Video detection method based on artificial intelligence

Abstract

The present invention provides a video detection method based on artificial intelligence, the method comprising: receiving a first part of a decryption private key; determining an integrity index associated with the video image; combining the first part of the decryption private key with the integrity index to generate the decryption private key; using the generated decryption private key to attempt to decrypt the video image content; if the integrity index indicates that the video image has not been tampered with, and the generated decryption private key is generated in a correct manner, then the content is successfully decrypted. The present invention proposes a video detection method based on artificial intelligence, which can verify whether the forensic video has been tampered with by embedding authentication data into a video file without additional information, thereby improving the operating efficiency of the video detection process.

Description

Technical Field

[0001] The present invention relates to video security, and in particular to a video detection method based on artificial intelligence. Background Art

[0002] Authentication based on forensic video images is widely used in a variety of application scenarios. However, due to the development of digital technology, videos are becoming easier to forge, and anyone can edit videos through image processing software. Therefore, the security mechanism of forensic videos is becoming more and more urgent. Video authentication technology is used to ensure the consistency of forensic video content, and it is necessary to detect whether the forensic video has been tampered with. Traditional signature-based video authentication technology mainly extracts the feature information of the video, and after simplifying the feature information, the simplified feature information is saved independently as authentication data. When it is necessary to confirm whether the forensic video has been forged, the verifier must provide authentication data to authenticate the forensic video. However, its disadvantage is that the legitimate user must safely save the authentication data, otherwise it cannot be used during authentication. When the forensic video is forged, it is necessary to first extract the video compression encoding of the area, which puts higher requirements on time and space and has low operating efficiency. Summary of the invention

[0003] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes a video detection method based on artificial intelligence, comprising:

[0004] receiving a first portion of a private decryption key; determining an integrity index associated with the video image;

[0005] combining the first portion of the decrypted private key with the integrity index to generate the decrypted private key;

[0006] Use the generated decryption private key to attempt to decrypt the video image content;

[0007] If the integrity index indicates that the video image has not been tampered with and the generated decryption private key is generated in a correct manner, the content is decrypted successfully;

[0008] Determining the integrity index associated with the video image further includes:

[0009] Based on the target security level, determining a first private key matrix;

[0010] Analyze the video image to obtain a sample detection matrix UM;

[0011] Each sample detection matrix and its transposed matrix UM T Multiply to generate a second-order symmetric matrix M corresponding to each sample detection matrix;

[0012] performing at least one arithmetic matrix calculation between each second-order symmetric matrix and the first private key matrix to generate a second private key matrix; and

[0013] The second private key matrix is ​​used as an integrity index corresponding to the video image.

[0014] Advantageously, said determining an integrity index associated with said video image further comprises determining a checksum associated with a portion of said video image.

[0015] Preferably, parsing the video image to obtain the sample detection matrix UM further includes parsing the video image to obtain at least one subset matrix of static length or dynamic length.

[0016] Preferably, the method further comprises:

[0017] Based on the target security level, an initial parameter set is determined, and the initial parameter set is sent to the smart cloud, so that the smart cloud uses the initial parameter set to determine the integrity of the video image stored in the smart cloud to generate a sample summary.

[0018] Preferably, the method further comprises:

[0019] The initial parameter set is sent to the client; the client uses the initial parameter set to determine the integrity of the video image stored in the client to generate a video summary to be detected; and the sample summary is compared with the video summary to be detected to determine whether the tampering of the copy of the video image is detected.

[0020] Preferably, the comparing the sample summary with the summary of the video to be detected further comprises transmitting the summary of the video to be detected to the intelligent cloud for performing the comparison on the intelligent cloud, or transmitting the sample summary to the client, wherein the client performs the comparison.

