Advertisement material infringement detection method and device, equipment and storage medium

Abstract

The invention discloses an advertisement material infringement detection method and device, equipment and a storage medium. The method comprises the following steps: obtaining advertisement putting data and advertisement copyright materials, wherein the advertisement putting data comprises a plurality of advertisement candidate materials; determining a candidate material visual feature corresponding to each advertisement candidate material and a copyright material visual feature corresponding to the advertisement copyright material, and determining a suspected infringement material based on each candidate material visual feature and the copyright material visual feature; candidate material semantic features corresponding to each suspected infringement material and copyright material semantic features of each advertisement copyright material are extracted, infringement confidence is determined based on each candidate material semantic feature and the copyright material semantic features, and the suspected infringement material with the infringement confidence greater than a confidence threshold is determined as an infringement confirmation material. According to the scheme, through multi-dimensional feature extraction and comparison, in combination with deep fusion analysis of visual features and semantic features, the accuracy and robustness of advertisement material infringement recognition are effectively improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, device, and storage medium for detecting copyright infringement of advertising materials. Background Technology

[0002] With the rapid development of the digital advertising industry, the deployment of multimedia advertising materials such as videos and images on major platforms has grown exponentially. However, the problem of advertising content infringement has also become increasingly serious, especially the unauthorized use of others' original video materials and the editing and recombining of existing content for advertising promotion, which has become one of the core issues troubling platforms and content creators.

[0003] Traditional methods for identifying advertising infringement often rely on manual review or comparison based on content fingerprints. These methods suffer a significant drop in accuracy when advertising materials have been edited, compressed, obscured, or watermarked, making them unsuitable for complex infringement detection scenarios. Furthermore, existing methods generally lack the ability to collaboratively analyze advertising materials at both the visual and semantic levels, easily leading to misjudgments and missed detections due to visual similarity but semantic irrelevance or semantic similarity but different presentation. Simultaneously, advertising content distribution platforms exhibit differentiated dissemination patterns and infringement risk characteristics. Without dynamic analysis of platform data for candidate material selection, infringement detection efficiency is often low. Summary of the Invention

[0004] This application provides a method, apparatus, device, and storage medium for detecting advertising material infringement. It obtains advertising delivery data from an advertising platform and extracts candidate advertising materials, while simultaneously acquiring copyrighted advertising materials for infringement comparison. Based on these materials, visual features are extracted from both candidate and copyrighted materials, and the visual feature similarity between them is calculated. Candidate materials with similarity exceeding a threshold are identified as suspected infringing materials. Furthermore, semantic features are extracted from both suspected and copyrighted materials. The degree of matching between the semantic features of candidate and copyrighted materials is then used to ultimately confirm the infringing material. This technical solution effectively solves the problems of insufficient ability to identify complex processing forms such as material reconstruction and editing / masking in existing advertising content infringement detection methods, as well as high false positive rates and response delays. It achieves the ability to efficiently screen and accurately locate infringing behavior from a large number of advertising materials, significantly improving the automated detection level and accuracy of advertising content copyright protection.

[0005] Firstly, this application provides a method for detecting copyright infringement in advertising materials, including: Acquire the collected advertising placement data and the advertising copyright materials used for infringement investigation, wherein the advertising placement data includes several candidate advertising materials; Determine the visual features of the candidate material corresponding to each of the advertising candidate materials, and the visual features of the copyright material corresponding to each of the advertising copyright materials. Calculate the visual feature similarity between each of the candidate material visual features and the copyright material visual features. Identify the advertising candidate materials whose visual feature similarity is greater than a similarity threshold as suspected infringing materials. Extract the semantic features of candidate materials corresponding to each of the suspected infringing materials, as well as the semantic features of copyright materials for each of the advertising copyright materials. Determine the infringement confidence level based on the semantic features of each candidate material and the semantic features of the copyright materials. Identify the suspected infringing materials whose infringement confidence level is greater than the confidence level threshold as confirmed infringing materials.

[0006] Secondly, this application provides an advertising material infringement detection device, comprising: The acquisition module is used to acquire collected advertising placement data and advertising copyright materials used for infringement investigation. The advertising placement data includes several candidate advertising materials. The filtering module is used to determine the visual features of the candidate materials corresponding to each of the advertising candidate materials and the visual features of the copyright materials corresponding to each of the advertising copyright materials, calculate the visual feature similarity between each of the candidate material visual features and the copyright material visual features, and identify the advertising candidate materials whose visual feature similarity is greater than a similarity threshold as suspected infringing materials; The determination module is used to extract the semantic features of candidate materials corresponding to each suspected infringing material, as well as the semantic features of copyright materials of each advertising copyright material, determine the infringement confidence level based on the semantic features of each candidate material and the semantic features of the copyright material, and determine the suspected infringing material with the infringement confidence level greater than the confidence level threshold as confirmed infringing material.

[0007] Thirdly, this application provides an advertising material infringement detection device, comprising: One or more processors; A memory that stores one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the advertising material infringement detection method as described in the first aspect.

[0008] Fourthly, this application provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the advertising material infringement detection method as described in the first aspect.

[0009] This application constructs a feature comparison system for identifying advertising material infringement by acquiring collected advertising placement data and copyrighted advertising materials used for infringement investigation. First, based on image processing and computer vision technologies, visual features corresponding to each candidate advertising material and the copyrighted material are extracted. Then, according to similarity calculation rules in the feature space, the visual feature similarity between the candidate materials and the copyrighted material is evaluated. Further, candidate materials with similarity exceeding a preset threshold are marked as suspected infringing materials. For suspected infringing materials, a semantic-level feature extraction process is implemented to extract the embedding representations of the candidate materials and the copyrighted material in the semantic space. Combining deep feature matching and content expression consistency calculation mechanisms, the semantic consistency between materials is evaluated, and a corresponding infringement confidence score is generated accordingly. Finally, materials with an infringement confidence score exceeding the confidence threshold are automatically determined as confirmed infringing materials, serving as the basis for subsequent processing and legal compliance assessment. This method, by constructing a dual-channel analysis link of visual and semantic aspects, effectively improves the accuracy and automation level of advertising material infringement identification, taking into account both image similarity and semantic consistency dimensions, and reducing the false positive and false negative rates. Meanwhile, the proposed feature extraction and judgment process has good model compatibility and platform adaptability, and can be widely applied to infringement monitoring tasks in various advertising scenarios, with high technical practical value and promotion potential. Attached Figure Description

