Multimedia content monitoring analysis method and system based on content distribution

Abstract

The invention discloses a multimedia content monitoring analysis method and system based on content distribution, and particularly relates to the technical field of multimedia monitoring. The method comprises the following steps: generating a globally unique identifier based on nodes of a distribution network, user identity information and content characteristics, embedding the identifier into a multimedia video stream of a low dynamic change region by using a robust digital watermarking algorithm, and recording a resolution switching frequency and a distribution log in a dynamic adaptive stream distribution process, the dynamic characteristics of the video are analyzed by combining the resolution switching frequency and the frame difference degree, the integrity loss rate is calculated based on the watermark extraction success rate, the integrity of watermark embedding is evaluated by using a machine learning model, the distribution strategy is optimized according to the evaluation result, and a complete evidence chain of extraction, analysis and tracing is generated. The robustness and traceability of the watermark in the dynamic switching environment are remarkably improved, the problems that the content is difficult to trace and copyright protection is weakened are effectively solved, and efficient and reliable technical support is provided for multimedia content distribution.

Description

Technical Field

[0001] The present invention relates to the field of multimedia monitoring technology, and in particular to a multimedia content monitoring and analysis method and system based on content distribution. Background Art

[0002] Multimedia content monitoring and analysis based on content distribution refers to the process of using technical means to monitor and analyze multimedia content in real time during its dissemination process through content distribution networks (CDN) or other distribution mechanisms. This technology can cover multimedia forms such as video, audio, and pictures, and aims to track key data such as content flow paths, access frequency, and user interaction.

[0003] This method is often used in scenarios such as copyright protection, content optimization, and user behavior analysis. By monitoring the distribution link, illegal use can be discovered in a timely manner, dissemination bottlenecks can be identified, and distribution efficiency can be optimized. At the same time, decision support for content recommendation or marketing strategies can be provided based on the analysis results. This plays an important role in maintaining content security, improving user experience, and supporting business growth. Digital watermarking is an invisible identifier embedded in multimedia content for copyright tracking and illegal distribution detection. For example, when a video stream is distributed through a CDN, a unique watermark identifier is embedded. When the content is illegally downloaded or disseminated without authorization, the monitoring system can identify the source of the leak by extracting the watermark information. This technology is widely used on streaming platforms (such as Netflix or YouTube) to prevent the spread of pirated content.

[0004] The prior art has the following deficiencies:

[0005] When video streams are distributed through CDNs, they usually use dynamic adaptive streaming (Adaptive Bitrate Streaming) technology to optimize bandwidth utilization. However, dynamic switching of different resolutions may accidentally destroy the integrity of watermark embedding, resulting in watermark extraction failure. In addition, dynamic adaptive stream switching causes the integrity of watermark embedding to be destroyed, which may cause the monitoring system to be unable to accurately extract the watermark, and thus unable to track the source of the content or determine the responsibility for the leak. This may be used maliciously, making it difficult to trace the leaked content, seriously weakening the effectiveness of copyright protection and content security policies, while increasing the risk of misjudgment and failure of legal accountability. Summary of the invention

[0006] The purpose of the present invention is to provide a multimedia content monitoring and analysis method and system based on content distribution to solve the shortcomings of the background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a multimedia content monitoring and analysis method based on content distribution, comprising the following steps:

[0008] S1: Generate a globally unique identifier for multimedia content based on the nodes of the distribution network, user identity information and content characteristics;

[0009] S2: using a robust digital watermarking algorithm, embedding the unique identifier into a video stream of the multimedia content, including selecting a target area for embedding the watermark, the target area being a low dynamic change area, and embedding the identifier information in the target area based on a discrete cosine transform technique;

[0010] S3: Distribute the multimedia content embedded with the watermark through dynamic adaptive streaming technology. During the distribution process, generate and record distribution logs, including content switching frequency and resolution change information;

[0011] S4: discovering suspected illegal multimedia content through web crawling or user reports, extracting watermark identifiers from suspected illegal content, and using inverse transformation algorithms to recover embedded identifiers from the frequency domain or pixel domain;

[0012] S5: extract the resolution switching frequency of the content from the distribution record, calculate the frame difference of the multimedia content, and measure the structural similarity between consecutive frames;

[0013] S6: Based on the watermark extraction success rate, the integrity loss rate is calculated, and a machine learning model is built using the integrity loss rate and frame difference to evaluate the integrity of watermark embedding during the distribution process;

[0014] S7: Optimize the distribution strategy based on the evaluation results, including limiting the switching frequency and improving the watermark embedding area, and generate a complete chain of evidence containing extraction, analysis and traceability information.

[0015] Preferably, in S2, the video stream is decomposed into single-frame images and divided into pixel blocks of fixed size, the pixel change value between the current frame and the previous frame is calculated based on the frame difference method, and the area with the smallest dynamic change is used as the target area for embedding the watermark. Within the target area, a low-frequency component area is further selected, and the spatial domain pixel value is converted into a frequency domain representation using a discrete cosine transform, and the unique identifier is encoded into a binary form and bit-wise embedded in the low-frequency coefficient. Watermark embedding is achieved by fine-tuning the coefficient, and the frequency domain data is restored to the spatial domain pixel value through an inverse discrete cosine transform.

