False video analysis processing system and method
By designing a false video analysis and processing system, using the collaborative work of big data collection, feature recognition, anomaly detection and deep learning models, the problem of false video recognition is solved, and the detection and blocking of false videos with high accuracy and effectiveness is achieved.
Patent Information
- Application Number
- CN202411573908.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-06
- Publication Date
- 2025-02-18
AI Technical Summary
It is difficult for the existing technology to effectively identify and crack down on false videos spread on the Internet, resulting in false information causing social riots and panic.
A false video analysis and processing system was designed, including a big data collection module, a feature recognition module, anomaly picture detection module, a deep learning big data model and a review module. Through the coordinated work of these modules, they can identify feature points in the video and collect information. Evidence, detect screen abnormalities, conduct iterative updates of deep learning models to ensure accurate judgment of authenticity of videos.
The accuracy of false video recognition can be improved, the spread of false videos can be effectively blocked, social order can be maintained, and the accuracy and adaptability of detection can be further improved through the review process and iterative update of deep learning models.
Smart Images
Figure FT_1 
Figure FT_2 
Figure FT_3
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of information engineering, and in particular to a false video analysis and processing system and method. Background Art
[0002] The essence of video is that a series of static images (called frames) are played continuously at a certain time interval to create a medium with dynamic effects. At the same time, the basic unit of an image is a pixel, which represents the smallest point in an image. An image is composed of thousands of pixels, and each pixel contains color information. Color information is stored in binary form in computing information devices and packaged as data.
[0003] Due to the widespread popularity of AI, creations using AI are becoming more and more numerous, and AI-generated videos are becoming more and more realistic, causing some lawless elements to use AI-generated videos to spread false information, causing riots and spreading panic in society, thereby disrupting social order. Therefore, a false video analysis and processing system and method is needed. Summary of the invention
[0004] In view of the deficiencies in the prior art, the present invention provides a false video analysis and processing system and method, which aims to identify and combat false videos spread on the Internet, block their spread, and maintain social order.
[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A false video analysis and processing system comprises a big data collection module, a feature recognition module, an abnormal picture detection module, a deep learning big data model and a review module. The big data collection module extracts key information through the video name and the feature points in the video identified by the feature recognition module, and collects data from the entire network to judge the authenticity of the video. The abnormal picture detection module performs picture detection by cutting video clips to find false synthesis traces. The deep learning big data model is used to provide samples for picture recognition, and simultaneously conducts learning and iterates model versions. The review module is used for multiple rapid repeated recognition tasks, and cooperates with the deep learning big data model to review and analyze the video.
[0006] Preferably, the big data collection module includes an efficient search engine, an information matching unit and a chain association unit which are electrically connected to each other. The efficient search engine is used to quickly search for information on the Internet as proof material. The search process is performed by the information matching unit matching entries and information features, and the chain association unit is used to retrieve similar entries to speed up the acquisition of information evidence.
[0007] Preferably, the feature recognition module includes a feature identification unit, an anchor positioning unit, an audio conversion unit and a text screening unit which are electrically connected to each other. The feature identification unit is used to identify AI-generated trace features in the video, and use the anchor positioning unit to perform the same type of anchoring to quickly identify subsequent content of the same video as comparison material. The audio conversion unit is used to convert audio information into text and cooperate with the text screening unit to screen the text as a retrieval feature to cooperate with the acquisition of network information evidence.
[0008] Preferably, the abnormal picture detection module includes a video frame extraction unit, a continuity loss recognition unit and a picture distortion recognition unit which are electrically connected to each other. The video frame extraction unit is used to segment the frame pictures of the video, and cooperate with the continuity loss recognition unit and the picture distortion recognition unit to identify abnormal pictures in the video to determine the authenticity of the video.
[0009] Preferably, the review module includes a feature point determination threshold unit and a suspicious point determination unit electrically connected to each other, and the feature point determination threshold unit is used to accumulate the threshold when analyzing and processing each video determined to be authentic. After the threshold reaches a certain value, the suspicious point determination unit will conduct multiple reviews to further confirm the authenticity of the video.
[0010] Preferably, the deep learning big data model includes a sample database, a learning unit and an update iteration unit that are electrically connected to each other, the sample data is used to provide reference samples for the false video analysis and processing work, the learning unit is used to synchronously perform learning and recognition in the false video analysis and processing work, and load new sample data into the sample database, and the update iteration unit is used to organize the data in the sample database after a round of workflow is completed, and establish associations and classifications for similar materials.