[0010] Figure 1 This is a flowchart of the advertising material infringement detection method provided in the embodiments of this application; Figure 2 This is a flowchart illustrating the frequency control of advertising data collection provided in an embodiment of this application; Figure 3 This is a flowchart illustrating the confirmation of advertising candidate materials provided in this application embodiment; Figure 4 This is a flowchart of the material visual feature extraction method provided in the embodiments of this application; Figure 5 This is a flowchart illustrating the keyframe extraction of copyrighted material provided in an embodiment of this application; Figure 6 This is a flowchart of the semantic feature extraction of candidate materials provided in the embodiments of this application; Figure 7 This is a flowchart illustrating the confirmation of infringing material warning provided in an embodiment of this application; Figure 8 This is a flowchart illustrating the steps for determining infringing materials as provided in an embodiment of this application; Figure 9 This is a structural diagram of the advertising material infringement detection device provided in the embodiments of this application; Figure 10 This is a structural diagram of the advertising material infringement detection device provided in the embodiments of this application. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as being processed sequentially, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. A process can be terminated when its operation is completed, but it may also have additional steps not included in the drawings. A process can correspond to a method, function, procedure, subroutine, subroutine, etc.

[0012] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0013] In the context of rapid dissemination and precise targeting of digital advertising content, the copyright protection of advertising materials is facing unprecedented challenges. With the diversification of advertising channels and the increasing frequency of content production, the unauthorized copying, modification, or reuse of original materials is becoming increasingly serious, directly threatening the intellectual property security of advertisers and creators, disrupting market order, and weakening the exclusivity and credibility of brand value. In actual operation, infringement often exists covertly in the form of material recombination, minor modifications, and migration between platforms. Traditional detection mechanisms relying on manual review are insufficient to meet the current needs for large-scale, real-time, and high-precision infringement identification. Existing advertising infringement detection systems generally suffer from lengthy detection processes, limited material coverage, subjective fluctuations in judgment standards, and delayed response times. This leads to a heavy reliance on human experience for infringement identification, long detection cycles, and a lack of unified evaluation mechanisms, failing to effectively support rapid review of cross-platform, high-frequency, and diverse materials. Especially with the explosive growth in the number of materials and the increasing prevalence of AI-generated content, the technical shortcomings of traditional models in terms of accurate identification and timely response are becoming increasingly apparent, hindering the intelligent evolution of the copyright protection system.

[0014] Therefore, there is an urgent need to build an advertising creative infringement identification system with automated processing, intelligent analysis, and high-accuracy judgment capabilities. This system should introduce a large-scale advertising data collection mechanism, construct a unified visual and semantic feature comparison model, and combine AI-driven similarity calculation and confidence assessment logic to achieve comprehensive coverage, rapid identification, and dynamic tracking of cross-platform advertising creatives. This system will not only significantly improve the efficiency and accuracy of infringement detection, reduce labor costs and omission rates, but also provide copyright holders and platforms with full-process visualized rights protection support through real-time feedback and automatic reporting mechanisms, comprehensively promoting the compliant operation and healthy development of the digital advertising content ecosystem.

[0015] To address the aforementioned issues, this embodiment provides a method for detecting advertising material infringement. By constructing a dual-feature modeling mechanism for advertising materials, encompassing both visual and semantic features, it achieves automated screening and accurate infringement judgment of large-scale candidate advertising materials, thereby building an efficient and intelligent advertising infringement identification system. This method supports acquiring collected advertising placement data and copyright material data used for infringement screening. The advertising placement data includes multiple candidate advertising materials to be compared. First, corresponding visual feature information is extracted from both candidate and copyright materials. A deep learning-based similarity calculation model is used to determine the visual matching degree between candidate and copyright materials, and based on a preset similarity threshold, candidate materials suspected of having infringement risks are automatically filtered out. Subsequently, semantic features are further extracted from the suspected infringing materials. Combined with the semantic expression of the copyright materials, a semantic-level content alignment mechanism is constructed. By fusing and analyzing the consistency between candidate and copyright materials in the semantic space, the infringement confidence level is assessed, and intelligent identification of confirmed infringing materials is completed based on the confidence threshold. This method enables a two-way review path from visual appearance to content semantics, significantly enhancing the accuracy and comprehensiveness of advertising infringement detection and replacing the inefficient process of traditional manual review. This method not only improves the automation and scalability of infringement identification, but also has good cross-platform adaptability and model scalability. It can be widely applied to various advertising channels and content creation scenarios, and comprehensively promote the intelligent upgrading and effectiveness improvement of the copyright protection system in the digital advertising field.

[0016] The advertising material infringement detection method provided in this embodiment can be executed by an advertising material infringement detection device. This device can be implemented through software and / or hardware, and can consist of two or more physical entities, or a single physical entity. For example, the advertising material infringement detection device could be an operations and maintenance server used to maintain the normal operation of business.

[0017] The advertising material infringement detection device is equipped with at least one type of operating system, including but not limited to Android, Linux, and Windows. The device can install at least one application based on the operating system; this application can be a built-in application of the operating system or an application downloaded from a third-party device or server. In this embodiment, the advertising material infringement detection device has at least one application capable of executing the advertising material infringement detection method.

[0018] For ease of understanding, this embodiment uses an operation and maintenance server as the subject of the advertising material infringement detection method for example.

[0019] Figure 1 A flowchart of an advertising material infringement detection method provided in an embodiment of this application is given. (Reference) Figure 1 The specific methods for detecting copyright infringement in this advertising material include: S110. Obtain the collected advertising placement data and advertising copyright materials used for infringement investigation, wherein the advertising placement data includes several candidate advertising materials.

[0020] In some embodiments, the process first involves acquiring collected advertising delivery data and copyrighted advertising materials used for infringement investigation. Advertising delivery data refers to advertising-related information obtained from various advertising platforms, used to extract content to be detected. Copyrighted advertising materials refer to original advertising materials used for comparison and determination of potential infringement. The advertising delivery data includes several candidate advertising materials, which are advertising content fragments that have not yet been confirmed as infringing but require further analysis.

[0021] In one embodiment, the method for collecting advertising delivery data may be: periodically obtaining structured or unstructured advertising display content from various advertising platforms through web crawlers or platform interfaces deployed in the system, including images, videos, text and their corresponding delivery metadata.

[0022] In one embodiment, the way to obtain advertising copyright materials may be: by providing advertising source files with confirmed ownership from the original content creator, advertiser, or platform, or by obtaining them from a standardized material library registered by the rights holder in the copyright database.