[0016] Preferably, in S5, the resolution switching frequency of the content is extracted from the distribution record, the frame difference of the multimedia content is calculated, and the structural similarity between consecutive frames is measured, specifically:

[0017] Extract consecutive frames F from multimedia content t and F t-1, assume that each frame is a W×H grayscale image matrix, where W is the frame width, H is the frame height, and the pixel value range is [0,255]; if the frame is a color image, it must be converted to a grayscale image first: I(x,y)=0.2989·R(x,y)+0.5870·G(x,y)+0.1140·B(x,y); where R(x,y), G(x,y), B(x,y) are the red, green, and blue channel values ​​of the pixel; calculate the current frame F pixel by pixel t and the previous frame F t-1 The absolute difference D(x,y): D(x,y)=|F t (x,y)-F t-1 (x,y)|; where: D(x,y) is the difference value of the pixel position (x,y), F t (x, y) is the pixel value at position (x, y) in the current frame, F t-1 (x, y) is the pixel value at position (x, y) of the previous frame; all pixel difference values ​​are accumulated to obtain the total pixel difference between the two frames, which is expressed as: Normalize the total difference value to the average difference value to represent the frame difference: Where: FDI is the frame difference, W·H is the total number of pixels in the frame, which is used to normalize the difference value range.

[0018] Preferably, in S6, the integrity loss rate is calculated based on the watermark extraction success rate, specifically:

[0019] Define the known average extraction success rate as μ, set the total number of simulated trials N, and use normal distribution to generate random extraction success rate samples PSWE i : in, represents a normal distribution, μ is the mean, σ is the standard deviation; each trial generates a PSWE i Value; for each test sample PSWE i , calculate the corresponding watermark integrity loss rate WILR i , the expression is: WILR i =100%-PSWE i ; Summarize all WILR in N simulations i The value of N trials is used to find the average loss rate AWILR, which is expressed as: The standard deviation of the loss rate is calculated based on the average loss rate of N tests as the integrity loss rate.

[0020] Preferably, a machine learning model is constructed using the integrity loss rate and the frame difference degree to evaluate the integrity of watermark embedding during the distribution process. Specifically: the frame difference degree and the integrity loss rate are converted into a comprehensive feature vector, and the comprehensive feature vector is used as the input of the machine learning model. The machine learning model takes predicting the integrity value label of watermark embedding during the distribution process for each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of the prediction errors of the integrity value labels of watermark embedding during all distribution processes as the training target. The machine learning model is trained until the sum of the prediction errors reaches convergence and then the model training is stopped. The integrity value of watermark embedding during the distribution process is determined according to the model output result, where the machine learning model is a polynomial regression model.

[0021] Preferably, in S7, the obtained integrity value of watermark embedding during the distribution process is compared with the reference threshold of the integrity value preset according to historical data. If the integrity value of watermark embedding during the distribution process is greater than or equal to the preset reference threshold of the integrity value, it indicates that the integrity of watermark embedding during the distribution process is high, and no warning signal is generated at this time; if the integrity value of watermark embedding during the distribution process is less than the preset reference threshold of the integrity value, it indicates that the integrity of watermark embedding during the distribution process is low, and a warning signal is generated at this time to optimize the distribution strategy.

[0022] Preferably, optimizing the distribution strategy specifically includes: if frequent resolution switching affects the integrity of watermark embedding, set the upper limit value SF of the switching frequency max ; where SF is the resolution switching frequency, N switch is the total number of resolution switches that occur during the playback process, and T playback is the total playback time;

[0023] According to the analysis result of the frame difference degree FDI, avoid embedding the watermark in areas with剧烈 dynamic changes, and preferably select low-frequency areas. The formula: TargetArea = {(x,y)∣FDI(x,y) ≤ FDI max}; TargetArea is the target area for watermark embedding, FDI(x,y) is the frame difference degree at the pixel position (x,y), and FDI max is the upper limit value of the frame difference degree;

[0024] Obtain the integrity value Ypred of watermark embedding calculated during the distribution process and the preset integrity reference threshold Ythreshold; if Ypred ≥ Ythreshold, it indicates that the integrity of watermark embedding is high and no warning signal needs to be generated; if Ypred < Ythreshold, it indicates that the integrity of watermark embedding is low, a warning signal is generated, and a complete evidence chain needs to be formed.

[0025] The present invention also provides a multimedia content monitoring and analysis system based on content distribution, including an identifier generation module, a watermark embedding module, a content distribution module, a content monitoring module, a switching frequency and frame difference analysis module, an integrity assessment module, and an optimization and response module:

[0026] Identifier generation module: Generates a globally unique identifier for multimedia content based on the nodes of the distribution network, user identity information and content characteristics;

[0027] Watermark embedding module: using a robust digital watermark algorithm to embed the unique identifier into the video stream of the multimedia content, specifically including selecting a target area for embedding the watermark, the target area being a low dynamic change area, and embedding the identifier information in the target area based on discrete cosine transform technology;

[0028] Content distribution module: distributes the multimedia content embedded with watermarks through dynamic adaptive streaming technology. During the distribution process, it generates and records distribution logs, including content switching frequency and resolution change information;

