Sensitive information leakage protection system for video transmission
By disassembling, classifying, extracting features, and assessing sensitivity of video content, and combining this with information leakage prevention strategies, the problem of information leakage during video transmission was solved, thereby improving the security and confidentiality of video transmission.
Patent Information
- Application Number
- CN202410969117.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing video transmission technologies pose a risk of information leakage and lack security and confidentiality.
The system uses a target video attribute information acquisition module to decompose and classify content in a structured manner, a transmission feature content information acquisition module to extract and integrate features, a sensitive content data set determination module to identify sensitive information, a content data desensitization algorithm information determination module to conduct multi-dimensional sensitivity assessment, and an information leakage prevention module to build a protection strategy library for information desensitization and protection.
It achieves precise, comprehensive, and efficient protection of sensitive information in video transmission, reduces the risk of information leakage, and enhances the security and confidentiality of video information transmission.
Smart Images

Figure CN119031182A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of information protection, and in particular to a sensitive information leakage protection system for video transmission. Background Art
[0002] With the rapid development of digital technology and the widespread application of the Internet, the generation, transmission and application of video data are becoming more and more frequent. Video content has become an important carrier in the fields of information dissemination, entertainment, education, and business. However, in the process of processing and transmitting video data, since it may contain a large amount of sensitive information, such as personal identity information, business secrets, financial data, and privacy content, the risk of information leakage also increases.
[0003] The existing technology has the risk of information leakage caused by video transmission and lacks technical problems such as security and confidentiality. Summary of the invention
[0004] The present application provides a sensitive information leakage protection system for video transmission, which is used to solve the technical problems in the prior art that video transmission may cause information leakage risks and lack of security and confidentiality.
[0005] In view of the above problems, the present application provides a sensitive information leakage protection system for video transmission.
[0006] The present application provides a sensitive information leakage protection system for video transmission, the system comprising:
[0007] a target video attribute information acquisition module, the target video attribute information acquisition module is used to acquire target video attribute information, perform content decomposition on the target video attribute information, acquire a video content partition set, and perform structured classification on the video content partition set to acquire a partition image content set, a partition audio content set, and a partition text content set; a transmission feature content information acquisition module, the transmission feature content information acquisition module is used to construct a data processing multi-channel, and based on the data processing multi-channel, feature extraction and integration are performed on the partition image content set, the partition audio content set, and the partition text content set, respectively, to obtain video partition transmission feature content information; a sensitive content data set determination module, the sensitive content data set determination module is used to perform sensitive information extraction and integration on the video content partition set respectively, and the sensitive content data set determination module is used to perform sensitive information extraction and integration on the video content partition set respectively, and the sensitive content data set determination module is used to perform sensitive information extraction and integration on the video content partition set respectively, and the sensitive content data set determination module is used to perform sensitive information extraction and integration on the video content partition set respectively, and the sensitive content data set determination module is used to perform sensitive information extraction and integration on the video content partition set respectively, and the sensitive content data set determination module is used to perform sensitive information extraction and integration on the video content partition set respectively, and the sensitive content data set determination module is used to perform sensitive information extraction and integration on the video content partition set respectively, and the sensitive content data set determination module is used to perform structured classification ... The module comprises a module for analyzing information standards, obtaining a partition sensitive content information discrimination standard, performing sensitive data identification on the video partition transmission feature content information based on the partition sensitive content information discrimination standard, and determining a partition sensitive content data set; a content data desensitization algorithm information determination module, wherein the content data desensitization algorithm information determination module is used to perform multi-dimensional sensitivity evaluation on the partition sensitive content data set, obtain partition content data sensitivity information, and determine partition content data desensitization algorithm information based on the partition content data sensitivity information; an information leakage protection module, wherein the information leakage protection module is used to construct a video sensitive information protection policy library, and based on the partition content data desensitization algorithm information and the video sensitive information protection policy library, performs information desensitization processing and information leakage protection on the video content partition set.
[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:
[0009] The target video attribute information acquisition module acquires the target video attribute information, performs content disassembly, acquires the video content partition set, and performs structured classification on the video content partition set to acquire the partition image content set, partition audio content set, and partition text content set; the transmission feature content information acquisition module constructs a data processing multi-channel, extracts and integrates the features of the partition image content set, partition audio content set, and partition text content set, respectively, to obtain the video partition transmission feature content information; the sensitive content data set determination module performs sensitive information standard analysis, obtains the partition sensitive content information discrimination standard, performs sensitive data identification on the video partition transmission feature content information, and determines the partition sensitive content data set; the content data desensitization algorithm information determination module performs multi-dimensional sensitivity evaluation on the partition sensitive content data set, obtains the partition content data sensitivity information, and determines the partition content data desensitization algorithm information based on the partition content data sensitivity information; the information leakage protection module constructs a video sensitive information protection strategy library, and performs information desensitization processing and information leakage protection on the video content partition set. The technical effect of accurately, comprehensively, and efficiently protecting sensitive information in video transmission, reducing the risk of information leakage, and improving the security and confidentiality of video information transmission has been achieved. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0011] Figure 1 A structural diagram of a sensitive information leakage protection system for video transmission is provided for an embodiment of the present application;
[0012] Figure 2 A schematic diagram of the execution structure of a transmission feature content information acquisition module of a sensitive information leakage protection system for video transmission is provided for an embodiment of the present application.
[0013] Explanation of the reference numerals: target video attribute information acquisition module 10 , transmission feature content information acquisition module 20 , sensitive content data set determination module 30 , content data desensitization algorithm information determination module 40 , information leakage protection module 50 . DETAILED DESCRIPTION
[0014] The present application provides a sensitive information leakage protection system for video transmission, which is used to solve the technical problems in the prior art that video transmission may cause information leakage risks and lack of security and confidentiality. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0015] Embodiment 1, as Figure 1 As shown, the present application provides a sensitive information leakage protection system for video transmission, the system comprising:
[0016] The target video attribute information acquisition module 10 is used to obtain the target video attribute information, perform content decomposition on the target video attribute information, obtain a video content partition set, and perform structured classification on the video content partition set to obtain a partition image content set, a partition audio content set and a partition text content set.