[0023] In one embodiment, the identification method for advertising candidate materials may be: filtering out advertising segments containing multimedia content and possessing infringement risk characteristics from the collected advertising delivery data, and using them as target objects for subsequent comparison and analysis.

[0024] Optionally, Figure 2 A flowchart illustrating the advertising data collection frequency control provided in an embodiment of this application is given. (Reference) Figure 2 The specific methods for controlling the frequency of advertising data collection include: S1101. Obtain the number of historical infringing materials on the advertising platform, and generate an infringement probability index based on the number of historical infringing materials.

[0025] For example, the first step is to obtain the number of historical infringing materials from the advertising platform. Here, the advertising platform refers to an online platform that provides advertising services, and the number of historical infringing materials refers to the cumulative number of advertising materials identified as infringing on the platform within a specified time period, reflecting the platform's risk level regarding content compliance.

[0026] After obtaining the number of historical infringing materials, a corresponding infringement probability index is generated based on this number. The infringement probability index is a quantitative indicator that represents the likelihood of current or future infringement on an advertising platform. It is typically used to assess the platform's content risk level and provide a basis for subsequent content review strategies, platform classification management, or risk control resource allocation.

[0027] In one embodiment, the method for obtaining the number of historical infringing materials can be: retrieving the historical judgment result logs from the infringement detection system, counting the material entries that were judged to be confirmed as infringing from the material records of the same advertising platform, and obtaining the number of times the platform infringed within the specified statistical window.

[0028] In one embodiment, the infringement probability index can be generated by: calculating the infringement rate based on the number of historical infringing materials and the total number of materials on the platform, and mapping the infringement rate to a preset index model to output a standardized infringement probability index.

[0029] S1102. Generate an advertising collection frequency based on the infringement probability index, and collect advertising data from the advertising platform at the advertising collection frequency.

[0030] For example, the ad collection frequency is generated based on the aforementioned infringement probability index. The ad collection frequency refers to the time interval or trigger frequency at which the system performs ad data collection operations on a specific ad platform, used to control the execution density of collection tasks. A higher infringement probability index indicates a higher risk of infringement on the platform, and the corresponding ad collection frequency should be increased accordingly to strengthen content monitoring of the platform.

[0031] After generating the ad collection frequency, ad delivery data is collected from the target ad delivery platform according to this frequency. Ad delivery data refers to the ad content, display records, and related metadata collected from the ad platform within a specified time period, which is used in subsequent infringement analysis and processing procedures.

[0032] In one embodiment, the ad collection frequency can be generated by inputting an infringement probability index into a preset frequency mapping function. This function can be a linear function, a step function, or an exponential function with upper and lower limits, and outputting a collection trigger frequency proportional to the exponent. For example, when the infringement probability index exceeds a set threshold, the collection frequency can be increased from once a day to once an hour.

[0033] In one embodiment, the data collection method for ad delivery can be: by using a data access interface established with the ad platform, requesting the latest ad content, including the ad creative itself, delivery time, display location, audience attributes, and other information, as the input data source for subsequent infringement detection.

[0034] Optionally, Figure 3 A flowchart illustrating the process of confirming advertising candidate materials according to an embodiment of this application is provided. (See reference...) Figure 3 The specific methods for confirming candidate advertising creatives include: S111. Extract the advertising materials from the advertising data, obtain the dissemination volume of the advertising materials, and generate a dissemination popularity index based on the dissemination volume.

[0035] For example, the first step is to extract advertising creatives from the advertising data. Advertising creatives refer to the core multimedia elements that constitute the advertising content, such as images, videos, or text content, used to represent the visual or informational presentation of the advertisement.

[0036] After obtaining the advertising creative, the next step is to obtain the reach of that creative. Reach refers to the number of times the advertising creative has been displayed, clicked, shared, or otherwise interacted with across different channels, reflecting the creative's dissemination scope and audience engagement.

[0037] Based on the reach of the advertising material, a corresponding popularity index is generated. The popularity index indicates the current level of activity in the dissemination of the advertising material and is an important indicator for measuring the material's influence and dissemination effectiveness.

[0038] In one embodiment, the method for extracting advertising materials can be: parsing the multimedia fields in the advertising data to identify and separate specific images, videos, and text content.

[0039] In one embodiment, the way to obtain the reach of the creative material can be by collecting data such as the exposure, click-through rate, and number of shares of the creative material through an advertising platform or a third-party data monitoring interface.

[0040] In one embodiment, the dissemination popularity index can be generated by normalizing the collected material dissemination data and combining it with historical dissemination trends, and then calculating the dissemination popularity index using a weighted algorithm or machine learning model.

[0041] S112. Based on the infringement probability index of the advertising platform corresponding to the advertising material and the dissemination popularity index corresponding to the advertising material, determine the material weight index corresponding to the advertising material.

[0042] For example, based on the infringement probability index of the advertising platform corresponding to the advertising creative and the dissemination popularity index corresponding to the advertising creative, the creative weight index is determined. Among them, the infringement probability index indicates the likelihood of infringement on the advertising platform, the dissemination popularity index reflects the dissemination activity of the advertising creative, and the creative weight index combines these two indicators to measure the priority and attention of the advertising creative in infringement risk management.

[0043] In one embodiment, the infringement probability index can be obtained by extracting and calculating the historical infringement statistics of the target advertising platform from the infringement risk assessment module.

[0044] In one embodiment, the method for obtaining the popularity index can be: based on the popularity indicators such as the exposure and clicks of advertising materials, a popularity score can be generated through normalization and model calculation.

[0045] In one embodiment, the method for determining the material weight index can be: using a weighted algorithm, linearly combining the infringement probability index and the dissemination popularity index according to predetermined weights or non-linearly fusing them to obtain a weight value that comprehensively reflects the infringement risk and dissemination influence, which is used to guide subsequent adjustments to the collection frequency or key monitoring.

[0046] S113. Ad campaign materials whose material weight index is greater than the weight index threshold are identified as candidate ad campaign materials.

[0047] For example, ad creatives with a weight index greater than a weight index threshold are identified as candidate ad creatives. The weight index is a metric derived by comprehensively considering the ad platform's infringement probability index and the ad creative's popularity index, reflecting the infringement risk and dissemination impact of the creative; the weight index threshold is a preset standard value used to screen high-risk or high-attention creatives.

[0048] In one embodiment, the weight index threshold can be set by determining one or more grading thresholds based on historical data statistical analysis and risk control strategies to distinguish advertising materials with different risk levels.