[0029] Content monitoring module: discovers suspected illegal multimedia content through web crawling or user reports, extracts watermark identifiers from suspected illegal content, and uses inverse transformation algorithms to recover embedded identifiers from the frequency domain or pixel domain;

[0030] Switching frequency and frame difference analysis module: extracts the resolution switching frequency of the content from the distribution record, calculates the frame difference of the multimedia content, and measures the structural similarity between consecutive frames;

[0031] Integrity assessment module: Based on the watermark extraction success rate, the integrity loss rate is calculated, and a machine learning model is established using the switching frequency and frame difference to evaluate the integrity of watermark embedding during the distribution process;

[0032] Optimization and response module: Optimizes the distribution strategy based on the evaluation results, including limiting the switching frequency and improving the watermark embedding area, and generates a complete chain of evidence including extraction, analysis and traceability information.

[0033] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0034] 1. The present invention generates a globally unique identifier and embeds it into the video stream in the low dynamic change area using a robust digital watermark algorithm, and combines discrete cosine transform technology to enhance the watermark's anti-compression and anti-noise capabilities; in the dynamic adaptive stream distribution process, the switching frequency and resolution change information are recorded in real time to ensure the traceability of the watermark information. In addition, through the analysis and modeling of frame difference and integrity loss rate, the machine learning model is used to dynamically evaluate the watermark embedding integrity, providing early warning and optimization basis for the potential watermark damage risk in the distribution process, thereby improving the robustness of watermark embedding and the security of distributed content.

[0035] 2. The present invention also introduces a flexible optimization strategy and a complete evidence chain generation mechanism, which can optimize the distribution strategy according to the evaluation results, such as limiting the resolution switching frequency and avoiding embedding watermarks in areas with drastic dynamic changes, thereby further improving the success rate of watermark extraction and the accuracy of tracing. By extracting and tracing the watermarks of illegally disseminated content, combined with an automated network monitoring and reporting mechanism, copyright protection and security management of content distribution are ensured. The system can be widely used in the fields of digital copyright protection, streaming media content distribution, and information security, providing strong technical support for maintaining the legal use of content and reducing copyright disputes. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0037] Figure 1 The present invention is a flow chart of the method.

[0038] Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0040] Example 1, please refer to Figure 1 As shown, the multimedia content monitoring and analysis method based on content distribution described in this embodiment includes the following steps:

[0041] S1: Generate a globally unique identifier for multimedia content based on the nodes of the distribution network, user identity information and content characteristics;

[0042] S2: using a robust digital watermarking algorithm, embedding the unique identifier into a video stream of the multimedia content, including selecting a target area for embedding the watermark, the target area being a low dynamic change area, and embedding the identifier information in the target area based on a discrete cosine transform technique;

[0043] S3: Distribute the multimedia content embedded with the watermark through dynamic adaptive streaming technology. During the distribution process, generate and record distribution logs, including content switching frequency and resolution change information;

[0044] S4: discovering suspected illegal multimedia content through web crawling or user reports, extracting watermark identifiers from suspected illegal content, and using inverse transformation algorithms to recover embedded identifiers from the frequency domain or pixel domain;

[0045] S5: extract the resolution switching frequency of the content from the distribution record, calculate the frame difference of the multimedia content, and measure the structural similarity between consecutive frames;

[0046] S6: Based on the watermark extraction success rate, the integrity loss rate is calculated, and a machine learning model is built using the integrity loss rate and frame difference to evaluate the integrity of watermark embedding during the distribution process;

[0047] S7: Optimize the distribution strategy based on the evaluation results, including limiting the switching frequency and improving the watermark embedding area, and generate a complete chain of evidence containing extraction, analysis and traceability information.

[0048] In S1, a globally unique identifier is generated for the multimedia content based on the nodes of the distribution network, user identity information and content characteristics, specifically:

[0049] Node information of the distribution network: Each node in the distribution network (such as the edge server of the CDN) usually has a unique identifier, such as an IP address, node ID, or geographic location code.

[0050] Utilization of node information: Encode the unique identification of the node as part of the identifier, such as in the form of a string or number. Example: If the IP address of the node is 192.168.0.1, it can be converted to 192168001 as part of the identifier. Ensure that the content distributed by different nodes can be distinguished, so as to achieve source tracking.

[0051] User identity information: For each end user or subscriber, a unique user information portion is generated through a user ID, account ID, or device identifier (such as MAC address or device IMEI).

[0052] Dynamic association: To prevent the identifier from being easily forged, a timestamp or session identifier can be added to ensure the uniqueness of the user identifier within a specific time. Example: If the user ID is U12345 and the timestamp is 20241228T1030, the combination is U12345-20241228T1030. This enables accurate tracking of each user or session, making it easier to identify the party responsible for content leakage.

