Advertisement material effect evaluation method, device and equipment based on weighted fusion analysis

Abstract

The invention relates to the technical field of image processing, and provides an advertisement material effect evaluation method, device and equipment based on weighted fusion analysis, aiming at solving the problems that in the prior art, an advertisement piece cannot be associated with an original material, and the accuracy of evaluating the putting effect of the material in different scenes is low. The method comprises the following steps: performing scene switching detection and blind watermark information embedding on an original video material to obtain a target video collection; blind watermark information extraction is carried out on the target video collection, and material fragment information is determined; according to the material fragment information, second-level statistics is carried out on pre-collected delivery data, real-time evaluation parameters and the number of times of use of original video materials are determined, and the real-time evaluation parameters comprise the real-time material playing number; and when the original video material is used for multiple times, according to the real-time material playing number, performing weighted averaging calculation on each single-time material evaluation result, and determining a multiple-time material evaluation result. According to the invention, accurate material performance data is provided, and an advertiser is helped to optimize material use.

Description

[0001] This application is a divisional application of the invention patent application filed on October 24, 2024, with the invention name “Method, device, equipment and storage medium for evaluating the effect of advertising video material delivery” and application number 202411487805.3. Technical Field

[0002] The present invention relates to the field of image processing technology, and in particular to a method, device and equipment for evaluating the effect of advertising materials based on weighted fusion analysis. Background Art

[0003] Amid the rapid growth of digital advertising, advertisers are increasingly relying on diverse video assets for personalized, targeted delivery to boost conversion rates and brand exposure. However, as the number and frequency of assets increase, scientifically and in real time, evaluating the performance of each original video asset in real-time has become a critical issue in optimizing ad delivery. Especially in scenarios where multiple assets and multiple versions are mixed, relying solely on overall delivery performance makes it difficult to discern the strengths and weaknesses of specific assets, which can easily lead to wasted resources and deviations from optimal optimization targets.

[0004] However, existing technologies have the following limitations: material management usually relies on manual operations, which can easily lead to the reuse of materials, version confusion and loss, and weak correlation between the finished advertisements and the original materials, resulting in the inability to accurately track the effectiveness of material use; the monitoring and analysis of advertising effectiveness mostly relies on overall data and lacks detailed analysis of specific material performance.

[0005] Existing Chinese patent CN104954819A discloses a method for intelligently monitoring new media advertisements using digital watermark coding technology. The method includes: first, the advertising material management module categorizes and manages new media advertisements to be broadcast; then, the digital watermark coding module inserts watermarks into the new media advertisements to be broadcast and uniquely codes them; the terminal device transmits the user terminal's parsed results (containing the unique code of the digital watermark) back to the national advertising material library via the Internet or the broadcasting network operator's two-way return network; then, by comparing the collected advertising parsed data from the user terminal with the front-end advertising material library, it obtains advertising monitoring data such as which new media advertising content the user has viewed and the frequency of exposure to specific advertisements. Although the above patent discloses a technical solution for obtaining advertising monitoring data by comparing advertising analysis data with a material library, the correlation between the finished advertisement and the original material is not strong. Once the material is used in multiple finished films or delivered on different platforms, it is difficult to track its specific performance and contribution, which limits the improvement and optimization of material utilization; and there is a lack of detailed analysis of the performance of specific materials, which makes it impossible to accurately evaluate the effects of the materials in different scenarios, making it difficult to implement targeted optimization strategies.

[0006] Therefore, how to associate the finished advertisement with the original material and accurately evaluate the effectiveness of the material in different scenarios is an urgent problem to be solved. Summary of the Invention

[0007] In view of this, the present invention provides a method, device and equipment for evaluating the effect of advertising materials based on weighted fusion analysis, which is used to solve the problem in the prior art that the advertising film cannot be associated with the original material and the evaluation accuracy of the delivery effect of the material in different scenarios is low.

[0008] The technical solution adopted in the present invention is:

[0009] In a first aspect, the present invention provides a method for evaluating the effectiveness of advertising materials based on weighted fusion analysis, the method comprising:

[0010] Performing scene switching detection and blind watermark information embedding on at least one original video material in an advertisement delivery scenario to obtain a target video collection;

[0011] Extracting blind watermark information from the target video collection, and determining material segment information based on the extracted blind watermark information, wherein the material segment information at least includes the name and location of the original video material used in the target video collection;

[0012] Based on the material segment information, pre-collected delivery data is statistically analyzed at the second level to determine real-time evaluation parameters related to the evaluation of the original video material and the number of times the original video material is used in the advertisement delivery, wherein the real-time evaluation parameters include at least: the number of real-time material plays, the real-time material click-through rate, and the real-time material loss rate;

[0013] According to the number of uses, when the original video material is used multiple times, a weighted average calculation is performed on the single material evaluation results corresponding to each original video material according to the number of real-time material playbacks to determine multiple material evaluation results.