[0049] In one embodiment, the candidate advertising creatives can be determined by comparing the weight index of each advertising creative with a threshold, marking creatives that exceed the threshold as candidate creatives, and then proceeding to the subsequent infringement detection and review process.

[0050] In one embodiment, the weight index threshold can be dynamically adjusted based on different advertising platforms, creative types, or time periods to achieve more accurate and flexible creative selection.

[0051] S120. Determine the visual features of the candidate material corresponding to each of the advertising candidate materials and the visual features of the copyright material corresponding to each of the advertising copyright materials, calculate the visual feature similarity between the visual features of each candidate material and the visual features of the copyright material, and determine the advertising candidate materials whose visual feature similarity is greater than the similarity threshold as suspected infringing materials.

[0052] In some embodiments, candidate material visual features corresponding to each advertising candidate material and copyright material visual features corresponding to each advertising copyright material are determined. Candidate material visual features refer to feature vectors or embedded representations extracted from advertising candidate materials through image processing or multimodal feature extraction algorithms to represent their visual content; copyright material visual features refer to standardized visual feature representations generated based on copyright materials, used for subsequent similarity calculations and matching comparisons.

[0053] In one embodiment, the method of extracting visual features may include: using a convolutional neural network, a visual Transformer, or other pre-trained deep learning models to encode image frames or video keyframes in advertising materials to obtain feature vectors that represent multi-dimensional visual characteristics such as color distribution, structural texture, and spatial layout.

[0054] In one embodiment, the method for calculating the visual feature similarity between the visual features of candidate materials and the visual features of copyrighted materials may include: using cosine similarity, Euclidean distance, contrastive learning model or dual-tower structure model to perform matching analysis on the feature representations of the two materials and quantify the degree of similarity of their visual content.

[0055] In one embodiment, if the visual features of a candidate material are more similar to the visual features of any copyrighted material than a preset similarity threshold, the candidate material is identified as a suspected infringing material for further verification and processing.

[0056] Optionally, Figure 4 A flowchart for visual feature extraction of materials provided in an embodiment of this application is given. (Reference) Figure 4 The specific methods for extracting visual features from this material include: S1201. Extract several candidate material keyframes and several copyright material keyframes from the candidate advertising material and the copyright advertising material, respectively.

[0057] For example, several keyframes are extracted from both candidate advertising materials and copyrighted advertising materials. Candidate advertising materials refer to advertising content selected through a weighted index that has a high risk of infringement or high dissemination popularity, while copyrighted advertising materials refer to standard advertising materials with legal ownership used for comparison.

[0058] In one embodiment, keyframes refer to several static frames in video footage that can represent the overall content characteristics. Frames with rich information and obvious visual differences are usually selected to reduce the amount of subsequent comparison calculations and improve matching efficiency.

[0059] In one embodiment, keyframes can be extracted by: selecting video frames with significant scene changes or prominent visual features based on inter-frame difference analysis; or by using a pre-trained deep learning model to automatically identify key content frames.

[0060] In one embodiment, the number of keyframes and the extraction strategy can be dynamically adjusted based on factors such as the length of the source material, the complexity of the content, and computing resources, so as to ensure that the keyframes can effectively represent the overall content of the source material.

[0061] Optionally, Figure 5 A flowchart for extracting keyframes of copyrighted material according to an embodiment of this application is provided. (Reference) Figure 5 The keyframe extraction method for this copyrighted material specifically includes: S12011. Calculate the histogram similarity of the color histograms between adjacent frames of the advertising copyright material, and determine the shot switching point of the advertising copyright material based on the histogram similarity.

[0062] For example, the color histogram similarity between adjacent frames of advertising copyright material is calculated. Here, color histogram similarity refers to an indicator that quantifies the degree of visual content variation by comparing the color distribution characteristics of consecutive frames, and is used to reflect the coherence of the image content.

[0063] In one embodiment, the shot transition points of the advertising copyright material are determined based on the calculated color histogram similarity. A shot transition point refers to a location in the video where the scene or image undergoes a significant change, typically manifested as a moment when the color histogram similarity between adjacent frames decreases significantly, representing a switch or segmentation of video content.

[0064] In one embodiment, the color histogram similarity can be calculated by using algorithms such as cosine similarity, Bach distance, chi-square distance, or cross-entropy to measure the color histograms of consecutive frames.

[0065] In one embodiment, the method for determining the shot switching point can be: setting a similarity threshold, and when the color histogram similarity of adjacent frames is lower than the threshold, marking the frame as a shot switching point for subsequent keyframe extraction and video segmentation processing.

[0066] S12012. Based on the shot switching point, the advertising copyright material is divided into several continuous shot segments, and several copyright material keyframes of the advertising copyright material are extracted from each of the continuous shot segments.

[0067] For example, advertising copyright materials are divided into several consecutive shot segments based on shot transition points. Shot transition points indicate locations where the video content undergoes significant changes; these points divide the video into multiple logically continuous and relatively consistent shot segments.

[0068] In one embodiment, several copyright material keyframes are extracted from the advertising copyright material within each consecutive shot segment. These copyright material keyframes are static frames that represent the main visual content of the shot segment and are used for subsequent feature extraction and similarity matching.

[0069] In one embodiment, the specific method of shot segmentation may be: after identifying the shot switching point according to a preset threshold, the video frame is divided into multiple consecutive frame sequences according to the position of the switching point, and each sequence corresponds to a shot segment.

[0070] In one embodiment, the extraction method of keyframes of copyrighted material may include: selecting several frames with the greatest inter-frame differences within a shot segment, or determining keyframes through methods such as uniform sampling and scene semantic analysis, so as to ensure that the keyframes can accurately reflect the visual features of the shot segment.

[0071] S1202. Extract the visual features of the copyright material corresponding to the advertising copyright material based on the keyframes of the copyright material, perform feature weighting on the keyframes of the candidate material based on the visual features of the copyright material, and extract the visual features of the candidate material corresponding to the advertising candidate material based on the keyframes of the candidate material.

[0072] For example, the visual features of the copyrighted material corresponding to the advertising copyrighted material are extracted based on the keyframes of the copyrighted material. Here, the visual features of the copyrighted material refer to the feature vectors extracted from the keyframes of the copyrighted material through image processing or deep learning models, used to characterize its visual content.