[0053] Extract the unique characteristics of multimedia content to generate fingerprints (for example, calculate the MD5 value or SHA value of the content based on a hash algorithm). Fingerprint calculation can be based on the metadata of the content (such as file name, size, creation time) or the content itself (such as pixel features of video key frames). For example: MD5 calculation is performed on the key frame data of a video stream, and the result is AB12CD34EF56. For different resolution versions of the same content, additional bit rate or resolution information is used to ensure fingerprint uniqueness. Example: The resolution is marked as 1080p and the bit rate is 5Mbps, and the fingerprint is AB12CD34EF56-1080p-5Mbps. Avoid identifier conflicts due to similar content or multiple versions. A globally unique identifier is formed by combining distribution nodes, user identity information, and content characteristics in a regularized manner. Rule design: Format: [node identifier]-[user identifier]-[content fingerprint] Example: Node identifier: 192168001; User identifier: U12345-20241228T1030; Content fingerprint: AB12CD34EF56-1080p-5Mbps; Final identifier: 192168001-U12345-20241228T1030-AB12CD34EF56-1080p-5Mbps.

[0054] S2: using a robust digital watermark algorithm, embedding the unique identifier into the video stream of the multimedia content, including selecting a target area for embedding the watermark, the target area being a low dynamic change area, and embedding the identifier information in the target area based on discrete cosine transform technology.

[0055] Decompose the video stream into single-frame images, and select several frames per second (such as key frames) for watermark embedding. For each frame of the image, divide it into multiple fixed-size blocks (such as 8×8 or 16×16 pixel blocks). Calculate the dynamic change degree of each block, and select the low dynamic change area as the target area: Use the frame difference method to calculate the pixel change value D(x,y) of the corresponding block of the current frame and the previous frame. The expression is: Among them, F t and F t-1are the pixel values ​​of the current frame and the previous frame respectively, x, y are the coordinates of the block, W, H are the width and height of the block. The area with the smallest change value D(x, y) is selected as the embedding target area. In the target area, the area where the low-frequency component is located is further selected because the low-frequency component has a high robustness to compression and noise.

[0056] Perform DCT on the selected target area block to convert the pixel value in the spatial domain into the frequency domain C(u,v): Among them, f(x,y) is the pixel value in the block, C(u,v) is the frequency domain coefficient, and α(u), α(v) are normalization factors.

[0057] The unique identifier is encoded into binary form and embedded bit by bit into the low-frequency component of the DCT coefficient: the low-frequency coefficient of the target area is selected (such as C(1,1), C(1,2), etc.).

[0058] According to the value of the identifier bit (0 or 1), the coefficients are fine-tuned: C′(u,v)=C(u,v)+Δ; where: C′(u,v) is the coefficient after embedding. Δ is the modulation amplitude (to ensure that the embedded information does not affect the human eye perception).

[0059] Perform an inverse discrete cosine transform (IDCT) on the modified DCT coefficients to restore the frequency domain data to the spatial domain pixel value f′(x,y), expressed as: Replace the corresponding pixel blocks in the video frame to complete the watermark embedding.

[0060] Extract the target area block from the video stream embedded with the watermark and perform DCT calculation. Compare the changes in low-frequency coefficients and decode the embedded binary identifier based on the change value. Simulate compression, cropping or noise addition to the video stream after the watermark is embedded to verify whether the extraction result is accurate. Ensure that the watermark information can still be correctly extracted under common content processing operations.

[0061] S3: The multimedia content embedded with the watermark is distributed through the dynamic adaptive streaming technology. During the distribution process, a distribution log is generated and recorded, including content switching frequency and resolution change information.

[0062] Transcode the multimedia content embedded with watermark into versions with different resolutions and bit rates (e.g. 144p, 360p, 720p, 1080p, etc.).

[0063] Adaptive Bitrate (ABR) technology is used to package each version into segmented files (such as HLS or MPEG-DASH format) to support resolution switching under different network conditions.

[0064] Divide the video content into multiple time segments (e.g., 2-10 seconds per segment), and generate multiple resolution versions for each segment. Example: Generate 144p, 360p, and 720p resolution versions for a 10-second video segment. Record segment index and watermark information for subsequent tracing.

[0065] Multi-resolution content is pushed to nodes around the world through the content distribution network (CDN). The client dynamically selects the appropriate resolution for playback based on the user's network status (such as bandwidth, latency): when network conditions are poor, a lower resolution (such as 360p) is automatically selected. When network conditions improve, it switches to a higher resolution (such as 720p or 1080p).

[0066] The client triggers the resolution switching mechanism by monitoring the playback status (such as buffering rate and loading speed) in real time. The dynamic switching is controlled by the client's player, which obtains the adapted segment through the protocol (such as HLS's .m3u8 file or DASH's MPD file).

[0067] Switching event detection: Monitor resolution switching events in the player and record the time and frequency of each resolution change. Example: Switching time: T1, T2, T3; Resolution sequence: 360p→720p→360p→1080p; Frequency: Calculate the ratio of switching times to playback time.

[0068] Resolution track: Records the resolution selection of each segment during the entire playback process. Example: Playback segment sequence: Segment 1: 360p; Segment 2: 720p; Segment 3: 1080p; Resolution track: [360p, 720p, 1080p]. Bitrate change record: Combined with the resolution, record the actual bitrate of each segment to analyze the impact of the network on the distribution effect.

[0069] Store logs as structured data for easy retrieval and analysis: Timestamp: records the exact time when the switching event occurs. Resolution: records the resolution value before and after the switching. User ID: identifies the user of the playback session. CDN node: records the source node of the distributed content. Network status: records the bandwidth and latency information at the time of the switching.