[0014] Preferably, performing scene switching detection and blind watermark information embedding on at least one original video material in the advertisement delivery scenario to obtain a target video collection includes:

[0015] Decomposing the original video material into multiple frames of original video images, and obtaining a color histogram corresponding to each original video image;

[0016] Performing difference calculation on the color histograms of the original video images of adjacent frames to obtain the color histogram difference value;

[0017] If the color histogram difference value is greater than a preset color histogram threshold, the adjacent frames of original video images are classified into two scene video segments;

[0018] By using the image processing technology based on frequency domain transformation, the preset blind watermark information is embedded into each frame image in each scene video clip, and the video clips of each scene after the blind watermark information is embedded are edited to obtain the target video highlights.

[0019] Preferably, the method of embedding preset blind watermark information into each frame image in each scene video clip using an image processing technology based on frequency domain transformation, and editing each scene video clip after embedding the blind watermark information to obtain a target video collection includes:

[0020] Perform color space conversion on each frame image in each scene video clip to obtain the target color channel;

[0021] Perform discrete cosine transform on the target color channel to convert each frame image in each scene video clip into a scene frequency domain image;

[0022] According to a preset key, the high-frequency area in the scene frequency domain image is modified and the preset blind watermark information is embedded in the scene frequency domain image.

[0023] Preferably, the preset color histogram threshold is determined by the following steps:

[0024] Obtaining a current frame image from adjacent frames of original video images;

[0025] Determining, based on the current frame image, a color histogram set of all frame images before the current frame image;

[0026] The color histogram threshold is determined based on the color histogram set in combination with a preset smoothing parameter and an initial threshold.

[0027] Preferably, the modifying the high-frequency area in the scene frequency domain image according to the preset key and embedding the preset blind watermark information into the scene frequency domain image includes:

[0028] According to the preset blind watermark information, obtain the template size of the area to be embedded;

[0029] Selecting any target area in the scene frequency domain image according to the size of the template of the area to be embedded, wherein the size of the target area is the same as the size of the template of the area to be embedded;

[0030] Performing frequency domain coefficient analysis on the target area, and calculating the square sum of all frequency domain coefficients in the target area as the regional energy value;

[0031] Comparing the energy value of the region with a preset energy threshold, and if the energy value of the region is greater than the energy threshold, classifying the target region as a high-frequency region;

[0032] If the energy value of the region is less than or equal to the energy threshold, classifying the target region as a low-frequency region;

[0033] Repeatedly select any target area in the scene frequency domain image to perform frequency domain coefficient analysis until all areas in the scene frequency domain image are classified as high-frequency areas or low-frequency areas;

[0034] Acquire multiple frames of scene frequency domain images, and select a scene frequency domain image as the current frame image;

[0035] Using the optical flow method, the positions of each pixel in the high-frequency area of ​​the current frame image and the positions of each pixel in the previous frame image are analyzed to determine the optical flow field corresponding to each pixel;

[0036] Calculate the mean of the optical flow field corresponding to each pixel point to determine the mean optical flow corresponding to the high-frequency area in the current frame image;

[0037] When the optical flow mean is less than a preset optical flow threshold, modifying the high-frequency area in the current frame image and embedding the preset blind watermark information into the scene frequency domain image;

[0038] Repeatedly select a scene frequency domain image as the current frame image to embed blind watermark information until the preset blind watermark information is embedded in all scene frequency domain images.

[0039] Preferably, when the original video material is used multiple times according to the number of times of use, a weighted average calculation is performed on the single material evaluation results corresponding to each original video material according to the number of real-time material playbacks, and after determining the multiple material evaluation results, the method further includes:

[0040] Obtain the advertisement type and delivery platform for the target video highlights;

[0041] The material evaluation result is corrected twice according to the advertisement type and delivery platform, and the corrected evaluation result is used as the delivery effect analysis result.

[0042] Preferably, the single material evaluation result is determined by the following steps:

[0043] Obtaining the total duration of the target video collection, and determining a material position according to the original video material position and the total duration, wherein the material position is used to identify a relative time interval of the original video material in the target video collection;

[0044] Determining standard parameters according to the material position gear, wherein the standard parameters at least include a standard material play count, a standard material click rate, and a standard material loss rate;

[0045] The standard parameter and the evaluation parameter are calculated to determine the single material evaluation result.