[0021] Compared with the prior art, the present invention has the following advantages:

[0022] The present invention proposes an artificial intelligence-based video detection method, which embeds authentication data into a video file and can verify whether the forensic video has been tampered with without additional information, thereby improving the operating efficiency of the video detection process. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a flow chart of a video detection method based on artificial intelligence according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following and the accompanying drawings illustrate the principles of the present invention. Figure 1The present invention is described in detail in conjunction with one or more embodiments of the present invention. The present invention is described in conjunction with such embodiments, but the present invention is not limited to any embodiment. The scope of the present invention is limited only by the claims, and the present invention encompasses numerous alternatives, modifications, and equivalents. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. These details are provided for illustrative purposes, and the present invention may be implemented according to the claims without some or all of these specific details.

[0025] One aspect of the present invention provides a video detection method based on artificial intelligence. Figure 1 4 is a flow chart of a video detection method based on artificial intelligence according to an embodiment of the present invention.

[0026] The present invention stores the detected video in the smart card of the client, and the untampered version of the video is stored in the smart cloud. The video detection agent is used to perform an initial analysis of multiple storage locations including the video in the smart cloud. The storage location analyzed may include the location of all data or any selected continuous or non-continuous storage location. The location analyzed may also include the location where the data value is stored. The video detection agent performs multiple calculations on the data obtained from the smart card to generate a summary so that an illegal user cannot tamper with the video and maintain its summary.

[0027] By analyzing the video data storage location of the client, the data obtained can be compared with the values ​​obtained from the intelligent cloud to determine whether tampering has occurred. Preferably, the video detection agent can be reconfigured to perform multiple detections in the video, making it more difficult for illegal users to crack the system.

[0028] The integrity of the video concerned can be determined in the client. That is, the video detection agent can generate a summary based on the integrity of the video in the client smart card. The summary can then be analyzed using the video detection agent component and the result feedback module. If the analysis result indicates that the video has been tampered with, a variety of actions are used to protect the content, including transmitting tampering information to the content provider, preventing the client from receiving the content, and prohibiting the content from being decrypted.

[0029] In another embodiment of the present invention, the intelligent cloud can run the video detection agent, and the video detection agent results can then be transmitted to other modules of the intelligent cloud. Based on the results, the intelligent cloud can then determine whether to continue transmitting content to the client or stop sending content to the client.

[0030] When combining the video detection agent results, the intelligent cloud can determine the integrity value of the transmission, so that the decryption private key is used to interpret the content. The decryption private key can be based on a combination of multiple parts of the algorithm. Each time a new private key is used, the algorithm combining the multiple parts is changed. The algorithm can also be encrypted and transmitted to the client.

[0031] In addition, the video detection agent can be downloaded to the smart card periodically before the content is sent or any combination. The video detection agent can also delete itself after the analysis is completed, so that illegal users cannot browse.

[0032] The intelligent cloud determines a summary, for example, based on the detected video and the untampered version of the video. A copy of the video detection agent is then provided to the client. The client then executes the video detection agent on the copy of the detected video, using in part the parameters provided by the intelligent cloud. The result of the client executing the video detection agent includes a summary based on the integrity of the detected video. The client then provides its summary to the intelligent cloud, which compares the summary with the required summary to determine whether tampering of the client copy of the video has occurred.

[0033] Alternatively, the intelligent cloud can use the required digest to determine the decryption private key, and then transmit a portion of the decryption private key to the client to provide the required digest to the client to initiate the comparison of the client. The portion provided to the client is combined with the digest of the client to form the decryption private key.

[0034] The intelligent cloud includes an agent update module configured to start tampering detection of the selected video. The agent update module can periodically download the video detection agent to the client and execute the video detection agent in a non-tampered video copy to obtain a set of initialization parameters. The initialization parameters are used to configure the operation of the video detection agent on the client.

[0035] Preferably, the agent update module includes a result feedback module configured to receive from a client an execution result of the video detection agent based on the client. The result feedback module compares the received results to determine whether the video of the client has been tampered with. If tampering is detected, the result feedback module can perform a variety of actions, including prohibiting the transmission of content to the client, transmitting information about the tampering or a variety of other activities to the owner.