[0070] Bind a unique watermark identifier to each segment in the log to ensure that the distribution link is traceable. Example: Segment 1 (360p): watermark identifier A1B2C3; Segment 2 (720p): watermark identifier A1B2C3. Dynamically record the mapping relationship between switching frequency and resolution trajectory and watermark to ensure that the specific distribution path can be traced back when monitoring illegal content.

[0071] S4: Through web crawling or user reporting, multimedia content suspected of illegal dissemination is discovered, watermark identifiers are extracted from the suspected illegal content, and the embedded identifiers are recovered from the frequency domain or pixel domain using an inverse transformation algorithm.

[0072] Set crawling targets, such as video sharing platforms, pirated websites, P2P networks, or social media. Use crawler tools (such as Scrapy, Beautiful Soup) to scan target platforms regularly. Quickly locate suspected illegal content through specific keywords (such as video titles, content descriptions) or hash fingerprints. Download matching content files for subsequent analysis. Provide reporting channels (such as online forms or API interfaces) to allow users to upload suspected illegal content or links. Verify the legitimacy of the reported content and download sample content.

[0073] Decompose the suspected illegal video content into single-frame images and divide them into pixel blocks, which is consistent with the block division method when embedding watermarks. According to the information of the embedding algorithm, locate the low-dynamic change area or the specific position of the frequency domain, and extract the target area for watermark recovery. According to the algorithm when embedding watermarks, select the appropriate domain for watermark extraction: Frequency domain extraction: Perform discrete cosine transform (DCT) on the target area block to obtain frequency domain coefficients. Extract the change values ​​in low-frequency components (such as C(1,1), C(1,2), etc.), which contain the embedded identifier information. Extract watermark information directly from the pixel domain, such as reading the embedded binary data through the least significant bit (LSB).

[0074] Recover the watermark identifier from the extracted frequency domain or pixel domain data using the inverse transform algorithm: Frequency domain inverse recovery: Decode the corresponding binary data by comparing the embedded offset value (such as the presence or absence of Δ) of the frequency domain coefficients. Pixel domain inverse recovery: Extract the watermark value of each bit from the pixel data and combine them into a complete binary identifier.

[0075] Convert the extracted binary data back into a unique identifier (such as a combined string of distribution node, user ID, and content characteristics). Verify the integrity of the identifier to ensure that the extraction result is consistent with the embedded identifier format.

[0076] Verify the integrity of the watermark extraction result to ensure that it can still be correctly extracted under common processing (such as compression, cropping, noise interference): Check whether there is any damage or information loss in the identifier. If the identifier is incomplete, try to recover the lost part based on redundant coding.

[0077] The extracted watermark identifier is compared with the records in the legal content distribution log to verify the distribution source and distribution path of the content. The specific source of the illegal content is determined by matching the distribution node, user ID and timestamp in the database.

[0078] Based on the extracted watermark identifier, determine the source of the illegally disseminated content, including: distribution node (such as CDN server or cache node). Responsible user (such as account ID or device identifier). Distribution time and resolution version (such as 720p version was leaked at a specific time point). Save illegal content samples, watermark extraction process and results to generate a complete chain of evidence for legal accountability or internal analysis. Compare the extracted identifier with the distribution log and record it.

[0079] S5: Extract the resolution switching frequency of the content from the distribution record, calculate the frame difference of the multimedia content, and measure the structural similarity between consecutive frames.

[0080] Extract consecutive frames F from multimedia content t and F t-1 , assume that each frame is a W×H grayscale image matrix, where W is the frame width, H is the frame height, and the pixel value range is [0,255]; if the frame is a color image, it must be converted to a grayscale image first: I(x,y)=0.2989·R(x,y)+0.5870·G(x,y)+0.1140·B(x,y); where R(x,y), G(x,y), B(x,y) are the red, green, and blue channel values ​​of the pixel. Calculate the current frame F pixel by pixel t and the previous frame F t-1 Absolute difference: D(x,y)=F t (x,y)-F t-1 (x,y)|; where: D(x,y) is the difference value of the pixel position (x,y), F t (x, y) is the pixel value at position (x, y) in the current frame, F t-1 (x,y) is the pixel value at position (x,y) in the previous frame.

[0081] All pixel difference values ​​are accumulated to obtain the total pixel difference between the two frames, which is expressed as: Normalize the total difference value to the average difference value to represent the frame difference: Where: FDI is the frame difference, which represents the average difference value per pixel, and W·H is the total number of pixels in the frame, which is used to normalize the difference value range.

[0082] Frame difference is an indicator that measures the degree of pixel change between consecutive frames. The larger the value, the more significant the pixel change between frames, and the lower the structural similarity between consecutive frames. This usually occurs when there is fast motion, drastic scene switching, or dynamic content changes in the video, such as chase scenes in action movies or transition switching in videos. This low similarity may affect the efficiency of the compression algorithm because more bit rate needs to be allocated for more changing details.

[0083] On the contrary, the smaller the frame difference, the less pixel changes between frames, and the higher the structural similarity between consecutive frames. This usually occurs when the video content is relatively static, such as a fixed-lens dialogue scene or a low-dynamic background. High similarity indicates that there is more redundant information between frames, which is very beneficial for video compression and content distribution, reducing bitrate requirements and improving distribution efficiency. Small frame differences may also mean that it is more suitable for embedding watermarks or other hidden information.