[0046] In a second aspect, the present invention provides an advertising material effect evaluation device based on weighted fusion analysis, the device comprising:

[0047] A target video collection determination module is used to perform scene switching detection and blind watermark information embedding on at least one original video material in an advertisement delivery scenario to obtain a target video collection;

[0048] a blind watermark information extraction module, configured to extract blind watermark information from a target video collection and determine material segment information based on the extracted blind watermark information, wherein the material segment information includes at least the name and location of the original video material used in the target video collection;

[0049] a delivery data statistics module, configured to perform second-level statistics on the pre-collected delivery data based on the material clip information, and determine real-time evaluation parameters related to the evaluation of the original video material and the number of times the original video material is used in the advertisement delivery, wherein the real-time evaluation parameters include at least: the number of real-time material plays, the real-time material click-through rate, and the real-time material loss rate;

[0050] The multiple-use material evaluation module is used to perform weighted average calculation on the single material evaluation results corresponding to each original video material according to the number of uses, when the original video material is used multiple times, based on the number of real-time material playbacks, to determine the multiple material evaluation results.

[0051] In a third aspect, an embodiment of the present invention further provides an electronic device comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method of the first aspect of the above-mentioned embodiment.

[0052] In a fourth aspect, an embodiment of the present invention further provides a storage medium having computer program instructions stored thereon, which implements the method of the first aspect of the above-mentioned embodiment when the computer program instructions are executed by a processor.

[0053] In summary, the beneficial effects of the present invention are as follows:

[0054] The present invention provides a method, device, and apparatus for evaluating the effectiveness of advertising materials based on weighted fusion analysis. The method comprises: obtaining at least one original video material uploaded by a target object in an advertising delivery scenario; performing scene switching detection on the original video material and dividing the original video material into multiple scene video segments; utilizing image processing technology based on frequency domain transformation to embed preset blind watermark information into each frame image in each scene video segment, and editing each scene video segment after embedding the blind watermark information to obtain a target video collection; performing blind watermark information extraction on the target video collection, and determining material segment information based on the extracted blind watermark information, wherein the material segment information includes at least the name and location of the original video material used in the target video collection; when the target video collection is used for advertising delivery, evaluating each material segment based on pre-collected delivery data information and the material segment information, and summarizing all material evaluation results to obtain a delivery effect evaluation result. By introducing a blind watermark information embedding and extraction mechanism in the advertising delivery scenario, the present invention solves the problem in the prior art of the difficulty in accurately associating advertising films with original video materials. When generating the target video highlights, the original video material is subjected to scene switching detection and a unique blind watermark is embedded in it, so that each piece of material is still traceable after synthesis; subsequently, by extracting the blind watermark information from the final advertisement after delivery, the name of the original material contained therein and its specific position in the final advertisement can be accurately identified. This process realizes the automatic positioning and identification of the material segments, so that the system can perform second-level statistical analysis on the playback, click and loss data of each piece of material at the specific delivery time and in different scenes. By combining these real-time evaluation parameters with the number of times the material is used, the solution further supports the weighted average evaluation of multiple uses of the material, effectively improving the evaluation accuracy of the material delivery effect, providing advertisers with a more targeted and effective optimization basis, and significantly overcoming the limitations of existing technologies in terms of coarse granularity and poor accuracy in material evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work, and these are all within the scope of protection of the present invention.

[0056] Figure 1 This is a schematic diagram of the overall working process of the advertising material effect evaluation method based on weighted fusion analysis in Example 1 of the present invention;

[0057] Figure 2 Schematic diagram of the process of performing scene switching detection on the original video material in embodiment 1 of the present invention;

[0058] Figure 3 Schematic diagram of the process of determining the color histogram threshold in Example 1 of the present invention;

[0059] Figure 4 Schematic diagram of the process of embedding preset blind watermark information into each frame image in each scene video clip in embodiment 1 of the present invention;

[0060] Figure 5 Schematic diagram of a process for extracting blind watermark information from a target video collection obtained by editing video clips of each scene after embedding blind watermark information in Example 1 of the present invention;

[0061] Figure 6 Schematic diagram of the process of evaluating each material clip in Example 1 of the present invention;

[0062] Figure 7 Schematic diagram of the process of determining a single material evaluation result in Example 1 of the present invention;

[0063] Figure 8 This is a structural block diagram of an advertising material effect evaluation device based on weighted fusion analysis in Example 2 of the present invention;

[0064] Figure 9 This is a schematic diagram of the structure of an electronic device in Example 3 of the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions 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. It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. In the description of the present invention, it should be understood that the orientation or position relationship indicated by the terms "center", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like is based on the orientation or position relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further limitations, elements defined by the phrase "comprising..." do not preclude the presence of additional identical elements in the process, method, article, or apparatus comprising the elements. The embodiments of the present invention and the features thereof may be combined with each other if there is no conflict, and all are within the scope of protection of the present invention.