[0017] Specifically, the target video attribute information acquisition module 10 acquires various attribute information of the target video, which includes basic information such as the format, duration, resolution, frame rate, etc. of the video, and also covers relevant descriptions such as the source and purpose of the video. After acquiring the attribute information of the target video, the content is disassembled. This disassembly process is not carried out arbitrarily, but according to a series of preset rules, based on the time sequence, the video is divided according to a fixed time interval (such as every 5 seconds, 10 seconds) to form a preliminary partition, and a comprehensive and orderly video content partition set is obtained. This video content partition set is deeply structured and classified. During the classification, the image recognition algorithm, audio spectrum analysis technology, and natural language processing technology are used to accurately analyze the content of each partition. For partitions that are mainly composed of visual images and contain elements such as objects, characters, and scenes, they will be classified into the partition image content set. If the information in the partition exists mainly in the form of sound, such as character dialogue, background music, environmental sound effects, etc., it will be classified into the partition audio content set. For those contents presented in text form, such as clear subtitles, narration, annotations or tags in the video, they will be accurately divided into partitioned text content sets. It can accurately obtain and classify various types of content in the video, so as to accurately identify sensitive information in it, provide targeted protection for sensitive information in the video transmission process, and greatly reduce the risk of sensitive information leakage.
[0018] The transmission characteristic content information acquisition module 20 is used to construct a data processing multi-channel, and based on the data processing multi-channel, the partition image content set, the partition audio content set and the partition text content set are respectively extracted and integrated to obtain the video partition transmission characteristic content information.
[0019] Specifically, the transmission feature content information acquisition module 20 creates multiple independent and parallel data processing paths, each channel has specific functions and processing capabilities to adapt to different types of partition content. For the partition image content set, image feature extraction technology, such as edge detection, texture analysis, object recognition and other algorithms, is used to extract the color, shape, texture, key objects and other features of the image. For the partition audio content set, audio feature extraction methods, such as frequency analysis, sound intensity detection, speech recognition, etc., are used to obtain key information such as frequency characteristics, duration, intonation, and sound intensity of the audio. For the partition text content set, natural language processing technology, including word frequency statistics, keyword extraction, semantic analysis, etc., is used to extract the vocabulary, sentence structure, theme, emotional tendency and other features of the text. After completing the feature extraction of the partition image, audio and text content sets respectively, these features are integrated, and different types of features are fused, associated or uniformly encoded to form comprehensive feature information that can fully reflect the characteristics of the video partition content, that is, the video partition transmission feature content information. By extracting and integrating the fine features of partitioned images, audio and text content, it is possible to more accurately identify features that may contain sensitive information, provide key basis for subsequent sensitive information judgment and protection, and significantly improve the accuracy of sensitive information identification.
[0020] The sensitive content data set determination module 30 performs sensitive information standard analysis based on the video content partition set respectively, obtains partition sensitive content information identification standards, performs sensitive data identification on the video partition transmission feature content information based on the partition sensitive content information identification standards, and determines the partition sensitive content data set.
[0021] Specifically, the sensitive content data set determination module plays a key role in screening and defining sensitive information. In-depth analysis of sensitive information standards is performed on the video content partition set. This means that it will carefully study the characteristics, uses, and possible information types of each partition, and formulate unique and accurate sensitive content information identification standards for different video content structure areas. Due to the different nature and importance of different partitions, the sensitive information identification standards of each partition are different. For image content, it will detect whether there are specific logos, patterns, barcodes, and other elements that may imply sensitive information; for audio parts, it will identify specific voice tones, keywords, password phrases, etc.; for text, it will look for specific digital combinations, professional terms, confidential codes, etc. After such a comprehensive and in-depth sensitive data identification process, the partition sensitive content data set is finally accurately determined. This set lists in detail the specific data that is identified as sensitive in each video partition. Through such highly precise and targeted operations, the sensitive content data set determination module can accurately and thoroughly identify all sensitive information in complex video content, providing a definite basis for subsequent strict protection measures, thereby effectively ensuring the security of sensitive information during video transmission and eliminating any possible risk of leakage.
[0022] The content data desensitization algorithm information determination module 40 is used to perform a multi-dimensional sensitivity assessment on the partitioned sensitive content data set, obtain partitioned content data sensitivity information, and determine partitioned content data desensitization algorithm information based on the partitioned content data sensitivity information.
[0023] Specifically, when the content data desensitization algorithm information determination module 40 performs a comprehensive and in-depth analysis of the partition sensitive content data set during the multi-dimensional sensitivity assessment, it will conduct the assessment work from multiple key perspectives. For example, from the perspective of data type, if the partition sensitive content contains personal financial account information, such as credit card number, bank account number, etc., since such information is directly related to personal property security, it will be given extremely high sensitivity; while if it is some general personal interests and hobbies data, the sensitivity is relatively low. From the perspective of the spread of data, if the sensitive content data may be widely spread and affect many users, such as a large-scale customer list, its sensitivity will be significantly improved; on the contrary, if it is only used in a small range or has a limited impact on data, the sensitivity will be reduced accordingly. From the perspective of data update frequency, if it is highly timely and frequently updated data, such as real-time stock trading data, the sensitivity will be extremely high within a specific time period; while relatively stable and infrequently changing data, such as historical financial statements, the sensitivity may be relatively stable. Through such a comprehensive and detailed multi-dimensional assessment, the sensitivity information of the partition content data can be accurately obtained. This information can describe the sensitivity level of each partition sensitive data in detail, which is divided into different levels such as high, medium and low, and can be further subdivided into specific degrees. Subsequently, based on the obtained partition content data sensitivity information, the partition content data desensitization algorithm information is determined in a targeted manner. For partition content data that is assessed as highly sensitive, advanced encryption technology is selected, such as using a high-intensity asymmetric encryption algorithm to encrypt the data to ensure that only authorized personnel with specific keys can decrypt and access it; or using data replacement to replace key sensitive information with identifiers that have no practical significance but can be recovered. For medium-sensitivity data, partial obfuscation methods are used, such as hiding some key numbers or characters; or data truncation is performed to retain only some key fragments of the data. For low-sensitivity data, simple anonymization processing is selected to remove obvious identification that can directly identify individuals or organizations. Through this precise and adaptive processing method, the security of sensitive data can be effectively protected, and the availability of data and business needs can be balanced to the greatest extent, ensuring that normal business processes and reasonable use of data are not affected under the premise of ensuring information security.