[0084] S6: Based on the watermark extraction success rate, the integrity loss rate is calculated, and the machine learning model is constructed using the integrity loss rate and frame difference to evaluate the integrity of watermark embedding during the distribution process.

[0085] Define a known average extraction success rate, denoted as μ (e.g., μ = 95%). Define the fluctuation range of the extraction success rate (e.g., standard deviation σ = 2%) based on historical data or experimental observations. Set the total number of simulation trials N (usually more than 10,000 times to ensure the stability of the results).

[0086] Generate a random sample of extraction success rates PSWE using a normal distribution (or other suitable distribution, such as Beta distribution) i : in, represents a normal distribution, μ is the mean, and σ is the standard deviation. Each trial generates a PSWE i Ensure that the generated extraction success rate is within a reasonable range (i.e. 0≤PSWE i ≤100%), values ​​outside the range need to be regenerated. For each test sample PSWE i , calculate the corresponding watermark integrity loss rate WILR i , the expression is: WILR i =100%-PSWE i ; Summarize all WILR in N simulations i The value of N trials is used to find the average loss rate AWILR, which is expressed as: The standard deviation of the loss rate is calculated based on the average loss rate of N tests as the integrity loss rate.

[0087] The frame difference and integrity loss rate are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the integrity value label of the watermark embedded in the distribution process as the prediction target, and takes minimizing the sum of prediction errors of the integrity value labels of the watermark embedded in all distribution processes as the training target. The machine learning model is trained until the sum of prediction errors converges and the model training is stopped. The integrity value of the watermark embedded in the distribution process is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0088] The method for obtaining the integrity value of watermark embedding during distribution is as follows: from the comprehensive feature vector training data of the trained machine learning model, obtain the corresponding function expression: LQ = F(FDI, GEH); where F is the output function of the model, FDI is the frame difference degree, GEH is the integrity loss rate, and LQ is the integrity value of watermark embedding during distribution.

[0089] S7: Optimize the distribution strategy according to the evaluation results, including restricting the switching frequency and improving the watermark embedding area, and generate a complete evidence chain containing extraction, analysis, and traceability information.

[0090] Compare the obtained integrity value of watermark embedding during distribution with the reference threshold of the integrity value preset according to historical data. If the integrity value of watermark embedding during distribution is greater than or equal to the reference threshold of the preset integrity value, it indicates that the integrity of watermark embedding during distribution is high, and no warning signal is generated at this time; if the integrity value of watermark embedding during distribution is less than the reference threshold of the preset integrity value, it indicates that the integrity of watermark embedding during distribution is low, and a warning signal is generated at this time to optimize the distribution strategy.

[0091] The specific optimization of the distribution strategy includes: if frequent resolution switching (high switching frequency) has an obvious impact on the integrity of watermark embedding, the upper limit value SF of the switching frequency can be set max ; where SF is the resolution switching frequency (times / second), N switch is the total number of resolution switches that occur during playback, and T playback is the total playback time (seconds).

[0092] According to the analysis result of the frame difference degree FDI, avoid embedding the watermark in areas with剧烈 dynamic changes (high FDI areas), and preferentially select static background areas or low-frequency areas. The formula: TargetArea = {(x,y)∣FDI(x,y)≤FDI max}; TargetArea is the target area for watermark embedding, FDI(x,y) is the frame difference degree at the pixel position (x,y), and FDI max is the upper limit value of the frame difference degree, which is used to select low-dynamic areas.

[0093] Obtain the integrity value Ypred of watermark embedding calculated during distribution and the preset integrity reference threshold Ythreshold; if Ypred≥Ythreshold, it indicates that the integrity of watermark embedding is high, and no warning signal needs to be generated; if Ypred<Ythreshold, it indicates that the integrity of watermark embedding is low, and a warning signal is generated.

[0094] If an early warning signal is generated, a complete chain of evidence must be formed, including: the extracted watermark identifier and the corresponding extraction success rate and integrity loss rate, the frame difference FDI of the distributed data, the calculated integrity value Ypred, and the distribution node identifier (such as CDN node IP), the responsible user identifier (such as user ID or device ID), and the above information is organized into structured documents or log records for subsequent tracking and analysis.

[0095] In this embodiment, first, a globally unique identifier is generated for each content according to the nodes of the distribution network, user identity information and content characteristics; then, a robust digital watermarking algorithm is used to embed the identifier into the video stream, and a low dynamic change area is selected as the embedding target, and the watermark is hidden through discrete cosine transform technology. The content embedded with the watermark is distributed through dynamic adaptive streaming technology, and a log is generated during the distribution process to record the switching frequency and resolution transformation information of the content. At the same time, suspected illegal dissemination of content is found through network crawling or user reports, the watermark identifier is extracted from it, and the identifier information is restored from the frequency domain or pixel domain through the inverse transformation algorithm. The switching frequency is extracted from the distribution record and the frame difference is calculated to measure the structural similarity between consecutive frames. Based on the watermark extraction success rate, the integrity loss rate is calculated, and a machine learning model is constructed in combination with the switching frequency and frame difference to evaluate the integrity of the watermark embedding during the distribution process. Finally, the distribution strategy is optimized according to the evaluation results, such as limiting the switching frequency and improving the watermark embedding area, and a complete chain of evidence containing extraction, analysis and traceability information is generated to improve the robustness of watermark protection and traceability.