[0066] Example 1

[0067] See Figure 1 Embodiment 1 of the present invention discloses a method for evaluating the effect of advertising materials based on weighted fusion analysis, the method comprising:

[0068] S1: Obtain at least one original video material uploaded by the target object in the advertising delivery scenario;

[0069] Specifically, acquiring the original video material uploaded by the target audience in the advertising delivery scenario means first receiving the video file uploaded by the user and using it as the initial input data. The original video material includes various types of content, such as product advertisements, brand promotions, film and television drama clips, and user-generated content (UGC). The original video material covers different scenes, themes, and styles, and is diverse and complex. By further processing this original video material, the foundation is laid for the subsequent identification and evaluation of the material clips, thus ensuring that the usage and delivery effect of each material clip can be accurately extracted and analyzed when the original video material is edited and delivered.

[0070] S2: performing scene switching detection on the original video material, and dividing the original video material into a plurality of scene video segments according to the scene switching detection result;

[0071] Specifically, scene switching detection on the original video material refers to the use of image processing and computer vision technology to automatically identify the transition points between different scenes in the video, such as the moment of switching from one scene to another. Based on these scene switching detection results, the original video material is divided into multiple scene video clips, each of which represents a continuous visual and content unit. This process can not only accurately locate the starting and ending positions of each scene, but also provide a clear segmentation basis for subsequent blind watermark embedding and material management, thereby ensuring that each clip can be independently and accurately tracked and analyzed in subsequent editing, delivery and effect evaluation. In this way, by refining the processing of video materials, each scene clip can be managed and utilized more efficiently, improving the accuracy of overall advertising delivery and the accuracy of effect evaluation.

[0072] In one embodiment, see Figure 2 , said S2 includes:

[0073] S21: Decomposing the original video material into multiple frames of original video images, and obtaining a color histogram corresponding to each original video image;

[0074] Specifically, decomposing the original video material into multiple frames of raw video images is the first step in scene cut detection. First, the continuous original video material is decomposed into individual raw video images. Each second of video typically contains multiple frames (e.g., 30 frames per second). Each frame represents a moment in the video. Next, a color histogram is calculated for each raw video frame. A color histogram is a statistical chart that represents the number of pixels of each color in the image. By calculating the color histogram of each frame (denoted as H_i), the color distribution characteristics of each raw video frame can be obtained. These characteristics are used in subsequent scene cut detection.

[0075] S22: performing a difference calculation on the color histograms of the original video images of adjacent frames to obtain a color histogram difference value;

[0076] Specifically, after obtaining the color histogram of each frame of the original video image, the color histogram difference between adjacent frames is calculated. This is achieved by taking the difference between the color histograms of adjacent frames. Common methods include relative entropy (KL divergence) or L2 norm. For frames F_i and F_{i+1}, the color histogram difference D_i=d(H_i,H_{i+1}) is calculated, where d can be KL divergence or L2 norm. In this way, the degree of change in the color distribution of adjacent frame images is quantified. The calculation of color histogram difference values ​​is an important basis for detecting scene changes, because scene changes are usually accompanied by significant image color changes.

[0077] S23: If the color histogram difference value is greater than a preset color histogram threshold, the adjacent frames of original video images are classified into two scene video segments.

[0078] Specifically, to determine whether a scene cut exists, the calculated color histogram difference value D_i for each frame is compared with a preset color histogram threshold T_i. To improve the robustness of detection, an adaptive threshold is used, which is dynamically adjusted based on the changes in the current frame. If the color histogram difference value D_i of a frame is greater than the adaptive threshold T_i, it is considered that a scene cut has occurred between frames F_i and F_{i+1}. At this point, the original video images of adjacent frames are segmented into two different scene video clips, thereby achieving scene segmentation of the video material. This method ensures that each scene video clip has consistent visual characteristics and is significantly different from other scenes.

[0079] In one embodiment, see Figure 3 , the preset color histogram threshold is determined by the following steps:

[0080] S2301: Acquire a current frame image from adjacent frames of original video images;

[0081] Specifically, during the scene switch detection process, the original video material will be processed frame by frame. First, the current frame image in the adjacent frame original video image is obtained, that is, the frame currently being processed. This frame image is the target of the current analysis. By calculating the color histogram difference between it and the previous frame image, it can be determined whether there is a scene switch.

[0082] S2302: Determine, based on the current frame image, a color histogram set of all frame images before the current frame image;

[0083] Specifically, after acquiring the current frame, we collect a set of color histograms for all frames preceding the current one. This set includes the color histograms (H_1, H_2, ..., H_{i-1}) for each frame from the beginning of the video to the current one. These histograms represent the color distribution of each frame. By analyzing this data, we can understand the color changes in the previous frames of the video. This information is crucial for calculating the change trend of the current frame and subsequent threshold adjustments.

[0084] S2303: Determine the color histogram threshold according to the color histogram set, in combination with a preset smoothing parameter and an initial threshold.