[0024] The information leakage protection module 50 is used to construct a video sensitive information protection policy library, and based on the partition content data desensitization algorithm information and the video sensitive information protection policy library, perform information desensitization processing and information leakage protection on the video content partition set.
[0025] Specifically, the information leakage protection module 50 constructs a video sensitive information protection policy library, which is a comprehensive and comprehensive resource collection, which contains a series of protection policies, rules and standards for sensitive information of different types and degrees. These policies cover detailed provisions in multiple aspects, such as data access control, encryption technology application, transmission security protocol and user authority management. Based on the previously determined partition content data desensitization algorithm information and the constructed video sensitive information protection policy library, the video content partition set is processed in a fine and comprehensive manner. In terms of information desensitization processing, the sensitive data in the partition will be processed strictly in accordance with the selected desensitization algorithm. For example, if the data of a partition adopts a data fuzzification desensitization algorithm, the corresponding sensitive information will be fuzzy, so that it cannot be easily identified and understood while maintaining a certain degree of usability. In terms of information leakage protection, a variety of means will be used in combination. For example, strict access control is set up, and only authorized personnel can access sensitive information in specific partitions under specific conditions; encrypted transmission technology is used to ensure that data cannot be interpreted even if it is intercepted during data transmission; real-time monitoring and auditing of data access and use are carried out to promptly detect and prevent abnormal access behavior; and an emergency response mechanism is established to quickly take measures to handle and remedy information leakage incidents and minimize losses and impacts. Through such a comprehensive and rigorous processing process, the security of sensitive information in the video content partition set is guaranteed to the greatest extent, effectively preventing the risk of information leakage, and providing reliable security for the transmission and use of videos.
[0026] In a possible implementation, the target video attribute information acquisition module 10 further includes:
[0027] According to the target video attribute information, the video title, video tag and video description are obtained, and content classification and analysis are performed based on the video title, video tag and video description to generate video content directory information; video decoding and video frame extraction are performed on the target video to obtain a set of video segmentation frames; scene recognition is performed based on the set of video segmentation frames to obtain video frame scene feature information, and the video frame scene feature information is matched with the video content directory information to obtain a scene matching structure label set; the video segmentation frame set is clustered and partitioned based on the scene matching structure label set to obtain the video content partition set.
[0028] Specifically, after obtaining the attribute information of the target video, key elements will be extracted from it, namely the video title, video tags and video description. For the video title, its wording, theme expression and possible key information will be analyzed. Video tags are usually a brief summary of the video content or an annotation of key features, and each tag will be interpreted and considered separately. Video descriptions often contain richer and more detailed content, and natural language processing is performed on them, including lexical analysis, syntactic analysis, semantic understanding, etc. Combining the information provided by these three, the text classification model convolutional neural network is used to automatically extract features from the input text. For video-related text, it can capture the local relationship and pattern between words, extract representative features through multi-layer convolution and pooling operations, and then classify the main content of the video through the fully connected layer. After such an in-depth and comprehensive analysis and processing process, the final generated video content directory information will be presented in a highly structured and clear and easy-to-understand manner. The catalog will first identify the main categories, and then list the specific branches in detail in the subcategories. For each category, there will not only be a concise and accurate name description, but also a detailed text description that summarizes the main content features, key elements and unique value covered by the category.
[0029] The first step in processing the target video is video decoding. Videos are usually stored in a specific encoding format, such as H.264, H.265, etc. Video decoding is to convert these encoded and compressed data into a processable raw image data format. After decoding, the next step is video frame extraction, which means selecting key frames from a continuous video stream according to a certain time interval or specific rules. This time interval can be fixed, such as extracting a frame every 0.1 seconds; or it can be dynamically determined according to the degree of change in the video content, such as extracting frames more densely in the part where the picture changes dramatically, and extracting them at a longer interval in the relatively stable part. Through such decoding and extraction operations, a set of many video frames is finally obtained, that is, a video segmentation frame set.
[0030] The acquired video segmentation frame set is analyzed in depth for scene recognition. This process uses computer vision technology and image processing algorithms. For each video frame, a series of information that can characterize its scene features is extracted. This includes elements such as objects, people, background, color distribution, texture features, and spatial layout in the image. For example, specific objects such as cars and buildings in the frame are identified by image detection algorithms; the color distribution is analyzed by color histograms; and the texture characteristics of the image are extracted using texture analysis methods. After obtaining the scene feature information of the video frame, it is matched with the previously generated video content directory information. This matching process is a comparison and association operation that aims to find the consistency between the scene features of the video frame and the categories and features described in the video content directory. For example, if the video content directory mentions a scene of a city street, and the scene features of a video frame show dense buildings, traffic lights, and pedestrians, then the two can be considered to be matched. After matching all video frames, a set of scene matching structure labels is finally obtained. Each label in this set represents a specific scene structure type to which the matching video frame belongs.
[0031] After obtaining the scene matching structure label set, the video segmentation frame set is subjected to in-depth clustering and partitioning operations. The purpose of clustering and partitioning is to more accurately identify and divide different scene areas in the video so as to better manage and protect possible sensitive information. For example, in an internal training video of an enterprise, the scene of discussing sensitive projects in the conference room and the scene of employees' leisure communication need to be clearly distinguished. For each scene label, there is a specific sensitivity level and protection requirement. For example, the scene of discussing sensitive projects in the conference room contains highly sensitive content such as confidential business plans and undisclosed product information, while the scene of employees' leisure communication usually does not involve sensitive information. In the clustering process, the distance-based clustering algorithm will comprehensively consider multiple factors to measure the similarity between frames. From the perspective of visual features, similar layouts, personnel distribution, displayed documents or objects, etc. will be considered similar; from the perspective of content elements, the presence or absence of specific logos, symbols, charts, etc. will also affect the distance calculation. Temporal continuity is particularly important in the protection of sensitive information in video transmission. If continuous frames belong to the same sensitive scene category, a relatively complete sensitive information area can be more accurately determined to prevent sensitive information from being split or omitted during the partitioning process. Through clustering and partitioning operations, the final set of video content partitions can clearly define areas of different sensitivity levels. This allows stronger encryption, access control and other protection measures to be taken for highly sensitive areas during video transmission, while a relatively loose but still safe processing method is adopted for low-sensitivity areas. This ensures the security of sensitive information while optimizing transmission efficiency and resource utilization.