[0036] Optionally, the agent update module is also configured to determine a value combined with the result of the video detection agent of the client to form a decryption private key that can be used to interpret the received content. If the video of the client is determined to be tampered with, the decryption private key formed at the client cannot decode the received content. In addition, the client provides information indicating that video tampering has been detected to the agent update module and the result feedback module. The agent update module then performs a variety of actions based on the information including prohibiting the transmission of content to the client.

[0037] After the video detection agent is downloaded from the intelligent cloud, a digest may be generated, which may be used to detect tampering of the detected video. The video detection agent may use initialization parameters to perform detection. The result feedback module may receive a sample digest in encrypted form from the intelligent cloud, the sample digest being a digest generated based on the execution of the video detection agent of the untampered video copy. Preferably, the result feedback module is also used to combine the digest from the video detection agent with another value to generate a decryption private key to decrypt the content.

[0038] For example, any mechanism based on the integrity of the detected video can be used to generate the decryption private key. Preferably, the mechanism based on video integrity can be based on multiple static algorithms, such as checksum or ECC detection including a portion of the video to indicate whether the video has been tampered with. More preferably, the mechanism based on video integrity can be based on multiple dynamic algorithms. For example, different algorithms can be executed in the same or different parts of the video, or data can be randomly selected or rotated in time.

[0039] Preferably, a portion of a decryption private key can be received; and an integrity index associated with the video image can be determined; the portion of the decryption private key and the integrity index can be combined to generate a decryption private key; the generated decryption private key can be used to attempt to decrypt the video image content; if the integrity index indicates that the video image has not been tampered with and the generated decryption private key is generated in a correct manner, the content is successfully decrypted.

[0040] As one of the detection processes of the video detection agent, any binary digital sequence of actual measurement can be converted into a relatively small matrix. In the comparison process, both the summary to be detected and the sample summary are converted into matrices, so the comparison is performed using a matrix algorithm. If the comparison result is acceptable, the video detection agent can be configured to notify the intelligent cloud to transmit the content and the decryption private key required to decrypt the content to the client.

[0041] The summary to be detected and the sample summary may be dynamic matrices. The summary generation agent combines the binary data obtained from the smart card video and an initial parameter set of initialization parameters generated from a server. The initial parameter set may be generated using a random data generator.

[0042] The size of the summary and sample summary matrix can be determined according to the generation process and the required security level. The security level is taken as the probability of an illegal user obtaining the correct summary parameters. Assume that 0 or 1 is used in the video summary to represent different detection event results, forming a binary sequence. The vector formed by the binary sequence represents the selected sample detection. The number of sample detection summaries or matrices UM can further represent a binary sequence BS. BS is the overall sample space, which represents the number of all binary elements in the binary sequence of the detected video. UM is a sample measurement matrix with a measure of LL×NE, where LL represents the number of columns and NE represents the number of rows.

[0043] Determining the integrity index associated with the detected video may include determining a first private key matrix based on a target security level; parsing the detected video to obtain a sample detection matrix UM; and comparing each sample detection matrix with its transposed matrix UM. T Multiplying to generate a second-order symmetric matrix M corresponding to each sample detection matrix; performing at least one arithmetic matrix calculation between each second-order symmetric matrix and the first private key matrix to generate a second private key matrix; and using the second private key matrix as an integrity index corresponding to the video image.

[0044] In order to make the probability of similar points matching higher and the probability of dissimilar points matching lower, the present invention adopts an adjacency detection method, and uses a set of hash functions with preset constraints to establish multiple hash tables to map local features of the video to points in the Euclidean space, and hash these points so that the closer the Euclidean distance, the greater the probability of matching. In forensic video recognition, it includes establishing an index of the image in the histogram feature set and retrieving the neighboring values ​​of the forensic video features.