[0085] Specifically, in order to make the detection more robust, an adaptive threshold is used to determine whether there is a scene switch. The threshold (T_i) of the current frame will be dynamically adjusted based on the color histogram set, combined with the preset smoothing parameter (α) and the initial threshold (T). The specific calculation method is T_i = αT + (1-α)M_i, where Mi_i is the average difference of the previous n frames, expressed as Mi_i = (1 / n)Σ_{j = i-n+1}^{i}D_j. In this way, the threshold can be dynamically adjusted according to the changes in the current frame, making scene switch detection more sensitive and accurate. If the color histogram difference D_i between the current frame and the previous frame is greater than this adaptive threshold T_i, the system determines that there is a scene switch between the two frames. The setting of this adaptive threshold ensures flexible adjustment when processing different video materials, thereby improving the robustness and accuracy of detection.

[0086] In one embodiment, the S23 further includes:

[0087] S231: If the color histogram difference value is greater than a preset color histogram threshold, obtain adjacent audio frames corresponding to the original video images of adjacent frames;

[0088] Specifically, when the color histogram difference value of adjacent video frames exceeds the preset color histogram threshold, it is preliminarily judged that there may be a scene switch. At this time, in addition to relying on image information, further verification is required in combination with audio information. Therefore, the audio frames corresponding to the two video frames are extracted synchronously. The audio frame refers to a piece of audio data aligned with the video frame on the time axis. By obtaining these audio frames and analyzing the changes in the audio signal, it is possible to confirm whether a scene switch has actually occurred. This process ensures the synchronous processing of video and audio information, laying the foundation for subsequent multimodal analysis.

[0089] S232: Performing audio analysis on the adjacent audio frames using a preset audio feature extraction algorithm to determine a change range of audio features between adjacent audio frames;

[0090] Specifically, after obtaining the audio frame, the adjacent audio frames are analyzed, and an audio feature extraction algorithm is used to detect changes in the audio. The audio feature extraction algorithm includes spectrum analysis, pitch change detection, audio energy analysis, frequency band energy change, and audio event detection technology. The preset audio feature extraction algorithm is used to identify significant changes in the audio, such as the switching of background music, the beginning or end of a conversation, or sudden changes in ambient sound. By comparing the features of adjacent audio frames, the amplitude of the audio feature change is calculated. This amplitude of change can reflect the intensity of the change in the audio content in a short period of time, thereby providing an auxiliary basis for judging scene switching. If the audio feature changes significantly, it means that in addition to the visual scene, the audio environment has also changed significantly, which further supports the judgment of scene switching.

[0091] S233: Inputting adjacent original video frames and adjacent audio frames into a pre-trained multimodal large model, and outputting text content information related to the scene;

[0092] Specifically, the raw video images of adjacent frames are input into a pre-trained multimodal large model for processing. The multimodal large model can comprehensively process visual and audio information and automatically generate text descriptions related to the scene or extract text information in the scene through deep learning technology. This step includes optical character recognition (OCR), which is used to detect and extract text content in video frames, such as logos, subtitles, or written information within the scene. In addition, the multimodal large model can also analyze the combination of visual and audio features of the scene to generate descriptive text information. This text information may include an explanation of the scene content, the capture of semantic changes, or even the transcription of detected dialogues or important sounds. By generating or extracting this text information, the scene content can be further understood at the semantic level, providing richer information for scene switching.

[0093] S234: performing a weighted analysis on the audio feature change amplitude and the text content according to a preset weight, and determining whether to segment the adjacent frame original video image into two scene video segments based on the weighted analysis result.

[0094] Specifically, after completing the extraction of audio and text information, a weighted analysis is performed on the amplitude of changes in audio features and text content according to preset weights. The purpose of weighted analysis is to integrate multimodal information and make more accurate scene switching judgments. The preset weights can be adjusted according to the characteristics of the video content and the analysis objectives. For example, for music videos, the weight of audio features may be higher, while for videos with rich dialogue scenes, the weight of text content may be higher. Through this weighted analysis, audio, text, and image information are combined, and the amplitude of changes between them is comprehensively considered. If the weighted analysis results show obvious changes, it will be judged that there is a scene switch, and the adjacent video frames will be divided into two different scene video clips. In this way, not only is it based on visual features, but audio and text information are also fully utilized to ensure that the detection of scene switches is more accurate and reliable.

[0095] S3: Using image processing technology based on frequency domain transformation, the preset blind watermark information is embedded into each frame image in each scene video clip, and the video clips of each scene with the embedded blind watermark information are edited to obtain the target video highlights;

[0096] In one embodiment, see Figure 4 , the S3 includes:

[0097] S31: Perform color space conversion on each frame image in each scene video clip to obtain a target color channel;

[0098] Specifically, each frame in each scene video clip is converted from the RGB color space to the YUV color space. The YUV color space is divided into three channels: the Y channel (brightness), the U channel (chroma), and the V channel (density). The converted YUV image can more effectively separate color information from brightness information, allowing us to operate on the U and V channels without affecting the overall visual quality. Through this conversion, the target color channels (U and V) of each frame are obtained, laying the foundation for subsequent blind watermark embedding.