[0032] In a possible implementation, the target video attribute information acquisition module 10 further includes:
[0033] According to the video content directory information, a video scene entity label library is constructed, and the video scene entity label library is a type label library for each video scene entity; entity combination classification is performed based on the video scene entity label library to obtain a video application scene category set; video frame recognition training is performed based on the video application scene category set to construct a video scene classifier, and the video scene classifier includes a network encoder and a network decoder; the video segmentation frame set is sequentially input into the network encoder to obtain an image frame multi-scale feature set; based on the network decoder, the image frame multi-scale feature set is sequentially identified and classified, and the video frame scene feature information is output.
[0034] Specifically, based on the obtained video content directory information, key scene entity descriptions and features are extracted from it. For example, descriptions of scenes such as office meetings, campus sports games, and family dinners are found in the directory. Then, these extracted scene entities are classified and labeled to construct a special video scene entity label library. Each label in this label library clearly represents a specific type of video scene entity. For example, office meetings are labeled as work scenes-meetings, campus sports games are labeled as educational scenes-sports activities, and family dinners are labeled as life scenes-family gatherings, etc. This label library covers a wide variety of video scene types, and each label has a clear definition and classification, which can provide an accurate and comprehensive scene identification basis for subsequent video analysis, processing, and classification. In a sensitive information leakage protection system for video transmission, such a label library helps to quickly and accurately identify scene entities that may contain sensitive information in the video, so as to take targeted protection measures and reduce the risk of sensitive information leakage.
[0035] Based on the constructed video scene entity tag library, various entity tags in it are combined and analyzed. For example, the tag library may have separate entity tags such as production equipment, production products, and worker operations. When the two entity tags of production equipment and production products are combined together, a type of scene such as a production line can be constructed. For example, entity tags such as classrooms, students, and blackboards are combined to form a teaching type scene; hospital wards, medical equipment, patients, etc. are combined to form a medical care type scene. By combining and associating many possible entity tags in different ways, and comprehensively considering factors such as the relationship between entities, the frequency of occurrence, and common characteristics, a collection covering various scene categories is finally obtained. Each category in this collection represents a video application scenario composed of specific entities, such as industrial production scenarios, educational teaching scenarios, medical service scenarios, entertainment performance scenarios, etc. In the sensitive information leakage protection system used for video transmission, such a clear video application scenario category collection can help the system more accurately judge the types of sensitive information and risk levels that may exist in different scenarios, so as to adopt corresponding and targeted protection strategies and effectively ensure information security during video transmission.
[0036] Based on the set of video application scene categories obtained above, the video frame recognition training is carried out. During the training process, a large amount of video frame data with clear scene category annotations will be used. These data are input into the model to let the model learn the features and patterns of video frames under different scene categories. Through such training, a video scene classifier is constructed. This classifier consists of two parts: a network encoder and a network decoder. The function of the network encoder is to extract and encode the input video frames. It can convert the video frames into a compressed and representative feature representation, capturing the key information in the video frames, such as the shape, color, texture of the object, and the layout and structure of the scene. The network decoder receives the feature representation output by the encoder and is responsible for interpreting and classifying these features. It determines which video application scene category the input features belong to based on the knowledge and patterns learned in the previous training. In the sensitive information leakage protection system for video transmission, such a video scene classifier can quickly and accurately classify the scenes of the transmitted video frames, so as to timely identify the scenes where sensitive information may exist, provide an important basis for taking corresponding protection measures, and effectively ensure the security of sensitive information.
[0037] Each frame in the previously obtained video segmentation frame set is input into the constructed network encoder in sequence. The network encoder will conduct in-depth analysis and processing of each input frame. It does not extract features from a single scale or angle, but can extract features from multiple scales at the same time. For example, at a small scale, it will focus on the local details of the image, such as the texture and subtle changes in the edges of the image; at a large scale, it will capture the overall structure of the image, the shape and distribution of the main objects, etc. Through this multi-scale feature extraction method, the network encoder can comprehensively and richly describe the features of each frame of the image, thereby obtaining a multi-scale feature set of the image frame. This feature set contains the key information of each frame of the image at different scales, providing a comprehensive and detailed foundation for the subsequent network decoder to accurately classify the scene. In the sensitive information leakage protection system used for video transmission, this multi-scale feature set can more accurately reflect the information in the video frame, which helps to more effectively identify and protect possible sensitive content.
[0038] After receiving the multi-scale feature set of the image frame generated by the network encoder, the network decoder begins to process and analyze it one by one. The decoder contains various scene patterns and classification rules learned through training. It compares and matches the multi-scale features of each input image frame with these learned patterns. In the process of recognition and classification, the decoder comprehensively considers the performance and combination of features at different scales. For example, the local detail features of an image frame may suggest that it belongs to a certain scene, and the overall structural features further confirm or correct this judgment. Finally, the scene category to which each image frame belongs can be determined, and the corresponding video frame scene feature information can be output. These output video frame scene feature information describes in detail the specific scene characteristics represented by each frame of the image, such as indoor office scenes, outdoor sports scenes, or other specific types of scenes. In the sensitive information leakage protection system used for video transmission, these precise scene feature information is crucial for accurately identifying scenes that may contain sensitive information, so that corresponding protection measures can be taken in time to ensure information security during video transmission.