[0045] The steps to index the video feature set are:

[0046] The video high-dimensional feature vector p = (x1, x2, ..., x d ) is mapped into Euclidean space and converted into a binary sequence p' = Y c (x1)Y c (x2)…Y c (x d ). Among them, Y c (x i )(i∈[1,d]) represents the i 1 and cx i0, c is any element x in the feature vector p i The maximum value of , d is the vector dimension.

[0047] Randomly select k bits (k∈(0, c×d)) from the above binary string p' to form l functions: g1(p), g2(p), ..., g l (p), each function corresponds to a hash table.

[0048] Using the above hash function, the feature vector is mapped to the corresponding hash table.

[0049] The steps of using the proximity detection method to detect the forensic video features to be detected are:

[0050] For features e1, e2, ..., e in the forensic video to be detected k , the feature vector is also mapped to the Euclidean space according to the above indexing process, converted into a binary string, and then passed through the same l hash functions g1(p), g2(p), ..., g l (p) are mapped to the corresponding hash tables respectively.

[0051] Extraction i (e j )(i∈(0, l], j∈(0, k]) all hash table entries in the similarity domain, retain the entries whose distance to the feature vector of the forensic video to be detected is within the threshold range, and search the corresponding features by the feature set index as candidate neighboring features.

[0052] The obtained candidate neighboring feature set is sorted in ascending order according to the Euclidean distance between it and the query feature, and the first K features are returned as the neighboring features of the forensic video feature to be detected, which serves as the basis for subsequently sorting the images in the detection histogram feature set to obtain the video detection result.

[0053] The present invention converts the distance problem in the original feature space into the distance measurement problem in the Euclidean space, converts the indexing and query operations of large-scale data into a set of hash function operations, greatly shortens the video detection time, and can still maintain good performance in high-dimensional data space.

[0054] In the feature matching stage, if the first two feature points in the video of the histogram feature set whose Euclidean distance to a feature point in the forensic video to be detected satisfies that the ratio of the second closest distance to the closest distance is greater than a predetermined threshold, the two feature points are regarded as a pair of matching points.

[0055] Determine the fine-grained threshold and the coarse-grained threshold for matching videos in an iterative variable-step manner, and establish constraints based on the accurate matching results of the fine-grained threshold to delete the mismatches in the rough matching results of the coarse-grained threshold. During the iteration, set the number of fine-grained threshold matches e B , so as to ensure the accuracy of the matching and the establishment of its constraints. The threshold determination process is as follows:

[0056] (1) Parameter initialization. Include the threshold The loop times i1 = 0, i2 = 0; the maximum loop times i 1max , i 2max , and the required minimum number of matching points Q.

[0057] (2) Extract feature vectors from the matching images.

[0058] (3) Use the threshold to perform feature matching to obtain the current matching point set A = {(a i , a’ i )}, i = 1, 2,..., e A , where e A is the number of matches corresponding to the current , and let i1 ← i1 + 1.

[0059] (4) If the current number of matches e A < Q, and i1 < i 1max , then let μ1 be the preset first step size and return to step (3); if the number of matches is greater than Q, or i1 > i 1max , then go to step (5).

[0060] (5) Use the threshold to perform feature matching to obtain the current matching point set B = {(b j , b’ j )}, j = 1, 2,..., e B , where e B is the number of matches corresponding to the current , and let i2 ← i2 + 1.

[0061] (6) If the current number of matches e B < 2, and i2 < i2max, then let μ2 be the preset second step size and return to step (5); if the number of matches is greater than 2 and less than 5, and i2 < i 2max , then let μ3 be the preset third step size and return to step (5); otherwise, go to step (7).

[0062] (7) Complete the threshold selection to obtain the coarse-grained threshold and its corresponding coarse matching point set A, and fine-grained threshold and its corresponding exact matching point set B.

[0063] In one embodiment, the texture feature, histogram feature, and wavelet feature of the video summary are selected as feature metrics, which can be described as: watercolor = {texture; color; wavelet}. The idea is to calculate the similarity of video content features, which is expressed by the following formula:

[0064] det(video1, video2)=det(waterprint1, waterprint2).