[0099] S32: performing discrete cosine transform on the target color channel to convert each frame image in each scene video clip into a scene frequency domain image;

[0100] Specifically, after obtaining the target color channels, these channels are subjected to discrete cosine transform, or DCT. DCT is a frequency domain transform method that can convert images from the spatial domain to the frequency domain. Through DCT, the image is represented as a set of frequency components, where low-frequency components contain the main image information, while high-frequency components contain less detailed information. This transformation enables us to process in the frequency domain to hide blind watermark information while minimizing the impact on the visual quality of the image. After DCT processing, the U and V channels of each frame of the image are converted into frequency domain images, showing the weight of each frequency component.

[0101] S33: modifying the high-frequency region in the scene frequency domain image according to the preset key, and embedding the preset blind watermark information into the scene frequency domain image.

[0102] Specifically, a preset key is used to determine the specific location of the blind watermark information embedded in the frequency domain image. The key is a set of parameters used for positioning, ensuring that the location of the blind watermark is consistent and traceable for each frame. The high-frequency areas in the scene frequency domain image are selected for modification because these areas contain less visual information and have the least impact on the overall visual effect of the image after modification. By embedding the blind watermark information (i.e., a unique ID) in the high-frequency area, it is ensured that each frame of the image contains identifiable hidden information. This process converts the frequency domain image back to the spatial domain image through an inverse transform (IDCT), and finally completes the embedding of the blind watermark information. In this way, each frame of the image carries invisible identification information, ensuring that the material can be accurately identified and tracked when referenced in the film.

[0103] In one embodiment, the S33 includes:

[0104] S331: Obtain the size of the template of the area to be embedded according to the preset blind watermark information;

[0105] Specifically, the blind watermark's information content, such as text, images, or other forms of data, must first be determined. Based on the watermark's complexity and information content, an appropriate template area size must be calculated. For example, if the watermark is a short text, a template size of 8x8 or 16x16 is sufficient. However, if the watermark is an image or larger data, a larger area is required. The choice of template size affects the watermark's clarity and visibility. Too large a size can significantly affect the image after embedding, while too small a size can prevent the complete information from being transmitted. Therefore, a reasonable template size can improve the watermark's concealment and transmission efficiency.

[0106] S332: selecting any target region in the scene frequency domain image according to the size of the region template to be embedded, wherein the size of the target region is the same as the size of the region template to be embedded;

[0107] Specifically, a target region of the same size as the template is selected from the scene frequency domain image. This selection can be done randomly or based on image features. For example, if a frequency domain image of a static scene is selected, an 8x8 region is randomly selected. In practice, the selection of the target region takes into account the content characteristics of the image, such as selecting an area with a relatively flat background to ensure the concealment of the watermark. If the target region is located in an area with minimal texture variation, the visual effect after embedding the watermark will be better.

[0108] S333: Perform frequency domain coefficient analysis on the target area, and calculate the square sum of all frequency domain coefficients in the target area as the area energy value;

[0109] Specifically, the frequency domain coefficients of the selected target area are analyzed. By calculating the square sum of all the frequency domain coefficients in the area, an energy value reflecting the strength of the area is obtained. For example, when analyzing an 8x8 DCT coefficient matrix, each frequency domain coefficient in the matrix is ​​squared and summed up, and the result is the energy value of the area. This value can indicate the details and strength of the area and is usually used to assess whether it is suitable for watermarking. For example, if the energy value of a region is 50, it means that the area is relatively weak.

[0110] S334: Compare the energy value of the region with a preset energy threshold. If the energy value of the region is greater than the energy threshold, classify the target region as a high-frequency region.

[0111] Specifically, the calculated regional energy value is compared with a preset energy threshold. If the regional energy value is greater than the threshold, it is marked as a high-frequency area. For example, assuming the preset energy threshold is 40, if the energy value of a target area is 45, it means that the area has strong details and is suitable for carrying watermarks. Conversely, if the energy value is below the threshold, it is not suitable for watermark embedding. Through this classification, the system can effectively identify which areas can effectively carry watermark information without compromising the overall quality of the image.

[0112] S335: If the region energy value is less than or equal to the energy threshold, classify the target region as a low-frequency region;

[0113] Specifically, when the regional energy value is less than or equal to the preset energy threshold, the region is marked as a low-frequency region. Low-frequency regions usually contain fewer detail changes and are flatter, making them suitable for embedding watermarks without causing obvious visual interference. For example, if the energy value is 35, it means that the region is relatively smooth and suitable for carrying watermarks. Through such classification, a wise choice can be made between high-frequency and low-frequency regions to ensure the visual concealment of the watermark.