[0039] In one possible implementation, Figure 2 As shown, the transmission feature content information acquisition module 20 also includes:
[0040] The structured type of video content is obtained, and the structured type of video content includes image attribute data, audio attribute data and text attribute data; preprocessing steps are configured based on the image attribute data, audio attribute data and text attribute data respectively, and a set of video attribute data preprocessing steps is obtained; the image attribute data, audio attribute data and text attribute data are analyzed in turn by a feature extraction algorithm to obtain a set of video attribute data feature extraction steps; based on the set of video attribute data preprocessing steps and the set of video attribute data feature extraction steps, processing channels are loaded for the structured type of video content respectively to construct the data processing multi-channel.
[0041] Specifically, when analyzing and processing video content, we first need to clarify the structured type of the content contained in the video and parse and decompose the video. Among them, the structured types of video content include image attribute data, audio attribute data, and text attribute data. Image attribute data refers to the visual picture information presented in the video, including the shape, color, texture, appearance and movement of objects, and the layout of the scene. Audio attribute data covers the sound elements in the video, such as character dialogue, background music, and environmental sound effects. Text attribute data includes text information that appears in the video, such as subtitles, titles, annotations, and video descriptions. Comprehensively acquiring and determining these three types of data, namely image attribute data, audio attribute data, and text attribute data, lays the foundation for more in-depth and targeted processing and analysis in the future.
[0042] After obtaining the image attribute data, audio attribute data, and text attribute data, the preprocessing steps are planned and set for each type of data. For image attribute data, the preprocessing step configuration includes filtering and denoising, which is to remove noise interference in the image to make the image clearer and smoother, and image enhancement, which highlights the important features in the image by adjusting parameters such as contrast and brightness. In addition, operations such as image scaling and cropping are performed to meet the needs of subsequent processing. For audio attribute data, the preprocessing steps can include audio noise reduction to reduce the impact of background noise on effective sound, audio equalization to adjust the intensity of different frequency components to make the sound clearer and more balanced. It also includes audio segmentation to divide the continuous audio stream into meaningful segments for subsequent analysis. For text attribute data, the preprocessing steps include text cleaning, removing special characters, extra spaces and punctuation errors, etc., and word segmentation to divide the text into words or phrases. Stemming or word form restoration is performed to convert words into their basic form for subsequent analysis and comparison. By configuring such preprocessing steps for these three different types of data respectively, we finally get a set of preprocessing steps for image attribute data, audio attribute data and text attribute data. This set clarifies a series of operations that need to be performed in sequence when processing video attribute data, preparing for subsequent more accurate and effective data analysis and processing.
[0043] After obtaining the image attribute data, audio attribute data and text attribute data, the feature extraction algorithm is analyzed in turn. For image attribute data, the image feature extraction algorithm is used. The shape-based feature extraction obtains features by detecting and describing the outline and shape of objects in the image; the texture-based feature extraction analyzes the distribution law and repetitive pattern of pixels in the image; the color-based feature extraction focuses on the distribution and combination of colors in the image, etc. The application and analysis of these algorithms form the feature extraction steps for image attribute data. For audio attribute data, the feature extraction algorithms include spectrum-based feature extraction, which analyzes the energy distribution of audio signals at different frequencies; rhythm-based feature extraction, which focuses on the rhythm, intonation, stress and other features of audio; acoustic model-based feature extraction, which uses acoustic principles to obtain audio feature parameters, etc. The analysis of these algorithms determines the feature extraction steps for audio attribute data. For text attribute data, common feature extraction algorithms include the bag-of-words model, which represents text as a collection of words and the frequency of word occurrence; the TF-IDF (term frequency-inverse document frequency) algorithm, which measures the importance of words in the text; and topic models, such as LDA (latent Dirichlet allocation), which discovers potential topics in the text. By analyzing these algorithms, we get the feature extraction steps for text attribute data. Combining the above analysis of the feature extraction algorithms for image, audio and text attribute data, we finally form a set of feature extraction steps for these three different types of data.
[0044] After having a set of video attribute data preprocessing steps and a set of video attribute data feature extraction steps, build processing channels respectively. For the processing channel of image attribute data, first load the preprocessing steps, such as filtering and denoising operations, and then perform the corresponding feature extraction steps, such as shape or texture-based feature extraction. This channel is specifically used to process and analyze image information in the video to ensure that the image data can provide valuable input for subsequent analysis and application after optimization and feature extraction. For the processing channel of audio attribute data, also first load the preprocessing steps, such as audio noise reduction and equalization, and then perform feature extraction steps, such as feature extraction based on spectrum or rhythm. This channel focuses on processing audio information and extracting key features that can reflect the characteristics and content of the audio. For the processing channel of text attribute data, first perform preprocessing operations such as text cleaning and word segmentation, and then perform feature extraction steps such as bag of words model and TF-IDF. This channel is committed to effectively processing and extracting features from text content in the video. By loading the corresponding preprocessing and feature extraction steps for the three different structured types of video content, namely images, audio and text, we build multiple independent and collaborative processing channels, forming a data processing multi-channel architecture that can efficiently process multiple types of data at the same time.
[0045] In a possible implementation, the content data desensitization algorithm information determination module 40 further includes:
[0046] Acquire sensitive content description factor information, wherein the sensitive content description factor information includes content source partition, sensitive information type, sensitive information quantity and duration; perform feature descriptions on the partitioned sensitive content data sets based on the sensitive content description factor information to obtain partitioned sensitive data description feature sets; perform criticality assessment on each descriptive factor in the sensitive content description factor information to obtain sensitive description factor criticality factors; perform weighted quantitative assessment on the partitioned sensitive data description feature sets based on the sensitive description factor criticality factors to obtain the partitioned content data sensitivity information.
[0047] Specifically, when processing data involving sensitive content, it is necessary to obtain a series of key factor information used to describe these sensitive contents. Content source partitioning refers to clarifying the specific location of sensitive content in the entire data system or the specific area to which it belongs, which helps to determine the source of sensitive content and the possible scope of dissemination. Sensitive information type is to classify and define the nature of sensitive content, such as personal privacy information such as personal ID number, bank card password, medical records, or confidential information such as corporate trade secrets, R&D data, financial statements, etc. The number of sensitive information indicates the size of the sensitive data involved. This may be a specific number, such as how many personal information records are included, or a relative measure, such as a large amount, a small amount, etc. Duration refers to the length of time that sensitive content exists, remains valid, or is processed in a specific scenario or process. It can reflect the timeliness of sensitive content and the potential risk exposure cycle. Combining these sensitive content description factor information, it is possible to fully and accurately understand the characteristics and status of sensitive content, thereby providing the necessary basis and basis for subsequent evaluation, protection and management.
