (n) and F'q (') are the quantized coefficients.

As another example of perceptual hashing, the article by C.-S. Lu, H.-Y. M. Liao, "Structural Digital Signature for Image Authentication: An Incidental Distortion Resistant Scheme," IEEE T. Multimedia, Vol. 5, no. 2, pp. 161-173, June 2003 proposes to use the inter-scale relationships of the wavelet coefficients to obtain a perceptual hashing function that can extract the image structure; and the article by C.-S. Lu, C.-Y. Hsu, "Geometric Distortion-Resilient Image Hashing Scheme and Its Applications on Copy Detection and Authentication," Multimedia Systems, 11 (2), p. 159-173, 2005 makes this hash resistant to geometric distortions.

Perceptual hashing can be encrypted by generating the authentication code. The problem with this approach is that the authentication code must be stored and transmitted independently of the video, which is a drawback in practice. (B) Watermark

The authentication watermark is an invisible signal inserted into the video that allows you to detect changes. You can insert watermarks that do not depend on the video content. However, the most advanced authentication watermarks often calculate perceptual hashing of the video, encrypt the hashing and insert the resulting authentication code into the video itself [F. Bartolini, A. Tefas, M. Barni, I. Pitas, "Image Authentication Techniques for Surveillance Applications," Proceedings of the IEEE, vol. 89, no. 10, pp. 1403-1418, 2001].

For example, US Patent 5,960,081 of September 28, 1999 to T. Vynne, F. Jordan, entitled "Embedding a Digital Signature in a Video Sequence," proposes to embed a 32-bit digital signature in the motion vector. from the video. US Patent 6,064,764 of May 16, 2000 to V. Bhaskaran, V. Ratnakar entitled "Fragile watermarks for detecting tampering in images" proposes to insert the image signature directly into the high frequency coefficients of JPEG compressed images. . US Patent 9,419,671 of November 29, 2005 to M. Abdel-Mottaleb1 entitled "Image and video authentication system" proposes to extract the digital signature from the image / video and embed it in the high frequency coefficients of the image. / video in the frequency domain.

The problem with the watermark is that for a good level of security, a considerable amount of bits must be inserted into the video robustly, which considerably degrades the video quality. An alternative is to calculate the authentication code for already compressed videos and insert it into the compressed video assuming it will not change. In this case, the video cannot be changed, such as a digital to analog conversion or a change in the compression level.

US Patent 6,996,273 of February 7, 2006, "Robust Recognizer of Perceptually Similar Content" and US Patent 6,671,407, of December 30, 2003, "System and method for hashing digital images" have the objective of recognizing digital images. For this purpose, these documents propose to calculate a hashing that enables the visually similar images to be identified quickly. This allows you to quickly search a database for similar images, search for pirated copies, and so on.

OBJECT OF THE INVENTION

It is an object of the present invention to provide a video authentication method wherein information enabling authenticity to be verified is encrypted and inserted into the video itself visibly as a series of dots in a small spatial region of the video. SUMMARY OF THE INVENTION

Method for authenticating video by calculating the value of the video's perceptual hashing function; encrypting said perceptual hashing value to generate the authentication code; insert the bits of the authentication code as a visible mark on the video; decode the authentication code to generate the extracted authentication code; decrypt the extracted authentication code to generate the extracted hashing function; recalculate the video's perceptual hashing function to generate recalculated hashing; and compare extracted hashing with recalculated hashing to confirm the authenticity of the video.

BRIEF DESCRIPTION OF THE FIGURES AND ATTACHMENTS

Annex 1 is an illustration of dividing a frame into smaller units we call "space cubes."

Annex 2 is an illustration of an authenticated video frame.

Annex 3 is an illustration of a video frame with verified authenticity.

Appendix 4 is an illustration of a tampered with detected and localized video frame.

Figure 1 is a flowchart of the system implementation proposed in the present invention in a conventional computer.

Figure 2 is a flowchart of implementing authentication insertion in an embedded system or dedicated appliance.

DESCRIPTION OF THE PREFERRED EMBODIMENT

As the cost of video filming, streaming and storing equipment decreases, the use of video becomes increasingly popular in different applications.

However, due to the availability of sophisticated video manipulation tools, videos may be of little value as legal proof, as there may always be questions about whether they were manipulated fraudulently. Thus, it is necessary to authenticate them, that is, to have some means to prove that these videos have not been altered and to demonstrate their origins (ie to determine which camera or equipment generated which video).