[0114] S336: repeating steps S332 to S335 until all regions in the scene frequency domain image are classified as high-frequency regions or low-frequency regions;

[0115] Specifically, the entire scene frequency domain image is comprehensively analyzed, and all target regions are evaluated one by one by repeating steps S332 to S335. The purpose of this is to ensure that each region is classified so that the subsequent watermark embedding strategy can be based on comprehensive data. For example, if there are multiple 8x8 regions in an image, each region will be analyzed one by one until all regions are marked as high frequency or low frequency. This process ensures that the image can be comprehensively evaluated and the appropriate region can be selected for watermark embedding to the greatest extent possible.

[0116] S337: Acquire multiple frames of scene frequency domain images, and select a scene frequency domain image as the current frame image;

[0117] Specifically, the current frame image is obtained from multiple scene frequency domain images. Multiple frame images provide dynamic background information for watermark embedding, which helps to improve the robustness of the watermark. For example, in a video sequence, there may be dozens of frames of images. Selecting one frame from these frames as the current processing object can better analyze the motion of the current frame and compare it with the previous frame, which facilitates the embedding of dynamic watermarks.

[0118] S338: Using an optical flow method, analyze the positions of each pixel point in the high-frequency region of the current frame image and the positions of each pixel point in the previous frame image to determine the optical flow field corresponding to each pixel point;

[0119] Specifically, the optical flow method is applied to analyze the motion of high-frequency areas in the current frame. By calculating the changes in pixel positions between adjacent frames, the optical flow method can effectively extract motion information. For example, if a pixel in the high-frequency area of ​​the current frame in a video moves 5 pixels to the left, while the corresponding pixel in the previous frame remains unchanged, this motion information will be recorded and an optical flow field will be constructed. This process provides a foundation for subsequent watermark embedding, identifying which areas have less motion and are therefore more suitable for watermarking.

[0120] S339: Calculate the mean of the optical flow field corresponding to each pixel point to determine the mean optical flow corresponding to the high-frequency area in the current frame image;

[0121] Specifically, the mean of all high-frequency pixel points in the optical flow field is calculated to determine the overall motion state of the area. For example, if the optical flow values ​​of multiple pixels in the high-frequency area are 2, 3, 4, and 5 respectively, then their mean is 3.5. This mean can reflect the degree of motion in the area. If the mean is small, it means that the area has relatively little motion and is suitable for watermark embedding.

[0122] S3310: When the optical flow mean is less than a preset optical flow threshold, modify the high-frequency region in the current frame image and embed the preset blind watermark information into the scene frequency domain image;

[0123] Specifically, if the calculated optical flow mean is lower than the preset optical flow threshold, the high-frequency region of the current frame will be modified to embed the blind watermark information. For example, if the set optical flow threshold is 4 and the mean is 3.5, the region will be judged suitable for watermark embedding. At this time, the watermark information will be added to the frequency domain coefficients of the region to ensure the concealment and effectiveness of the watermark.

[0124] S3311: Repeat steps S337 to S3310 until the preset blind watermark information is embedded into all scene frequency domain images.

[0125] Specifically, steps S337 to S3310 are looped through, processing each scene frequency domain image frame by frame until all images are watermarked. This continuous processing ensures that the watermark information is effectively transmitted across multiple frames and improves the robustness and concealment of the watermark. Dynamic analysis of each frame ensures that the watermark information remains consistent and stable throughout the entire video sequence.

[0126] S4: extracting blind watermark information from the target video collection, and determining material segment information based on the extracted blind watermark information, wherein the material segment information at least includes the name and location of the original video material used in the target video collection;

[0127] Specifically, each frame in the target video collection is first processed. The image is converted from the frequency domain back to the spatial domain through the inverse discrete cosine transform (IDCT), and the blind watermark information embedded in the high-frequency area is extracted. Using a preset key, the specific position and content of the blind watermark information in the image are accurately located. Through the extracted blind watermark information, the system can identify the unique ID in each frame and determine which original material clips are used in the target video collection. This process allows the source and location of each material clip to be accurately tracked even if the material is edited, mixed, or reused in different scenes, ensuring that the references of the material in the finished film can be perfectly identified and accurate data reflow is achieved.

[0128] In one embodiment, see Figure 5 , said S4 includes:

[0129] S41: Decomposing the target video collection into multiple frames of target video images, performing discrete cosine transform on each of the target video images, and determining a target frequency domain image;

[0130] Specifically, the target video collection is decomposed into individual target video images, each frame representing a moment in the video. A discrete cosine transform (DCT) is applied to each frame, converting the target video image from the spatial domain to the frequency domain to determine the target frequency domain image. Through DCT, the image is decomposed into a series of frequency components, each representing a change in a different frequency within the image. These determined target frequency domain images reveal the image's high- and low-frequency information, providing the basis for the subsequent blind watermark extraction process. Through this transformation, the system is able to process and analyze hidden information in the image in the frequency domain.