[0096] Example 2, please refer to Figure 2 As shown, the multimedia content monitoring and analysis system based on content distribution described in this embodiment includes an identifier generation module, a watermark embedding module, a content distribution module, a content monitoring module, a switching frequency and frame difference analysis module, an integrity assessment module, and an optimization and response module:

[0097] Identifier generation module: Generates a globally unique identifier for multimedia content based on the nodes of the distribution network, user identity information and content characteristics;

[0098] Watermark embedding module: using a robust digital watermark algorithm to embed the unique identifier into the video stream of the multimedia content, specifically including selecting a target area for embedding the watermark, the target area being a low dynamic change area, and embedding the identifier information in the target area based on discrete cosine transform technology;

[0099] Content distribution module: distributes the multimedia content embedded with watermarks through dynamic adaptive streaming technology. During the distribution process, it generates and records distribution logs, including content switching frequency and resolution change information;

[0100] Content monitoring module: discovers suspected illegal multimedia content through web crawling or user reports, extracts watermark identifiers from suspected illegal content, and uses inverse transformation algorithms to recover embedded identifiers from the frequency domain or pixel domain;

[0101] Switching frequency and frame difference analysis module: extracts the resolution switching frequency of the content from the distribution record, calculates the frame difference of the multimedia content, and measures the structural similarity between consecutive frames;

[0102] Integrity assessment module: Based on the watermark extraction success rate, the integrity loss rate is calculated, and a machine learning model is established using the switching frequency and frame difference to evaluate the integrity of watermark embedding during the distribution process;

[0103] Optimization and response module: Optimizes the distribution strategy based on the evaluation results, including limiting the switching frequency and improving the watermark embedding area, and generates a complete chain of evidence including extraction, analysis and traceability information.

[0104] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0105] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0106] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0107] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A multimedia content monitoring and analysis method based on content distribution, characterized in that: The following steps are involved: S1: Generate a globally unique identifier for multimedia content based on the nodes of the distribution network, user identity information and content characteristics; S2: using a robust digital watermarking algorithm, embedding the unique identifier into a video stream of the multimedia content, including selecting a target area for embedding the watermark, the target area being a low dynamic change area, and embedding the identifier information in the target area based on a discrete cosine transform technique; S3: Distribute the multimedia content embedded with the watermark through dynamic adaptive streaming technology. During the distribution process, generate and record distribution logs, including content switching frequency and resolution change information; S4: discovering suspected illegal multimedia content through web crawling or user reports, extracting watermark identifiers from suspected illegal content, and using inverse transformation algorithms to recover embedded identifiers from the frequency domain or pixel domain; S5: extract the resolution switching frequency of the content from the distribution record, calculate the frame difference of the multimedia content, and measure the structural similarity between consecutive frames; S6: Based on the watermark extraction success rate, the integrity loss rate is calculated, and a machine learning model is built using the integrity loss rate and frame difference to evaluate the integrity of watermark embedding during the distribution process; S7: Optimize the distribution strategy based on the evaluation results, including limiting the switching frequency and improving the watermark embedding area, and generate a complete chain of evidence containing extraction, analysis and traceability information.

2. The method for monitoring and analyzing multimedia content based on content distribution according to claim 1, characterized in that: In S2, the video stream is decomposed into single-frame images and divided into pixel blocks of fixed size. The pixel change value between the current frame and the previous frame is calculated based on the frame difference method. The area with the smallest dynamic change is used as the target area for embedding the watermark. Within the target area, the low-frequency component area is further selected, and the spatial domain pixel value is converted into a frequency domain representation using discrete cosine transform. The unique identifier is encoded in binary form and embedded bit by bit in the low-frequency coefficient. Watermark embedding is achieved by fine-tuning the coefficient, and the frequency domain data is restored to spatial domain pixel values ​​through inverse discrete cosine transform.

3. The method for monitoring and analyzing multimedia content based on content distribution according to claim 2, characterized in that: In S5, the resolution switching frequency of the content is extracted from the distribution record, the frame difference of the multimedia content is calculated, and the structural similarity between consecutive frames is measured, specifically: Extract consecutive frames F from multimedia content t and F t-1 , assume that each frame is a W×H grayscale image matrix, where W is the frame width, H is the frame height, and the pixel value range is [0,255]; if the frame is a color image, it must be converted to a grayscale image first: I(x,y)=0.2989·R(x,y)+0.5870·G(x,y)+0.1140·B(x,y); where R(x,y), G(x,y), B(x,y) are the red, green, and blue channel values ​​of the pixel; calculate the current frame F pixel by pixel t and the previous frame F t-1 The absolute difference D(x,y): D(x,y)=|F t (x,y)-F t-1 (x,y)|; where: D(x,y) is the difference value of the pixel position (x,y), F t (x, y) is the pixel value at position (x, y) in the current frame, F t-1 (x, y) is the pixel value at position (x, y) of the previous frame; all pixel difference values ​​are accumulated to obtain the total pixel difference between the two frames, which is expressed as: Normalize the total difference value to the average difference value to represent the frame difference: Where: FDI is the frame difference, W·H is the total number of pixels in the frame, which is used to normalize the difference value range.