[0011] A false video analysis and processing method comprises the following steps: Step 1: Upload characteristic cases of fake videos to the sample database of the deep learning big data model to provide reference resources for subsequent analysis and identification work; Step 2: Upload the video that needs to be analyzed and processed, and the feature recognition module will work with the big data collection module to collect information evidence; Step 3: Using the abnormal images and the information evidence collected in step 2 to determine the authenticity of the video, and generating a determination record report for the video that is determined to be authentic; Step 4: Analyze the video determined to be true in step 3 to generate a determination record report, and review the video determined to be true; Step 5: After the workflow is completed, the data in the sample database is sorted, associations and classifications are established for similar materials, and the system is iteratively upgraded by updating the iteration unit.
[0012] Preferably, the specific steps of collecting information evidence in step 2 are: S1, first identify the AI-like generated trace features in the video through the feature identification unit, and then use the anchor positioning unit to anchor the same type; S2, quickly identify the subsequent content of the same video as comparison material based on the anchor content; S3, converting the audio information into text through an audio conversion unit, and using a text screening unit to screen questions as retrieval features; S4, using the comparison material in S2 and the search features in S3 as a benchmark to conduct a large-scale network search using an efficient search engine; S5. In the search process, the information matching unit matches the terms and information features, and cooperates with the chain association unit to retrieve similar terms to speed up the acquisition of information evidence.
[0013] Preferably, the specific steps of authenticity determination in step three are: S1, cutting the video frame pictures by a video frame extraction unit; S2. Determine the anchored AI-like generated trace features in the video based on the information evidence collected in step 2; S3, determining the continuity of the frame images through a continuity loss identification unit, and detecting whether the image continuity is lost; S4, identifying the distorted and blurred parts of the frame image by the image distortion identification unit; S5. If any one of the detections in S1 to S4 deviates from authenticity, the video is determined to be a false video; otherwise, it is determined to be a true video and a determination record report is generated.
[0014] Preferably, the specific steps of the review in step 4 are: S1, analyzing the determination record report generated in step 3, and accumulating the threshold value through the feature point determination threshold unit; S2, the threshold increases with the number of AI-generated traces and the number of image coherence loss judgments in the judgment record report; S3, when the threshold reaches a certain value, the suspicious point identification unit will conduct multiple reviews to further confirm the authenticity of the video; S4. If the video passes the review, it will be judged as a real video again. If the video fails the review, it will be judged as a fake video, and the record will be entered into the deep learning big data model.
[0015] The present invention provides a false video analysis and processing system and method, which has the following beneficial effects: 1. The present invention selects feature anchor points through the feature recognition module, and collects data efficiently through the big data collection module. It judges the authenticity of the video through the matching degree and detects whether there are traces of AI synthesis and modification, thereby improving the accuracy of false video identification.
[0016] 2. The present invention uses a review process to find suspicious points in the video that has passed the first detection, and repeats the re-detection and analysis work multiple times to prevent the video from being an iteratively upgraded AI-generated version with fewer vulnerabilities than the traditional version, thereby further improving the accuracy of detection and analysis.
[0017] 3. The deep learning big data model in the present invention can learn synchronously during the analysis and processing work, continuously load new sample data, organize the data in the sample database after the workflow is completed, and improve the sample database during updates and iterations to adapt to more different false videos. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 A schematic diagram of a false video analysis and processing system of the present invention; Figure 2 A schematic diagram of a big data collection module system of a false video analysis and processing system of the present invention; Figure 3 A schematic diagram of a feature recognition module system of a false video analysis and processing system of the present invention; Figure 4 A schematic diagram of an abnormal image detection module system of a false video analysis and processing system of the present invention; Figure 5 A schematic diagram of a review module system of a false video analysis and processing system of the present invention; Figure 6 A schematic diagram of a deep learning big data model system of a false video analysis and processing system of the present invention. DETAILED DESCRIPTION
[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0020] Example: Please refer to the attached Figure 1 -Attached Figure 6As one aspect of the present application, an embodiment of the present invention provides a false video analysis and processing system, including a big data collection module, a feature recognition module, an abnormal picture detection module, a deep learning big data model and a review module. The big data collection module extracts key information through the video name and the feature points in the video identified by the feature recognition module, and collects data from the entire network to determine the authenticity of the video. The abnormal picture detection module performs picture detection by cutting video clips to find traces of false synthesis. The deep learning big data model is used to provide samples for picture recognition, and simultaneously conducts learning and iterative model versions. The review module is used for multiple rapid repeated recognition tasks, and cooperates with the deep learning big data model to review and analyze the video.
[0021] The big data collection module is composed of an efficient search engine, an information matching unit and a chain association unit that are electrically connected to each other. The efficient search engine is used to quickly search for information on the Internet as evidence. The search process is carried out by the information matching unit matching entries and information features, and the chain association unit is used to retrieve similar entries to speed up the acquisition of information evidence.