[0048] For the sensitive content data of each partition, firstly, based on the factor of content source partition, the specific location of the partition in the video stream is clarified, such as whether it is the beginning of the video, a certain section in the middle, or the end. At the same time, its distribution area on the screen can also be determined, whether it is located in the upper left corner of the screen, the center, or other specific positions. Then, according to the sensitive information type factor, the specific category of the sensitive content of the partition is described in detail. In video transmission, this includes the image of the personal ID number that appears in the video, the company's confidential document screen, and confidential products with specific logos. Then, according to the sensitive information number factor, the number of sensitive information in the partition is counted. For example, there are 5 clearly visible facial features in a partition, or a confidential document display screen for 10 seconds. Finally, according to the duration factor, the duration of the continuous appearance of the sensitive content of the partition in the video is recorded. For example, the sensitive screen of a partition lasts for 8 seconds. By making such a comprehensive and detailed description of the sensitive content data of each partition based on the above sensitive content description factor information, a partition sensitive data description feature set is finally formed. This collection provides the protection system with precise and detailed information, enabling the system to accurately determine the risk level of sensitive content in each partition, and thus take targeted protection measures such as real-time encryption, screen masking, access permission control, etc. to ensure the security of sensitive information during video transmission.
[0049] It is crucial to evaluate the criticality of each descriptor in the sensitive content descriptor information. For content source partitions, the importance of the partition in the entire video is considered when evaluating its criticality. For example, if a partition is the core part of the video, such as a scene containing key decision discussions, then its criticality is high; while if it is a marginal, insignificant screen partition, its criticality is relatively low. The criticality assessment of sensitive information types depends on the sensitivity level of the information. For example, the type containing personal bank account information is considered to have extremely high criticality, while the type of information such as general personal hobbies is less critical. In terms of the amount of sensitive information, a large amount of sensitive information is obviously more critical than a small amount. If a large amount of confidential data appears in a partition, its criticality will increase accordingly. The criticality assessment of duration is based on the duration of sensitive content exposure. Sensitive information that is continuously displayed for a long time may bring a greater risk of leakage and has a higher criticality. Through such an assessment, each descriptor is assigned a corresponding weight or score, thereby obtaining the criticality factor of sensitive descriptors. These factors will play an important role in subsequent protection decisions, determining the intensity and priority of protection measures for sensitive content in different partitions.
[0050] In the sensitive information protection of video transmission, based on the previously determined key factors of sensitive description factors, a weighted quantitative evaluation is performed on the partition sensitive data description feature set. According to the key factors of sensitive description factors, a corresponding weight value is assigned to each description factor. For example, for the content source partition, if the partition weight of the company's core conference room is set to 0.3; in the sensitive information class, the weight of financial data is set to 0.4; when the amount of sensitive information is large, the weight is set to 0.2; the weight of the duration is long is set to 0.1. Then, the sensitive data description feature set of each partition is evaluated. Take a partition as an example, assuming that it is in the core conference room (the score is calculated according to the weight as 0.3×100=30), contains important financial data (0.4×100=40), has a large amount of sensitive information (0.2×80=16), and has a long duration (0.1×90=9). Add up the scores of each item: 30+40+16+9=95. Through such weighted calculation, the sensitivity evaluation value of the partition is 95. Similar calculations and evaluations are performed on all partitions to obtain the sensitivity information of the content data of each partition. This sensitivity information can clearly reflect the sensitivity of different partitions. During video transmission, the protection system can take stricter encryption, access control or real-time monitoring measures for highly sensitive partitions based on this sensitivity information, while using more conventional protection methods for relatively low-sensitivity partitions to achieve reasonable resource allocation and optimal protection effects, and minimize the risk of sensitive information leakage.
[0051] In a possible implementation, the information leakage protection module 50 further includes:
[0052] Construct a data desensitization algorithm list, wherein the data desensitization algorithm list includes data desensitization algorithms with different desensitization levels; obtain a video transmission security standard, perform a desensitization rating on the partition content data sensitivity information based on the video transmission security standard, and determine the partition content data desensitization level; and determine the partition content data desensitization algorithm information based on a matching mapping between the partition content data desensitization level and the data desensitization algorithm list.
[0053] Specifically, the data desensitization algorithm list is constructed to meet the needs of effectively protecting data of different sensitivity levels in video transmission. The data desensitization algorithms in this list are divided according to the degree of desensitization. The algorithms of the mild desensitization level include: partial data hiding, such as hiding only the middle digits of a mobile phone number. Data replacement, replacing part of the sensitive information with specific symbols or randomly generated characters. The algorithms of the medium desensitization level include: data truncation, retaining the front or back part of the sensitive data and truncating the middle content. Data randomization, making a certain degree of random modification to the sensitive data, but still maintaining the format and type of the data. The algorithms of the severe desensitization level include: data encryption, using a high-intensity encryption algorithm to encrypt sensitive data, and only with the correct key can it be decrypted and restored. Data anonymization, completely removing information that can directly identify individuals or sensitive subjects, so that the data cannot be traced back to the original source. By constructing such a list of data desensitization algorithms with different desensitization levels, it is possible to flexibly select and apply appropriate desensitization algorithms according to specific video transmission scenarios and data sensitivity requirements to ensure the security and compliance of sensitive information during transmission.