[0073] In one embodiment, keyframes of candidate materials are weighted based on the visual features of the copyrighted material. Feature weighting refers to assigning different weights to the visual features of different keyframes, emphasizing keyframes with a high degree of similarity to the visual features of the copyrighted material, thereby improving the overall accuracy of feature representation.

[0074] In one embodiment, visual features of candidate ad creatives are extracted based on weighted keyframe features. These visual features are obtained by integrating weighted keyframe visual features and are used for subsequent similarity calculations and infringement analysis.

[0075] In one embodiment, the visual features of copyrighted material can be extracted by encoding keyframe images using pre-trained models such as convolutional neural networks and visual Transformers to obtain high-dimensional visual feature vectors.

[0076] In one embodiment, feature weighting can be achieved by assigning weights based on the similarity scores between keyframes and copyright material keyframes, or by dynamically adjusting the contribution of different keyframes through an attention mechanism.

[0077] In one embodiment, the extraction of visual features of candidate materials can be achieved through weighted summation, pooling operations, or feature fusion networks to obtain comprehensive visual features representing the entire candidate material.

[0078] S130. Extract the semantic features of candidate materials corresponding to each of the suspected infringing materials, and the semantic features of copyright materials of each of the advertising copyright materials. Determine the infringement confidence level based on the semantic features of each candidate material and the semantic features of the copyright materials. Identify the suspected infringing materials whose infringement confidence level is greater than the confidence level threshold as confirmed infringing materials.

[0079] In some embodiments, the semantic features of candidate materials corresponding to each suspected infringing material, and the semantic features of copyrighted materials for each advertising copyrighted material are extracted. The semantic features of candidate materials refer to the information expression form used to characterize the semantic meaning of the advertising content, and are typically extracted from the text content, voice description, or visual tags of the materials through natural language processing or multimodal fusion methods. The semantic features of copyrighted materials represent the semantic features of the copyrighted materials, serving as a reference object for semantic comparison.

[0080] In one embodiment, the extraction of semantic features may include: using a pre-trained language model to embed text content, image descriptions, or audio transcriptions in advertising materials to obtain multi-dimensional semantic vectors, which are used to measure the semantic relevance between materials.

[0081] In one embodiment, the infringement confidence score for each suspected infringing material is determined based on the comparison between the semantic features of the candidate material and the semantic features of the copyrighted material. The infringement confidence score indicates the degree of semantic matching between the candidate material and the copyrighted material. It can be calculated using semantic vector similarity scoring, probability values ​​output by a multimodal fusion classifier, or by estimating the matching confidence score through a trained confidence model.

[0082] In one embodiment, when the infringement confidence level of a suspected infringing material exceeds a preset confidence level threshold, the material is identified as confirmed infringing material and enters the final infringement marking and handling process.

[0083] Optionally, Figure 6 A flowchart for the semantic feature extraction of candidate materials provided in an embodiment of this application is given. (Reference) Figure 6 The specific methods for extracting semantic features from candidate materials include: S1301. Extract the text content, audio content, and object category of the suspected infringing material.

[0084] For example, the text content, audio content, and object categories of the suspected infringing material are extracted first. The text content refers to the text information contained in the material, such as text overlays in images or videos, subtitles, etc.; the audio content refers to the audio segments or audio descriptions contained in the material; and the object categories refer to the main visual object types contained in the material, such as people, scenes, iconic items, etc.

[0085] In one embodiment, the method for extracting the text content of the source material can be: recognizing embedded text information in an image or video frame using OCR technology.

[0086] In one embodiment, the method for extracting the audio content of the material can be: using a speech recognition algorithm to transcribe the speech signal in the audio track to generate corresponding text or semantic tags.

[0087] In one embodiment, the method for extracting object categories from source material can be: using an image recognition model to analyze keyframes in the source material, identify and classify the main target objects within them.

[0088] S1302. Based on the text content of the material, the audio content of the material, and the object category of the material, determine the text semantic features, audio semantic features, and visual semantic features corresponding to the suspected infringing material, respectively.

[0089] For example, based on the text content, audio content, and object category of the material, the text semantic features, audio semantic features, and visual semantic features corresponding to the suspected infringing material are determined respectively. Among them, the text semantic features refer to the embedded vectors or label representations that characterize the semantic information conveyed by the text content in the material; the audio semantic features refer to the semantic expression results obtained through audio content analysis, which are used to reflect the actual meaning and context of the audio segment; and the visual semantic features refer to the image semantic representations generated based on the object categories and their spatial structures identified in the material, which are used to depict the visual meaning and context of the material.

[0090] In one embodiment, the method for determining the semantic features of the text may be: segmenting and encoding the extracted source text content, and inputting it into a pre-trained language model to generate a text embedding representation.

[0091] In one embodiment, the way to determine speech semantic features is to convert the speech content into text and then extract deep semantic features by combining emotion recognition and semantic understanding models.

[0092] In one embodiment, the visual semantic features can be determined by identifying the core objects in an image based on object detection and image classification models, and extracting their semantic vectors through a multi-layer convolutional neural network.

[0093] S1303. Generate candidate material semantic features for the suspected infringing material based on the text semantic features, the speech semantic features and the visual semantic features.

[0094] For example, based on textual semantic features, speech semantic features, and visual semantic features, candidate semantic features for suspected infringing materials are generated. Here, candidate semantic features refer to the comprehensive semantic representation formed by fusing multimodal semantic information, used to describe the overall content features of the suspected infringing materials. Textual semantic features refer to the vector representation characterizing the semantics of the text content in the material; speech semantic features refer to the semantic vector reflecting the meaning of the speech content in the material; and visual semantic features refer to the embedded representation describing the semantics of visual elements in the image.

[0095] In one embodiment, the way to generate semantic features of candidate materials can be: inputting textual semantic features, speech semantic features and visual semantic features into a unified multimodal fusion network, and fusing them into a unified semantic representation through an attention mechanism or feature splicing method.

[0096] In one embodiment, the way to fuse multimodal semantic features can be by using a cross-modal alignment model to map the feature space, ensuring that each modal feature has consistent representation capabilities in the same semantic space.

[0097] Optionally, Figure 7 A flowchart illustrating the infringing material confirmation warning provided in an embodiment of this application is given. (See reference...) Figure 7 The method for confirming and issuing early warnings of infringing materials specifically includes: S140. Obtain the advertising platform and advertising copyright material corresponding to the confirmed infringing material, and generate an infringement behavior analysis report based on the advertising platform, the confirmed infringing material and the advertising copyright material.