[0022] The feature recognition module is composed of a feature identification unit, an anchor positioning unit, an audio conversion unit and a text screening unit which are electrically connected to each other. The feature identification unit is used to identify AI-generated trace features in the video, and use the anchor positioning unit to perform the same type of anchoring to quickly identify subsequent content of the same video as comparison material. The audio conversion unit is used to convert audio information into text and cooperate with the text screening unit to screen the text as a retrieval feature to cooperate with the acquisition of network information evidence.
[0023] The abnormal picture detection module is composed of a video frame extraction unit, a continuity loss recognition unit and a picture distortion recognition unit which are electrically connected to each other. The video frame extraction unit is used to segment the video frames, and cooperates with the continuity loss recognition unit and the picture distortion recognition unit to identify abnormal pictures in the video to determine the authenticity of the video.
[0024] The review module is composed of a feature point determination threshold unit and a suspicious point discrimination unit which are electrically connected to each other. The feature point determination threshold unit is used to accumulate the threshold when analyzing and processing each video determined to be authentic. After the threshold reaches a certain value, the suspicious point discrimination unit will conduct multiple reviews to further confirm the authenticity of the video.
[0025] The deep learning big data model consists of a sample database, a learning unit and an update iteration unit that are electrically connected to each other. The sample data is used to provide reference samples for the false video analysis and processing work. The learning unit is used to perform learning and recognition synchronously during the false video analysis and processing work, and load new sample data into the sample database. The update iteration unit is used to organize the data in the sample database after a round of workflow is completed, and to establish associations and classifications for similar materials.
[0026] Based on the false video analysis and processing system provided above, as another aspect of the present application, a false video analysis and processing method includes the following steps: Step 1: Upload characteristic cases of fake videos to the sample database of the deep learning big data model to provide reference resources for subsequent analysis and identification work; Step 2: Upload the video that needs to be analyzed and processed, and the feature recognition module will work with the big data collection module to collect information evidence. The specific steps for collecting information evidence are as follows: S1, first identify the AI-like generated trace features in the video through the feature identification unit, and then use the anchor positioning unit to anchor the same type; S2, quickly identify the subsequent content of the same video as comparison material based on the anchor content; S3, converting the audio information into text through an audio conversion unit, and using a text screening unit to screen questions as retrieval features; S4, using the comparison materials in S2 and the search features in S3 as a benchmark to conduct a large-scale network search using an efficient search engine; S5. In the search process, the information matching unit matches the terms and information features, and cooperates with the chain association unit to retrieve similar terms to speed up the acquisition of information evidence.
[0027] Step 3: Use the abnormal images and the information evidence collected in step 2 to determine the authenticity of the video. The specific steps for authenticity determination are: S1, cutting the video frame pictures by a video frame extraction unit; S2, determining the anchored AI-like generated trace features in the video based on the information evidence collected in step 2; S3, determining the continuity of the frame images through a continuity loss identification unit, and detecting whether the image continuity is lost; S4, identifying the distorted and blurred parts of the frame image by the image distortion identification unit; If any of S5, S1 to S4 is judged to be out of authenticity, the video is judged to be a false video; otherwise, it is judged to be a real video and a judgment record report is generated.
[0028] Step 4: Analyze the videos judged to be true in step 3 to generate a judgment record report, and review the videos judged to be true. The specific steps of the review are as follows: S1. Analyze the determination record report generated in step 3 and accumulate the threshold value through the feature point determination threshold unit; S2, the threshold increases with the number of AI-generated traces and the number of image coherence loss judgments in the judgment record report; S3, when the threshold reaches a certain value, the suspicious point identification unit will conduct multiple reviews to further confirm the authenticity of the video; S4. If the video passes the review, it will be judged as a real video again. If the video fails the review, it will be judged as a fake video, and the record will be entered into the deep learning big data model.
[0029] Step 5: After the workflow is completed, the data in the sample database is sorted, associations and classifications are established for similar materials, and the system is iteratively upgraded by updating the iteration unit.
[0030] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A false video analysis and processing system, characterized in that: It includes a big data collection module, a feature recognition module, an abnormal image detection module, a deep learning big data model and a review module. The big data collection module extracts key information through the video name and the feature points in the video identified by the feature recognition module, and collects data from the entire network to determine the authenticity of the video. The abnormal image detection module performs image detection by cutting video clips to find traces of false synthesis. The deep learning big data model is used to provide samples for image recognition, and simultaneously conducts learning and iterates model versions. The review module is used for multiple rapid and repeated recognition tasks, and cooperates with the deep learning big data model to review and analyze the video.