There are two basic ways to authenticate video content:

1. An authentication code stored independently of the video. In this case, it is necessary to transmit and store the code separately from the video, which is a drawback in practice.

2. Authentication watermark, where the authentication code is stored within the video itself. In this case, either watermark insertion tends to considerably degrade video quality and / or authenticated video cannot be rendered harmless such as video recompression or brightness / contrast adjustment.

The present invention has the desirable characteristics of both types of systems: the authentication code is stored in the video itself; Inserting the code doesn't degrade the video (the resulting video just gets a little smaller); and authenticated video is recoding resistant.

The system proposed in the present invention first calculates the perceptual hashing function h of the video (also called robust hashing). A perceptual hashing is code that identifies the visual content of the video. Then the hashing (along with some other information that allows for frame insertion, removal, and replacement, described in detail in the "authentication code" section below) is encrypted, generating the authentication code. The bits of the authentication code are then inserted as a visible mark within the video. An example of the visible mark capable of storing the bits of the authentication code is a series of dots inserted into the side columns of the video, called the dot code here. However, this invention does not exclude other ways of storing bits visibly, for example by different colors and / or gray levels.

To verify authenticity, that is, to verify whether the video has been altered or not, the point code is decoded, generating the extracted authentication code. This code is decrypted, generating the extracted hashing he. On the other hand, perceptual hashing is recalculated from the video, generating recalculated hashing hr. If he = hr, the authenticity is confirmed. Otherwise, a tamper has been detected. We use the approximately equal symbol (=) instead of equal (=), as the video may suffer harmless distortions (for example, distortions introduced by different compression levels) without these distortions altering the visual content of the video.

The method of the present invention and the elements used in its application will now be described in detail and with reference to the Figures and the appendices.

Video Features

The method of the present invention can be applied to videos with any spatiotemporal resolution and any color system. However, for ease of description, in the present invention we will assume (without loss of generality) the authentication of 480x720 pixel and 29.97 frames per second digital video using the YCbCr color system (Y is luma, Cb and Cr are the difference-blue and difference-red chrominance components) with 4: 2: 0 chrominance subsampling (for every 4 volumes, there is an information of Cb and Cr).

Snippets and Cubes As an example in practicing the method of the present invention, the video is broken into 15-frame time slots approximately 0.5 second long. The stretch is the lowest temporal resolution for detecting changes. It was chosen 0.5 seconds because it is difficult to imagine any significant change in video shorter than this time interval. The first 8 and last 8 columns of all frames in the video will be used to store the dot code.

Annex 1 illustrates a digitized analog television frame divided into 16x24 pixel space cubes, totaling 30x29 space cubes. The starting 8 and ending 8 columns are reserved for storing the authentication code as a "point code" (CP). These columns are usually black in scanned analog videos. Thus, storing the point code in these columns does not lead to appreciable loss of information. The authentication code can be inserted into other areas of the video using different forms of visible encoding. Discarding these 16 columns, 480x704 pixels remain. A section is broken into space-time cubes. Each video cube has 15 frames χ 16 rows χ 24 columns. Thus, a section has 30x29 = 870 cubes (Annex 1). We chose multiples of 8 for the number of cube rows and columns, because many codecs work with 8x8 blocks.

Perceptual Hashing

Several perceptual hashings have been proposed in the academic literature to authenticate videos (see description of the state of the art). Many perceptual hashing uses transforms such as DCT, FFT, and wavelet that require high computational power. Without limiting the scope of the invention and as an example we describe the use of a very simple, intuitive and non-transforming perceptual hashing: the low resolution video itself quantized in space and time. The first advantage of this hash is the computational simplicity of calculating it, allowing the insertion of the authentication code to be performed on embedded systems with low processing power and low memory. The second advantage is its robustness to distortions introduced by different levels of compression. The third advantage is that the user can view the video in low resolution and decide if any detected changes were generated maliciously or accidentally.

The perceptual hashing used in the apparatus of the present invention is an 8- or 4-bit number vector composed of the arithmetic means of the Y, Cb, and Cr components within the cubes.

Upon insertion of the authentication signal, each element of this vector is quantized by 4 bits to decrease the number of bits. That is, an average of 12 bits per cube, since each cube has the components Y, Cb and Cr. Thus, the perceptual hashing h of a section will have 12 bits χ 870 cubes = 5220 bits, which must be entered as a point code.