[0098] For example, the process first involves acquiring the advertising platform and copyrighted advertising materials corresponding to the confirmed infringing materials. The confirmed infringing materials refer to advertising materials identified as infringing after semantic comparison and rule-based judgment. The advertising platform refers to the platform environment in which the materials are actually displayed or distributed, such as social media, search engines, or video platforms. The copyrighted advertising materials refer to the original, copyrighted advertising content, used as a basis for comparison and tracing. Subsequently, based on the advertising platform, the confirmed infringing materials, and the copyrighted advertising materials, an infringement analysis report is generated. This infringement analysis report is a structured document used to summarize, explain, and quantify the facts, pathways, and scope of impact of the infringement.

[0099] In one embodiment, the way to obtain the advertising platform may be by parsing and confirming the source field of the infringing material or by using a source tracing tag to locate the platform to which it belongs.

[0100] In one embodiment, the method for obtaining advertising copyright materials may be: by matching and confirming the semantic features of infringing materials, searching a pre-stored copyright material library, and obtaining the original material with the highest similarity.

[0101] In one embodiment, the infringement analysis report can be generated by automatically generating a report document based on the results of material comparison, the attributes of the advertising platform, and historical behavior records. The report document includes evidence of infringement, impact assessment, liability recommendations, and other content.

[0102] S150. Obtain the target terminal corresponding to the advertising copyright material, and send the infringement analysis report to the target terminal.

[0103] For example, the first step is to obtain the target terminal corresponding to the advertising copyright material. The target terminal refers to the receiving device or system node related to the ownership of the advertising copyright material, such as the copyright holder's management backend, legal compliance processing system, or content supervision platform. Subsequently, an infringement analysis report is sent to the target terminal. This infringement analysis report refers to a structured analysis result generated based on confirmed infringing materials, the advertising platform, and the copyright material, used to support rights protection, penalties, or filing.

[0104] In one embodiment, the target terminal can be obtained by querying the associated receiving system address or account identifier based on the registration information or distribution record of the advertising copyright material.

[0105] In one embodiment, the infringement analysis report can be sent by pushing the generated report to the target terminal's communication interface via a secure transmission protocol, ensuring data integrity and privacy protection.

[0106] Optionally, Figure 8 A flowchart illustrating the steps for determining infringing material according to an embodiment of this application is provided. (Refer to...) Figure 8The system automatically collects advertising data from multiple advertising platforms, relying on platform APIs for automated data collection, achieving scheduled synchronization and incremental updates to ensure data integrity and timeliness. The collected advertising data undergoes structured analysis to extract candidate advertising materials, including videos, images, and their associated text content. Data format parsing and content recognition algorithms, such as HTML structure parsing and video link extraction, are used to extract and archive these materials. A video similarity analysis algorithm is used to compare candidate advertising materials with copyrighted advertising materials extracted from original copyrighted videos, calculating their similarity scores. A deep learning-based video feature extraction method, combined with frame-level similarity calculation and content-level matching, sets a reasonable similarity threshold for initial screening. Based on the similarity analysis results, suspected infringing materials are identified and marked. Dynamic or fixed similarity thresholds are used for judgment; materials exceeding the threshold are classified as suspected infringing, and identification records are generated based on timestamps and contextual information. The identified suspected infringing materials are further input into an AI large-scale model for multi-dimensional analysis of semantics, structure, and style to determine whether they constitute substantial infringement. Multimodal deep learning models, such as CLIP, ViT, and multimodal BERT, are introduced to compare the multi-layered features of copyrighted advertising materials and suspected infringing materials, enabling fine-grained identification of infringement and analysis of behavioral patterns, thereby confirming infringing materials. Based on the AI ​​analysis results, a standardized infringement analysis report is automatically generated, including the basis for infringement determination, a summary of material comparison, and risk level. The analysis report is then pushed to the legal team through the company's internal collaboration system, where they verify the infringement and decide on subsequent rights protection or handling actions.

[0107] Based on the above embodiments, Figure 9 A schematic diagram of the advertising material infringement detection device provided in this application embodiment. (Reference) Figure 9 The advertising material infringement detection device provided in this embodiment specifically includes: an acquisition module 21, a filtering module 22, and a judgment module 23.

[0108] The acquisition module 21 is configured to acquire collected advertising placement data and advertising copyright materials used for infringement investigation, wherein the advertising placement data includes several advertising candidate materials; the screening module 22 is configured to determine the visual features of the candidate materials corresponding to each of the advertising candidate materials and the visual features of the copyright materials corresponding to each of the advertising copyright materials, calculate the visual feature similarity between the visual features of each candidate material and the visual features of the copyright materials, and identify the advertising candidate materials whose visual feature similarity is greater than a similarity threshold as suspected infringing materials; the determination module 23 is configured to extract the semantic features of the candidate materials corresponding to each of the suspected infringing materials and the semantic features of the copyright materials of each of the advertising copyright materials, determine the infringement confidence level based on the semantic features of each candidate material and the semantic features of the copyright materials, and identify the suspected infringing materials whose infringement confidence level is greater than a confidence threshold as confirmed infringing materials.

[0109] Based on the above embodiments, the acquisition module 21 includes: an infringement probability unit, configured to acquire the number of historical infringing materials on the advertising platform and generate an infringement probability index based on the number of historical infringing materials; and a collection frequency unit, configured to generate an advertising collection frequency based on the infringement probability index and collect advertising data from the advertising platform at the advertising collection frequency.

[0110] Based on the above embodiments, the advertising material infringement detection device further includes: a material delivery module configured to extract advertising materials from the advertising delivery data; a popularity index module configured to obtain the dissemination volume of the advertising materials and generate a dissemination popularity index based on the dissemination volume; a material weight module configured to determine the material weight index corresponding to the advertising materials based on the infringement probability index of the advertising platform corresponding to the advertising materials and the dissemination popularity index corresponding to the advertising materials; and a candidate material module configured to identify advertising materials whose material weight index is greater than the weight index threshold as advertising candidate materials.

[0111] Based on the above embodiments, the screening module 22 includes: a keyframe extraction unit, configured to extract a number of candidate material keyframes and a number of copyright material keyframes from the advertising candidate material and the advertising copyright material, respectively; and a visual feature extraction unit, configured to extract the copyright material visual features corresponding to the advertising copyright material based on the copyright material keyframes, perform feature weighting on the candidate material keyframes based on the copyright material visual features, and extract the candidate material visual features corresponding to the advertising candidate material based on the candidate material keyframes.