[0054] Obtain security standards specifically for video transmission, which are usually determined by relevant industry specifications, laws and regulations, and internal organizational policies. Then, based on these video transmission security standards, carefully analyze and evaluate the sensitivity information of the partition content data that has been obtained. During the evaluation process, multiple factors will be considered, such as the type of sensitive information (personal identity information, trade secrets, or other sensitive categories), the amount of sensitive information, the importance of the data, and potential risks. According to the specific rules and thresholds set in the security standards, if the sensitive data of a partition is mainly general personal information and the amount is small, it may be rated as a lower desensitization level, such as level 1. If the partition contains a large amount of critical trade secrets or highly sensitive personal financial information, it may be classified as a higher desensitization level, such as level 5. Through such strict evaluation and classification based on video transmission security standards, the required desensitization level of each partition content data can be accurately determined, providing clear guidance for the subsequent appropriate desensitization measures, thereby ensuring the security and compliance of sensitive data during video transmission.
[0055] After the desensitization level of the partition content data is determined, it needs to be matched and mapped with the previously constructed data desensitization algorithm list to clarify the specific desensitization algorithm that should be used for each partition. For example, if a partition is rated as the lowest desensitization level, such as level one, then according to the matching mapping, it will correspond to the mild desensitization algorithm in the list, such as simple data replacement or partial data hiding. If the partition is determined to be a higher desensitization level, such as level four, it will be matched to a more powerful desensitization algorithm, such as data encryption or data anonymization. When matching, the desensitization level will be matched with the desensitization strength and scope of application of the algorithm in the data desensitization algorithm list. Through this precise matching mapping, the most suitable desensitization algorithm information can be accurately determined for each partition content data. This ensures that while protecting sensitive data, the availability and security of data are balanced to the greatest extent to meet various security needs and business requirements in video transmission.
[0056] In a possible implementation, the information leakage protection module 50 further includes:
[0057] Based on the partition content data desensitization algorithm information, the partition sensitive content data set in the video content partition set is desensitized to obtain a partition video content desensitized data set; based on the partition content data sensitivity information of the video content partition set, a matching analysis is performed with the video sensitive information protection strategy library to obtain a partition content sensitive protection strategy set; based on the partition content sensitive protection strategy set, the partition video content desensitized data set is protected from information leakage.
[0058] Specifically, in the sensitive information protection process of video transmission, we have clarified the desensitization algorithm information of each partition content data. Next, we will carry out actual processing operations on the partition sensitive content data set in the video content partition set based on this information. By strictly following the predetermined partition content data desensitization algorithm, the sensitive content data of each partition is processed one by one, and the final result is the partition video content desensitized data set after information desensitization, which reduces the risk of sensitive information leakage and provides a safer guarantee for subsequent transmission and use.
[0059] During the video transmission process, the content data sensitivity information of each partition in the video content partition set has been determined. At this time, these sensitivity information are compared and analyzed with the pre-established video sensitive information protection policy library. This protection policy library stores a large number of response strategy solutions for different sensitivity levels. For example, if the sensitivity of a partition is evaluated as extremely high, it is matched with an extremely strict protection policy in the protection policy library, including the use of multiple encryption technologies, limiting access rights to only specific high-level personnel, and real-time monitoring of data access activities. For partitions with lower sensitivity, the matching protection policy may be relatively loose, such as only basic access authentication and regular data backup. Through such precise matching and analysis, the most suitable protection policy for each partition is obtained for its sensitivity status, and finally a partition content sensitive protection policy set is formed. Each policy in this set is accurately matched with the sensitivity of the corresponding partition, which can reasonably allocate protection resources and improve protection efficiency on the premise of ensuring the security of sensitive information, and minimize the risks faced by sensitive information during video transmission and storage.
[0060] Based on the partition content sensitive protection strategy set, further information leakage protection work is carried out on the partition video content desensitized data set that has completed the desensitization processing, including setting access permissions, monitoring data transmission paths, and performing data backup and recovery planning. A series of measures are taken to minimize the risk of sensitive information leakage during video transmission and ensure the security and confidentiality of video content.
[0061] In a possible implementation, the information leakage protection module 50 further includes:
[0062] Based on the matching analysis between the partition content data sensitivity information and the video sensitive information protection policy library, a partition matching protection policy set is obtained; a multi-dimensional protection test is performed on the partition matching protection policy set to obtain a policy security protection effect set, and the partition content sensitive protection policy set is preferentially obtained based on the policy security protection effect set.
[0063] Specifically, according to the determined partition content data sensitivity information, it is matched with the video sensitive information protection policy library. In this process, the system will filter out a series of applicable protection policies from the protection policy library based on the sensitivity level, type of sensitive information and other characteristics, thereby obtaining a partition matching protection policy set.
[0064] After obtaining the partition matching protection strategy set, in order to ensure that the selected strategy can truly and effectively protect the sensitive content in the video partition, it is necessary to conduct multi-dimensional protection tests on these strategies. Multi-dimensional protection tests will evaluate the effectiveness of the strategy from multiple different angles and aspects. For example, simulated attack tests are conducted, using various common attack methods and techniques, such as hacker attacks, malware intrusions, etc., to verify whether the strategy can successfully resist these attacks and protect sensitive data from being stolen or tampered with. Performance tests will also be conducted to observe whether the strategy has a negative impact on the speed, smoothness and quality of video transmission during implementation. If the strategy causes video freezes and excessive delays, it needs to be reconsidered or optimized. In addition, compatibility tests will also be conducted to ensure that the strategy is well compatible with the existing video system, hardware equipment, software applications and network environment, and there will be no conflicts or failures. Through these multi-dimensional tests, performance data of each strategy in different aspects can be obtained, thereby forming a strategy security protection effect set, which includes specific effect evaluations of each strategy in terms of attack resistance, performance impact, compatibility, etc. Then, based on this set of policy security protection effects, we conduct a comprehensive comparison and analysis of each policy, and give priority to those policies that perform well in multiple important dimensions, such as those that can effectively resist attacks, have little impact on performance, and have good compatibility. After this optimization process, we finally determine the most suitable content-sensitive protection policy set for each partition. The policies in this set have been rigorously tested and screened, and can maximize the security of sensitive content in video partitions, while achieving good results and stability in actual applications.
[0065] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0066] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.
[0067] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.