2. A false video analysis and processing system according to claim 1, characterized in that: The big data collection module includes an efficient search engine, an information matching unit and a chain association unit that are electrically connected to each other. The efficient search engine is used to quickly search for information on the Internet as proof data. The search process is performed by the information matching unit to match terms and information features, and the chain association unit is used to retrieve similar terms to speed up the acquisition of information evidence.
3. A false video analysis and processing system according to claim 1, characterized in that: The feature recognition module includes a feature identification unit, an anchor point positioning unit, an audio conversion unit and a text screening unit which are electrically connected to each other. The feature identification unit is used to identify AI-generated trace features in the video, and use the anchor point positioning unit to perform the same type of anchoring to quickly identify subsequent content of the same video as comparison material. The audio conversion unit is used to convert audio information into text and cooperate with the text screening unit to screen the text as a retrieval feature to cooperate with the acquisition of network information evidence.
4. A false video analysis and processing system according to claim 1, characterized in that: The abnormal picture detection module includes a video frame extraction unit, a continuity loss recognition unit and a picture distortion recognition unit which are electrically connected to each other. The video frame extraction unit is used to segment the frame pictures of the video, and cooperates with the continuity loss recognition unit and the picture distortion recognition unit to identify abnormal pictures in the video to determine the authenticity of the video.
5. A false video analysis and processing system according to claim 1, characterized in that: The review module includes a feature point determination threshold unit and a suspicious point discrimination unit electrically connected to each other. The feature point determination threshold unit is used to accumulate the threshold when analyzing and processing each video determined to be authentic. After the threshold reaches a certain value, the suspicious point discrimination unit will conduct multiple reviews to further confirm the authenticity of the video.
6. A false video analysis and processing system according to claim 1, characterized in that: The deep learning big data model includes a sample database, a learning unit and an update iteration unit that are electrically connected to each other. The sample data is used to provide reference samples for the false video analysis and processing work. The learning unit is used to synchronously perform learning and recognition during the false video analysis and processing work, and load new sample data into the sample database. The update iteration unit is used to organize the data in the sample database after a round of workflow is completed, and establish associations and classifications for similar materials.
7. A false video analysis and processing method, using a false video analysis and processing system as claimed in any one of claims 1 to 6, characterized in that: The following steps are involved: Step 1: Upload characteristic cases of fake videos to the sample database of the deep learning big data model to provide reference resources for subsequent analysis and identification work; Step 2: Upload the video that needs to be analyzed and processed, and the feature recognition module will work with the big data collection module to collect information evidence; Step 3: Using the abnormal images and the information evidence collected in step 2 to determine the authenticity of the video, and generating a determination record report for the video that is determined to be authentic; Step 4: Analyze the video determined to be true in step 3 to generate a determination record report, and review the video determined to be true; Step 5: After the workflow is completed, the data in the sample database is sorted, associations and classifications are established for similar materials, and the system is iteratively upgraded by updating the iteration unit.
8. A false video analysis and processing method according to claim 7, characterized in that: The specific steps of collecting information evidence in step 2 are: S1, first identify the AI-like generated trace features in the video through the feature identification unit, and then use the anchor positioning unit to anchor the same type; S2, quickly identify the subsequent content of the same video as comparison material based on the anchor content; S3, converting the audio information into text through an audio conversion unit, and using a text screening unit to screen questions as retrieval features; S4, using the comparison material in S2 and the search features in S3 as a benchmark to conduct a large-scale network search using an efficient search engine; S5. In the search process, the information matching unit matches the terms and information features, and cooperates with the chain association unit to retrieve similar terms to speed up the acquisition of information evidence.
9. A false video analysis and processing method according to claim 8, characterized in that: The specific steps of authenticity determination in step 3 are: S1, cutting the video frame pictures by a video frame extraction unit; S2. Determine the anchored AI-like generated trace features in the video based on the information evidence collected in step 2; S3, determining the continuity of the frame images through a continuity loss identification unit, and detecting whether the image continuity is lost; S4, identifying the distorted and blurred parts of the frame image by the image distortion identification unit; S5. If any one of the detections in S1 to S4 deviates from authenticity, the video is determined to be a false video; otherwise, it is determined to be a true video and a determination record report is generated.
10. A false video analysis and processing method according to claim 9, characterized in that: The specific steps of the review in step 4 are: S1, analyzing the determination record report generated in step 3, and accumulating the threshold value through the feature point determination threshold unit; S2, the threshold increases with the number of AI-generated traces and the number of image coherence loss judgments in the judgment record report; S3, when the threshold reaches a certain value, the suspicious point identification unit will conduct multiple reviews to further confirm the authenticity of the video; S4. If the video passes the review, it will be judged as a real video again. If the video fails the review, it will be judged as a fake video, and the record will be entered into the deep learning big data model.