When verifying authentication, 4-bit h and perceptual hashing is extracted from the point code and 8-bit hr perceptual hashing is recalculated from the video. Differences between h and hr elements are calculated element by element. If the difference of any element is greater than a threshold, it is considered that the cube represented by the element has changed. Experimentally, we found that threshold 28 is convenient as it does not cause false positives.

authentication code

The perceptual hashing h of a portion will be encrypted using a secret key k, generating the authentication code C which will be stored as point code. Secret key k is obtained by properly combining the following information:

1. A secret number stored inside the authentication insert device or program.

2. The current date and time.

3. The identity of the camera. It is a number that identifies the equipment.

4. The serial number of the video excerpt.

Information 2, 3 and 4 are also stored unencrypted in the video so that the video authenticity verification program can recreate the k key. For added security, a random number a is concatenated with perceptual hashing h before encryption: C = encriptak (a + h), where C is the authentication code and "+" is the bit concatenation operation. The encryption function can be any suitable cryptographic function. In decryption, the number a is extracted along with h and thrown away. We used a 32-bit random number, resulting in the 32 + 5220-bit C authentication code.

Point Code

In the present invention, we call point code the technique used to insert the bits into the side columns of the video. Our system uses one pixel to store each bit: one black pixel represents bit 0; and a white pixel represents bit 1. Thus, the 8 starting and ending 8 columns fit 2x8x480 = 7680 bits. Other ways of storing bits can be used visibly, for example by using different gray levels or by using the colors of pixels.

Embedded Information

In order for the authentication insertion system to be capable of running on low memory embedded systems as intended by the present invention, the one-pass authentication code / is stored as a point code on the next / / 1 leg. This allows the system to authenticate a video without storing any entire frames in memory. All 15 frames of the i + 1 section will contain the same point code, so that the point code will be static for approximately half a second, ensuring the robustness of the stored information. The following information is stored in the i + 1 snippet:

1. 32 sync bits that serve to indicate the beginning of each leg. If i is even 32 bits zero are entered, and if i is odd 32 bits one are entered. Thus, the authenticated video will have 15 consecutive frames with the zero sync signal, followed by 15 frames with the sync signal one, etc. The beginning of a stretch is indicated by changing the values of the sync bits.

2. 32-bit camera identity (a number that identifies the camera).

3. 32 bits used to represent date and time when the i snippet was recorded.

4. 32 bits representing the number i of the section.

5. The 32 + 5220-bit C authentication code, calculated as described above.

6. 32 bits that indicate "or-exclusive" of all previously entered DWORDs. This checksum indicates whether there was any error in writing / retrieving the point code.

Lighting and coloring correction

Annex 2 illustrates an authenticated dot-coded (CP) video frame on the eight side columns and six lighting / color correction rectangles on the lower corners (numbered 1, 2, 3, 4, 5, 6). A video may change in brightness, contrast and color. To make the system robust to these changes, six reference color rectangles for the automatic normalization of brightness, contrast and coloration are inserted at the bottom of the eight start and end columns (Annex 2). The six colors entered are, in color system, (Y, Cb, Cr):

• Dark gray (64,128,128) and light gray (192,128,128) to calibrate Y;

• Green (128,64,128) and purple (128,192,128) to calibrate Cb;

• Blue (128,128.64) and red (128,128,192) to calibrate Cr.

In authenticity checking, these color squares are used to return the authenticated video to the same level of brightness, contrast, and color as the video at the time of insertion. This ensures the robustness of the system to changes in brightness, contrast and color.

Authenticity Check

The authenticity check of a video is performed on a conventional computer. At this stage, the apparatus of the present invention asks the user for the identity of the encrypted camera. Only with this number will the system verify the authenticity. This allows you to sell individual licenses of the authenticity verification program.

In verifying authenticity, the Ce authentication code is extracted from the video's point code, along with the camera's identity, date and time, and the snippet number. Then, the checksum of the bits is calculated and compared to the extracted checksum to check for any errors in the dot code reading.

The method of the present invention verifies that the identity of the extracted camera matches the user supplied code. The method will only check if there is a match.

From the extracted information, the system recalculates the encryption key k. Then the Ce code is decrypted, generating the extracted hashing he = decriptak (Ce), quantized at 4 bits. On the other hand, 8-bit quantized perceptual hashing is recalculated from the video, generating the recalculated hashing hr. The differences between the elements of he and hr are calculated.