Claims
1. A sensitive information leakage protection system for video transmission, characterized in that: The system comprises: A target video attribute information acquisition module, the target video attribute information acquisition module is used to acquire target video attribute information, perform content decomposition on the target video attribute information, acquire a video content partition set, and perform structured classification on the video content partition set to acquire a partition image content set, a partition audio content set, and a partition text content set; A transmission feature content information acquisition module, the transmission feature content information acquisition module is used to construct a data processing multi-channel, based on which the sub-area image content set, the sub-area audio content set and the sub-area text content set are respectively subjected to feature extraction and integration to obtain video sub-area transmission feature content information; A sensitive content data set determination module, wherein the sensitive content data set determination module performs sensitive information standard analysis based on the video content partition set, obtains a partition sensitive content information discrimination standard, performs sensitive data identification on the video partition transmission feature content information based on the partition sensitive content information discrimination standard, and determines a partition sensitive content data set; a content data desensitization algorithm information determination module, the content data desensitization algorithm information determination module being used to perform a multi-dimensional sensitivity assessment on the partition sensitive content data set, obtain partition content data sensitivity information, and determine partition content data desensitization algorithm information based on the partition content data sensitivity information; An information leakage protection module is used to construct a video sensitive information protection policy library, and based on the partition content data desensitization algorithm information and the video sensitive information protection policy library, performs information desensitization processing and information leakage protection on the video content partition set.
2. The sensitive information leakage protection system for video transmission according to claim 1, characterized in that: The step of obtaining a set of video content partitions includes: According to the target video attribute information, a video title, a video tag and a video description are obtained, and content classification and analysis are performed based on the video title, the video tag and the video description to generate video content directory information; Perform video decoding and video frame extraction on the target video to obtain a set of video segmentation frames; Performing scene recognition based on the video segmentation frame set to obtain video frame scene feature information, and matching the video frame scene feature information with the video content directory information to obtain a scene matching structure label set; The video segmentation frame set is clustered and partitioned based on the scene matching structure label set to obtain the video content partition set.
3. The sensitive information leakage protection system for video transmission according to claim 2, characterized in that: The obtaining of video frame scene feature information includes: According to the video content directory information, a video scene entity tag library is constructed, wherein the video scene entity tag library is a type tag library of each video scene entity; Perform entity combination classification based on the video scene entity tag library to obtain a video application scene category set; Performing video frame recognition training based on the video application scene category set to construct a video scene classifier, wherein the video scene classifier includes a network encoder and a network decoder; Inputting the video segmentation frame set into the network encoder in sequence to obtain a multi-scale feature set of image frames; Based on the network decoder, the multi-scale feature set of the image frame is identified and classified in sequence, and the scene feature information of the video frame is output.
4. The sensitive information leakage protection system for video transmission according to claim 1, characterized in that: The construction of data processing multiple channels includes: Acquire a video content structured type, where the video content structured type includes image attribute data, audio attribute data, and text attribute data; Based on the image attribute data, the audio attribute data and the text attribute data, respectively, a preprocessing step configuration is performed to obtain a set of video attribute data preprocessing steps; Performing feature extraction algorithm analysis on the image attribute data, audio attribute data and text attribute data in sequence to obtain a set of video attribute data feature extraction steps; Based on the video attribute data preprocessing step set and the video attribute data feature extraction step set, the video content structured types are respectively loaded with processing channels to construct the data processing multi-channel.
5. The sensitive information leakage protection system for video transmission according to claim 1, characterized in that: The obtaining of partition content data sensitivity information includes: Acquire sensitive content description factor information, wherein the sensitive content description factor information includes content source partition, sensitive information type, sensitive information quantity and duration; Based on the sensitive content description factor information, feature descriptions are respectively performed on the partitioned sensitive content data sets to obtain partitioned sensitive data description feature sets; Performing a criticality assessment on each descriptive factor in the sensitive content descriptive factor information to obtain a criticality factor of the sensitive descriptive factor; A weighted quantitative evaluation is performed on the partition sensitive data description feature set based on the criticality factor of the sensitive description factor to obtain the partition content data sensitivity information.
6. The sensitive information leakage protection system for video transmission according to claim 1, characterized in that: The determining of partition content data desensitization algorithm information includes: Constructing a data desensitization algorithm list, wherein the data desensitization algorithm list includes data desensitization algorithms with different desensitization levels; Acquire a video transmission security standard, perform a desensitization rating on the partition content data sensitivity information based on the video transmission security standard, and determine a partition content data desensitization level; Based on matching and mapping the partition content data desensitization level and the data desensitization algorithm list, the partition content data desensitization algorithm information is determined.
7. The sensitive information leakage protection system for video transmission according to claim 1, characterized in that: The step of performing information encryption transmission and information leakage protection on the video content partition set includes: Performing information desensitization processing on the partition sensitive content data set in the video content partition set based on the partition content data desensitization algorithm information to obtain a partition video content desensitized data set; According to the partition content data sensitivity information of the video content partition set, matching analysis is performed with the video sensitive information protection strategy library to obtain a partition content sensitive protection strategy set; The partitioned video content desensitized data set is protected from information leakage based on the partitioned content sensitive protection strategy set.
8. The sensitive information leakage protection system for video transmission according to claim 7, characterized in that: The step of obtaining a partition content sensitive protection strategy set includes: Perform matching analysis based on the partition content data sensitivity information and the video sensitive information protection strategy library to obtain a partition matching protection strategy set; A multi-dimensional protection test is performed on the partition matching protection policy set to obtain a policy security protection effect set, and the partition content sensitive protection policy set is preferentially obtained based on the policy security protection effect set.
Citation Information
Patent Citations
Method and device for determining advertisement putting video, electronic equipment and storage medium
CN109819288A
Video cataloguing method and device
CN110121033A
Data grading method and device, and related equipment
CN110941956A
Multi-source multi-modal data processing system and method for applying system
CN111859451A
Scene clustering method and device in video and related equipment
CN114299435A
Cited By
Analysis and evaluation system for password security
CN120067359A
An analysis and evaluation system for password security
CN120067359B
Video conference security protection system based on multi-mode perception
CN121000936A
Dynamic desensitization processing method and system for sensitive information in video stream
CN121262428A
Method and system for dynamic desensitization of sensitive information in video stream
CN121262428B