[0065] The present invention uses the content of these three types of features to represent the features of the video summary. Each type of summary data feature set contains the features of this type of video file, which is expressed by the following formula:

[0066]

[0067]

[0068]

[0069] n represents the number of files contained in each video. The content similarity of texture features, histogram features, and wavelet features is calculated. By comparing each feature of the two videos, the summary feature similarity calculation formula can be derived as follows, indicating that the video summary feature similarity is equivalent to the similarity of all types of two video summaries:

[0070]

[0071]

[0072]

[0073] m represents the number of files contained in each video.

[0074] Furthermore, weights are assigned to the similarities of the above texture features, histogram features, and wavelet features, and the weighted similarities of the three summary data features are used to represent the video summary similarity. The weighted similarity formula is as follows:

[0075] det(video1, video2)=det(waterprint1, waterprint2)=

[0076] det_texture×α+det_color×β+det_wavelet×γ.

[0077] The above formula indicates that the similarity between video1 and video2 is equivalent to the similarity between the internal files of video1 and video2, which is equivalent to the weighted value of the similarity of the three features in the two video summaries. The values ​​of α, β, and γ change dynamically according to the differences of det_texture, det_color, and det_wavelet. After the above process, the similarity of the two video summaries is obtained. By comparing the similarity and the size of the threshold T, it can be determined whether the two videos belong to the same video, that is, whether they have been tampered with.

[0078] In summary, the present invention proposes an artificial intelligence-based video detection method, which embeds authentication data into video files and can verify whether the forensic video has been tampered with without additional information, thereby improving the operating efficiency of the video detection process.

[0079] Obviously, those skilled in the art will understand that the above modules or steps of the present invention can be implemented by a general computing system, they can be concentrated on a single computing system, or distributed on a network composed of multiple computing systems, and optionally, they can be implemented by program codes executable by the computing system, so that they can be stored in a storage system and executed by the computing system. Thus, the present invention is not limited to any specific hardware and video combination.

[0080] It should be understood that the above specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, and do not constitute a limitation of the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made without departing from the spirit and scope of the present invention should be included in the protection scope of the present invention. In addition, the appended claims of the present invention are intended to cover all changes and modifications that fall within the scope and boundaries of the appended claims, or the equivalent forms of such scope and boundaries.

Claims

1. A video detection method based on artificial intelligence, characterized in that: include: receiving the first part of the decryption private key; Determining an image integrity index associated with the video; combining the first portion with the integrity index to generate a decryption private key; Use the generated decryption private key to attempt to decrypt the video image content; If the integrity index indicates that the video image has not been tampered with, and the generated decryption private key is generated by combining the first part and the integrity index, then the content is successfully decrypted; Determining the integrity index associated with the video image further includes: Based on the target security level, determining a first private key matrix; Analyze the video image to obtain a sample detection matrix UM; Each sample detection matrix and its transposed matrix UM T Multiply to generate a second-order symmetric matrix M corresponding to each sample detection matrix; performing at least one arithmetic matrix calculation between each second-order symmetric matrix and the first private key matrix to generate a second private key matrix; and The second private key matrix is ​​used as an integrity index corresponding to the video image.

2. The method according to claim 1, characterized in that Determining an integrity index associated with the video image further includes determining a checksum associated with a portion of the video image.

3. The method according to claim 1, characterized in that The method further comprises: Based on the target security level, an initial parameter set is determined, and the initial parameter set is sent to the smart cloud, so that the smart cloud uses the initial parameter set to determine the integrity of the video image stored in the smart cloud to generate a sample summary.

4. The method according to claim 3, characterized in that The comparing the sample summary with the summary of the video to be detected further includes transmitting the summary of the video to be detected to the intelligent cloud for performing the comparison on the intelligent cloud, or transmitting the sample summary to the client, wherein the client performs the comparison.

Images