[0112] Based on the above embodiments, the keyframe extraction unit includes: a shot switching point subunit, configured to calculate the histogram similarity of color histograms between adjacent frames of the advertising copyright material, and determine the shot switching point of the advertising copyright material based on the histogram similarity; and a keyframe extraction subunit, configured to divide the advertising copyright material into several continuous shot segments based on the shot switching points, and extract several copyright material keyframes of the advertising copyright material in each of the continuous shot segments.

[0113] Based on the above embodiments, the determination module 23 includes: a text data extraction unit configured to extract the text content, audio content, and object category of the suspected infringing material; a text feature extraction unit configured to determine the text semantic features, audio semantic features, and visual semantic features corresponding to the suspected infringing material based on the text content, audio content, and object category; and a text feature synthesis unit configured to generate candidate material semantic features of the suspected infringing material based on the text semantic features, audio semantic features, and visual semantic features.

[0114] Based on the above embodiments, the advertising material infringement detection device further includes: a reporting module, configured to acquire the advertising platform and advertising copyright material corresponding to the confirmed infringing material, and generate an infringement behavior analysis report based on the advertising platform, the confirmed infringing material, and the advertising copyright material; and a sending module, configured to acquire the target terminal corresponding to the advertising copyright material and send the infringement behavior analysis report to the target terminal.

[0115] The advertising material infringement detection device provided in this application embodiment, through the construction of a structured component system consisting of multiple functional modules such as an acquisition module, a filtering module, and a judgment module, achieves complete process control for intelligent collection of advertising data, material similarity analysis, intelligent judgment of infringement behavior, and push of analysis reports. Based on access to multiple advertising platforms, the device can automatically extract historical infringement records and material dissemination data, dynamically adjust the collection frequency and material filtering strategy, thereby improving the accuracy and efficiency of identifying suspected infringing materials. The built-in acquisition module is responsible for collecting advertising data and copyrighted advertising materials from advertising platforms. The acquisition module integrates an infringement probability unit and a collection frequency unit, which can dynamically generate a platform-level infringement probability index based on the number of historical infringing materials, and adjust the advertising data collection frequency accordingly to achieve data-oriented collection from key platforms. In addition, the device also includes a material delivery module, a popularity index module, a material weight module, and a candidate material module, constructing a material weight filtering process. The device can calculate a dissemination popularity index based on the dissemination volume of advertising materials, and generate a material weight index in combination with the infringement probability index of the platform, thereby filtering out candidate advertising materials with high infringement risk and high dissemination influence. The module integrates a keyframe extraction unit and a visual feature extraction unit. The keyframe extraction unit analyzes shot transition points between copyrighted material frames based on color histogram similarity and extracts keyframes from shot segments to improve the representativeness of visual features. The visual feature extraction unit further extracts visual features from the copyrighted material's keyframes and weights the keyframes of candidate materials to generate candidate material visual features. The screening module calculates similarity values ​​between the candidate and copyrighted material visual features and identifies candidate materials exceeding a preset similarity threshold as suspected infringing materials. Based on this, the judgment module further analyzes suspected infringing materials by constructing a multimodal semantic recognition structure. The judgment module includes a text data extraction unit, a text feature extraction unit, and a text feature synthesis unit. The text data extraction unit extracts text content, audio content, and object category information from the material; the text feature extraction unit extracts text semantic features, audio semantic features, and visual semantic features, respectively. The text feature synthesis unit synthesizes these features to generate candidate material semantic features, matches them with the copyrighted material semantic features, calculates the infringement confidence score, and filters out confirmed infringing materials based on the confidence score threshold. To achieve closed-loop processing, the device further integrates a reporting module and a sending module. The reporting module acquires and confirms the platforms where infringing materials were distributed and related copyrighted materials, generating an infringement analysis report that includes material comparison information, similarity analysis results, semantic judgment criteria, and platform information. The sending module, based on preset target terminal configurations, pushes the analysis report to the corresponding legal personnel or processing systems, achieving automated legal process integration.Through its modular structure and intelligent judgment mechanism, this device breaks through the traditional infringement detection methods that rely on manual screening and rule matching, establishing a highly efficient detection system that integrates dynamic data collection, multi-dimensional feature analysis, and model-driven judgment. This system not only significantly improves the accuracy and real-time performance of infringement identification but also possesses strong adaptability and scalability, providing stable and intelligent technical support for large-scale advertising supervision and copyright protection.

[0116] The advertising material infringement detection device provided in this application embodiment can be used to execute the advertising material infringement detection method provided in the above embodiment, and has corresponding functions and beneficial effects.

[0117] Figure 10 This is a schematic diagram of the structure of an advertising material infringement detection device provided in an embodiment of this application, for reference. Figure 10 The advertising material infringement detection device includes: a processor 31, a memory 32, a communication device 33, an input device 34, and an output device 35. The number of processors 31 and the number of memories 32 in the advertising material infringement detection device can be one or more. The processor 31, memory 32, communication device 33, input device 34, and output device 35 of the advertising material infringement detection device can be connected via a bus or other means.

[0118] The memory 32, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as program instructions / modules corresponding to the advertising material infringement detection method in any embodiment of this application (e.g., the acquisition module 21, filtering module 22, and determination module 23 in the advertising material infringement detection device). The memory 32 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on the use of the device, etc. Furthermore, the memory 32 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0119] The communication device 33 is used for data transmission.

[0120] The processor 31 executes various functional applications and data processing of the device by running software programs, instructions and modules stored in the memory 32, thereby realizing the above-mentioned advertising material infringement detection method.

[0121] Input device 34 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the device. Output device 35 may include display devices such as a display screen.

[0122] The advertising material infringement detection equipment provided above can be used to execute the advertising material infringement detection method provided in the above embodiments, and has corresponding functions and beneficial effects.

[0123] This application embodiment also provides a storage medium containing computer-executable instructions. When executed by a computer processor, the computer-executable instructions are used to perform an advertising material infringement detection method. The advertising material infringement detection method includes: acquiring collected advertising placement data and advertising copyright materials for infringement investigation, wherein the advertising placement data includes a plurality of advertising candidate materials; determining the visual features of the candidate material corresponding to each of the advertising candidate materials and the visual features of the copyright material corresponding to each of the advertising copyright materials; calculating the visual feature similarity between the visual features of each candidate material and the visual features of the copyright material; identifying advertising candidate materials whose visual feature similarity is greater than a similarity threshold as suspected infringing materials; extracting the semantic features of the candidate material corresponding to each suspected infringing material and the semantic features of the copyright material of each of the advertising copyright materials; determining the infringement confidence level based on the semantic features of each candidate material and the semantic features of the copyright material; and identifying suspected infringing materials whose infringement confidence level is greater than a confidence threshold as confirmed infringing materials.