If no difference is greater than a certain threshold, the stretch is considered to be authentic. Annex 3 illustrates a video frame with verified authenticity. Note in the lower left corner the low resolution frame extracted and used to verify authenticity.

If there are any elements whose difference is greater than the threshold, then the cube corresponding to the element has been changed. Annex 4 illustrates a tampered with detected and localized video frame. Note in the lower left corner of attachment 4 is the extracted low resolution frame used to locate the tamper. Experimentally, it was found that the appropriate threshold is close to 28.

The system checks whether the segment number is in sequential order. This allows to detect removals, insertions and replacements of entire sections. The system ignores the removal of one or two frames from a portion, as this may happen accidentally during video transmission.

Safety

The system recognizes any piece of authenticated video as authentic, except some start and end frames of the piece whose authenticity cannot be verified by the system.

The proposed system can detect the following types of changes:

1. Significant changes in the averages of component Y, Cb and Cr of the cubes. In this case, in addition to detecting the change, the system spatially locates the change.

2. Removing, inserting, and replacing frames or snippets from video. This is possible by using the number of the inserted segment as a point code. The proposed system cannot detect the following changes:

1. Minor changes to the Y1 Cb and Cr component averages of the cubes.

A video authenticated by the proposed system may suffer the following distortions without the system detecting them as tampering:

1. It can be compressed at different compression levels using virtually any codec (encoder-decoder).

2. Resists analog-digital and digital-analog conversion.

3. Resists adjusting brightness, contrast and color.

Implementations:

The method of the present invention may be implemented in two ways:

1. Figure 1 is a flowchart of the implementation of authentication insertion on a conventional computer, where a program named inserts receives a video named input.ext and generates the authenticated video named marked.ext with the authentication code inserted as code. of points. Then another program named checks receives an authenticated video named marked.ext and generates a video named verified.ext and / or a report named report.txt describing whether or not the video is confirmed as authentic.

2. Figure 2 is a flowchart of the implementation of authentication insertion on an embedded computer or dedicated device, where the video signal from a camera (or some other device) is authenticated in real time by a device. Dedicated independent. The output is another video signal (analog or digital) with dot code. This signal is stored on a storage device (computer, DVD recorder, etc.) called labeled.ext. The authenticity of this video is verified on a conventional computer using the program called checks.

The proposed method has the following properties:

(A) The authentication code is stored in the video itself.

(B) Video does not suffer any quality degradation except in the small spatial region intended to receive the authentication code.

(C) Entering the authentication code can be done on embedded systems as it requires little processing and memory.

(D) The method can detect insertion, removal and replacement of frames and spatially locate the modified region of the video.

(E) Authenticated video resists different levels of compression, analog-to-digital conversion, brightness, contrast, and color adjustment, and can be compressed using a wide variety of codecs. (F) Depending on the encryption algorithm used, generating videos falsely authenticated by unauthorized persons can become extremely difficult.

Claims (5)

1. Method for authenticating video, where the authentication code is stored as a visible mark, characterized by the fact that it consists of: a) breaking the video into units called time slots; b) break the time segments into smaller units called space-time cubes; c) determining the arithmetic means of the Y1 Cb and Cr components of the space-time cubes and using said space-time cube averages as perceptual hashing of the video; d) encrypting said perceptual hashing function value to generate the authentication code; e) inserting the bits of the time pass authentication code as a visible mark on the next time pass of the video; f) decoding the authentication code to generate the extracted authentication code; g) decrypting the extracted authentication code to generate the extracted hashing function; h) recalculating the perceptual hashing function of the video to generate the recalculated hashing; and (i) compare extracted hashing with recalculated hashing to confirm the authenticity of the video. Method according to claim 1, characterized in that the authentication code of a portion is stored as a point code in the following portion, which occupies a small spatial region of the eight beginning and ending columns of the video. Method according to claim 1, characterized in that six rectangles (1 to 6) with the reference colors Y, Cb, Cr are inserted at the bottom of the eight starting and ending columns of the video for automatic normalization. of brightness, contrast and coloring. Method according to claim 1, characterized in that a sequential number is inserted which makes it possible to detect insertion, removal and replacement of video frames. Method according to claim 1, characterized in that an identifier of the camcorder generating the video is inserted, enabling the origin of the video to be determined.