4. The method for monitoring and analyzing multimedia content based on content distribution according to claim 3 is characterized in that: In S6, the integrity loss rate is calculated based on the watermark extraction success rate, specifically: Define the known average extraction success rate as μ, set the total number of simulated trials N, and use normal distribution to generate random extraction success rate samples PSWE i : i=1,2,…,N; where represents a normal distribution, μ is the mean, σ is the standard deviation; each trial generates a PSWE i Value; for each test sample PSWE i , calculate the corresponding watermark integrity loss rate WILR i , the expression is: WILR i =100%-PSWE i ; Summarize all WILR in N simulations i The value of N trials is used to find the average loss rate AWILR, which is expressed as: The standard deviation of the loss rate is calculated based on the average loss rate of N tests as the integrity loss rate.

5. The method for monitoring and analyzing multimedia content based on content distribution according to claim 4 is characterized in that: A machine learning model is constructed using the integrity loss rate and frame difference to evaluate the integrity of the watermark embedding during the distribution process. Specifically, the frame difference and integrity loss rate are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the integrity value label of the watermark embedded in the distribution process as the prediction target, and minimizes the sum of prediction errors of the integrity value labels of the watermark embedded in all distribution processes as the training target. The machine learning model is trained until the sum of prediction errors converges and the model training is stopped. The integrity value of the watermark embedded in the distribution process is determined according to the model output results. The machine learning model is a polynomial regression model.

6. The multimedia content monitoring and analysis method based on content distribution according to claim 1 is characterized in that: In S7, the integrity value of watermark embedding during the distribution process obtained is compared with the reference threshold of the integrity value preset according to historical data. If the integrity value of watermark embedding during the distribution process is greater than or equal to the reference threshold of the preset integrity value, it indicates that the integrity of watermark embedding during the distribution process is high, and no warning signal is generated at this time; If the integrity value of watermark embedding during the distribution process is less than the reference threshold of the preset integrity value, it indicates that the integrity of watermark embedding during the distribution process is low, and a warning signal is generated at this time to optimize the distribution strategy.

7. The method for monitoring and analyzing multimedia content based on content distribution according to claim 6, characterized in that: The optimization distribution strategy specifically includes: if frequent resolution switching affects the integrity of watermark embedding, set the upper limit of the switching frequency SF max ; SF≤SF max ; Where SF is the resolution switching frequency, N switch is the total number of resolution switching that occurs during playback, T playback is the total playing time; According to the frame difference FDI analysis results, avoid embedding the watermark in the area with drastic dynamic changes, and give priority to the low-frequency area. The formula is: Target Area = {(x, y) | FDI (x, y) ≤ FDI max }; Target Area is the target area for watermark embedding, FDI(x,y) is the frame difference of the pixel position (x,y), FDI max is the upper limit of frame difference; Obtain the integrity value of watermark embedding Ypred calculated during the distribution process and the preset integrity reference threshold Ythreshold; if Ypred≥Ythreshold, it indicates that the integrity of watermark embedding is high and no warning signal needs to be generated; if Ypred<Ythreshold, it indicates that the integrity of watermark embedding is low, a warning signal is generated, and a complete evidence chain needs to be formed.

8. A multimedia content monitoring and analysis system based on content distribution, used to implement the multimedia content monitoring and analysis method based on content distribution according to any one of claims 1 to 7, characterized in that: It includes an identifier generation module, a watermark embedding module, a content distribution module, a content monitoring module, a switching frequency and frame difference degree analysis module, an integrity evaluation module, and an optimization and response module: Identifier generation module: Generate a globally unique identifier for the multimedia content according to the nodes of the distribution network, user identity information, and content characteristics; Watermark embedding module: Use a robust digital watermarking algorithm to embed the unique identifier into the video stream of the multimedia content. Specifically, it includes selecting the target area for watermark embedding, where the target area is a low-dynamic change area, and based on the discrete cosine transform technology, embedding the identifier information in the target area; Content distribution module: Distribute the multimedia content with the embedded watermark through dynamic adaptive streaming technology. During the distribution process, generate and record the distribution log, including content switching frequency and resolution transformation information; Content monitoring module: Discover suspected illegally distributed multimedia content through web crawling or user reports, extract the watermark identifier from the suspected illegal content, and use the inverse transformation algorithm to recover the embedded identifier from the frequency domain or pixel domain; Switching frequency and frame difference degree analysis module: Extract the resolution switching frequency of the content from the distribution record, calculate the frame difference degree of the multimedia content, and measure the structural similarity between consecutive frames; Integrity evaluation module: Calculate the integrity loss rate based on the watermark extraction success rate, establish a machine learning model using the switching frequency and frame difference degree, and evaluate the integrity of watermark embedding during the distribution process; Optimization and response module: Optimize the distribution strategy according to the evaluation results, including restricting the switching frequency and improving the watermark embedding area, and generate a complete evidence chain containing extraction, analysis, and traceability information.

Images