[0124] Storage medium – any type of memory device or storage device. The term “storage medium” is intended to include: mounting media, such as CD-ROM, floppy disk, or magnetic tape devices; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (e.g., hard disk or optical storage); registers or other similar types of memory elements, etc. Storage medium may also include other types of memory or combinations thereof. Furthermore, storage medium may reside in a first computer system in which a program is executed, or it may reside in a different second computer system connected to the first computer system via a network (such as the Internet). The second computer system can provide program instructions to the first computer for execution. The term “storage medium” can include two or more storage media residing in different locations (e.g., in different computer systems connected via a network). Storage medium may store program instructions (e.g., specifically implemented as a computer program) executable by one or more processors.

[0125] Of course, the computer-executable instructions provided in the embodiments of this application are not limited to the above-mentioned advertising material infringement detection method, but can also perform related operations in the advertising material infringement detection method provided in any embodiment of this application.

[0126] The advertising material infringement detection device, storage medium, and advertising material infringement detection equipment provided in the above embodiments can execute the advertising material infringement detection method provided in any embodiment of this application. For technical details not described in detail in the above embodiments, please refer to the advertising material infringement detection method provided in any embodiment of this application.

[0127] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of this application. The scope of this application is determined by the scope of the claims.

Claims

1. A method for detecting copyright infringement in advertising materials, characterized in that, include: Acquire the collected advertising placement data and the advertising copyright materials used for infringement investigation, wherein the advertising placement data includes several candidate advertising materials; Determine the visual features of the candidate material corresponding to each of the advertising candidate materials, and the visual features of the copyright material corresponding to each of the advertising copyright materials. Calculate the visual feature similarity between each of the candidate material visual features and the copyright material visual features. Identify the advertising candidate materials whose visual feature similarity is greater than a similarity threshold as suspected infringing materials. Extract the semantic features of candidate materials corresponding to each of the suspected infringing materials, as well as the semantic features of copyright materials for each of the advertising copyright materials. Determine the infringement confidence level based on the semantic features of each candidate material and the semantic features of the copyright materials. Identify the suspected infringing materials whose infringement confidence level is greater than the confidence level threshold as confirmed infringing materials.

2. The advertising material infringement detection method according to claim 1, characterized in that, The acquisition of collected advertising delivery data includes: Obtain the number of historical infringing materials from the advertising platform, and generate an infringement probability index based on the number of historical infringing materials; Based on the infringement probability index, an advertising collection frequency is generated, and advertising data is collected from the advertising platform at the advertising collection frequency.

3. The advertising material infringement detection method according to claim 2, characterized in that, After acquiring the collected advertising placement data and the advertising copyright materials used for infringement investigation, the process also includes: Extract advertising creatives from the advertising data, obtain the reach of the advertising creatives, and generate a reach index based on the reach of the advertising creatives; Based on the infringement probability index of the advertising platform corresponding to the advertising material and the dissemination popularity index corresponding to the advertising material, the material weight index corresponding to the advertising material is determined. Ad creatives with a weight index greater than the weight index threshold are identified as candidate ad creatives.

4. The advertising material infringement detection method according to claim 1, characterized in that, The step of determining the visual features of the candidate materials corresponding to each of the advertising candidate materials, and the visual features of the copyright materials corresponding to each of the advertising copyright materials, includes: Extract several candidate material keyframes and several copyright material keyframes from the candidate advertising materials and the copyright advertising materials, respectively. Based on the keyframes of the copyright material, extract the visual features of the copyright material corresponding to the advertising copyright material. Based on the visual features of the copyright material, perform feature weighting on the keyframes of the candidate material. Based on the keyframes of the candidate material, extract the visual features of the candidate material corresponding to the advertising candidate material.

5. The advertising material infringement detection method according to claim 4, characterized in that, The step of extracting several copyright material keyframes from the advertising copyright material includes: Calculate the histogram similarity of the color histograms between adjacent frames of the advertising copyright material, and determine the shot switching point of the advertising copyright material based on the histogram similarity; Based on the shot switching point, the advertising copyright material is divided into several continuous shot segments, and several copyright material keyframes of the advertising copyright material are extracted from each of the continuous shot segments.

6. The advertising material infringement detection method according to claim 1, characterized in that, The step of extracting the semantic features of candidate materials corresponding to each of the suspected infringing materials includes: Extract the text content, audio content, and object categories from the suspected infringing materials; Based on the text content, audio content, and object category of the material, the text semantic features, audio semantic features, and visual semantic features corresponding to the suspected infringing material are determined respectively. Based on the text semantic features, the speech semantic features, and the visual semantic features, candidate material semantic features for the suspected infringing material are generated.

7. The advertising material infringement detection method according to claim 1, characterized in that, After determining the suspected infringing material with an infringement confidence level greater than the confidence threshold as confirmed infringing material, the method further includes: Obtain the advertising platform and copyrighted advertising materials corresponding to the confirmed infringing materials, and generate an infringement analysis report based on the advertising platform, the confirmed infringing materials, and the copyrighted advertising materials; Obtain the target terminal corresponding to the copyrighted advertising material, and send the infringement analysis report to the target terminal.

8. An advertising material infringement detection device, characterized in that, include: The acquisition module is used to acquire collected advertising placement data and advertising copyright materials used for infringement investigation. The advertising placement data includes several candidate advertising materials. The filtering module is used to determine the visual features of the candidate materials corresponding to each of the advertising candidate materials and the visual features of the copyright materials corresponding to each of the advertising copyright materials, calculate the visual feature similarity between each of the candidate material visual features and the copyright material visual features, and identify the advertising candidate materials whose visual feature similarity is greater than a similarity threshold as suspected infringing materials; The determination module is used to extract the semantic features of candidate materials corresponding to each suspected infringing material, as well as the semantic features of copyright materials of each advertising copyright material, determine the infringement confidence level based on the semantic features of each candidate material and the semantic features of the copyright material, and determine the suspected infringing material with the infringement confidence level greater than the confidence level threshold as confirmed infringing material.

9. An advertising material infringement detection device, characterized in that, include: One or more processors; A memory that stores one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the advertising material infringement detection method as described in any one of claims 1-7.

10. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the advertising material infringement detection method as described in any one of claims 1-7.

Images