[0131] S42: determining a target position containing blind watermark information in the target frequency domain image based on the key;

[0132] Specifically, after acquiring the target frequency domain images, a preset key is used to determine the specific location of the blind watermark information within these frequency domain images. The key is a set of predefined parameters used to identify the frequency components and locations where the blind watermark information is embedded. By referring to the key, the area containing the blind watermark information in the frequency domain image can be accurately found. This step ensures the accuracy of blind watermark extraction, because only the key can correctly decode the embedded hidden information. The use of the key enables the systematic location of the unique ID information hidden in each frame of the image.

[0133] S43: Extracting and restoring blind watermark information from the target frequency domain image according to the target position to determine the material clip information.

[0134] Specifically, after determining the target position of the blind watermark information, these positions are processed to extract the hidden blind watermark information. The extraction process includes reading the modification information in the high-frequency area from the frequency domain image and restoring the frequency domain information to the spatial domain through the inverse discrete cosine transform (IDCT). The extracted blind watermark information contains an embedded unique ID. These IDs correspond one-to-one to the original material clips. Through these IDs, it is possible to identify and confirm which original material clips are used in the target video highlights, ensuring accurate tracking of the material and the accuracy of data reflow. In this way, detailed usage and performance data can be provided for each material clip, thereby optimizing advertising delivery and content management.

[0135] S5: When the target video collection is used for advertisement delivery, each material segment is evaluated based on the pre-collected delivery data information and the material segment information, and all material evaluation results are summarized to obtain a delivery effect evaluation result.

[0136] Specifically, when the target video highlights are used for advertising delivery, the collected delivery data is correlated and analyzed with the previously extracted material clip information. Delivery data includes key indicators such as playback volume, click-through rate, viewing time, and user interaction. By matching this delivery data with the unique ID of each material clip, it is possible to accurately evaluate the performance of each material clip in different advertising delivery scenarios. During the evaluation process, various indicators are comprehensively considered, and the effectiveness of the material clips is quantitatively scored. A detailed evaluation report is generated, and finally, the material evaluation results are output to provide advertisers with targeted optimization suggestions, helping them to improve the efficiency of material utilization and advertising effectiveness in future advertising creation and delivery.

[0137] In one embodiment, see Figure 6 , the S5 includes:

[0138] S51: Based on the material segment information, pre-collected delivery data is statistically analyzed at the second level to determine real-time evaluation parameters related to the evaluation of the original video material and the number of times the original video material is used in the advertisement delivery, wherein the real-time evaluation parameters include at least: the number of real-time material plays, the real-time material click-through rate, and the real-time material loss rate;

[0139] Specifically, based on the material clip information, which includes at least the name and location of the original video material used in the target video collection, the collected advertising delivery data is refined and statistically analyzed in seconds. Specifically, for each original video material clip, its specific usage period in the video is first determined. For example, assuming that the original video material clip is used between the 2nd and 5th seconds of the video, the delivery data within these seconds will be focused on. For the number of real-time material plays, the number of plays from the 2nd second of the material's start of use will be used as the evaluation benchmark. Next, the average per-second dropout rate of the original video material clip during its usage period is calculated. The real-time material dropout rate refers to the proportion of viewers who leave the video within a certain second. The dropout rates from the 2nd to the 5th second are calculated separately, and then the average of these four values ​​is calculated to obtain the average per-second dropout rate of the material. Similarly, the click-through rate per second is also calculated. The real-time material click-through rate reflects the audience's interaction with the advertisement while watching the video. The click-through rates from the 2nd to the 5th second are counted separately, and their average is calculated to obtain the average per-second click-through rate of the material. Through such second-level statistics, key evaluation parameters related to material evaluation can be extracted, including the number of plays, average churn rate and average click-through rate. These parameters provide basic data for subsequent material evaluation, making the evaluation process more accurate and scientific, thereby helping advertisers understand the specific performance of each material in different time periods and delivery scenarios, and optimize advertising delivery strategies.

[0140] S52: Based on the number of uses, if the original video material is used once, determining a single material evaluation result based on the evaluation parameter and the position of the original video material;

[0141] Specifically, based on the number of uses, when the original video material is used once, the previously collected evaluation parameters and the original video material location will be used to calculate its single evaluation result. This step ensures that each single use of each material is evaluated in detail for subsequent comprehensive evaluation.

[0142] In one embodiment, see Figure 7 , the S52 includes:

[0143] S521: Obtain the total duration of the target video collection, and determine the material position level according to the original video material position and total duration;

[0144] Specifically, based on the material clip information, determine the specific usage position of the material in the target video collection, including the start and end time of the material, as well as the specific time period in the video. This step helps to understand the position of the material in the video and provides a reference for subsequent evaluation. The material position is divided into different gears according to the position of the material clip in the video and the total length of the video. For example, a video can be divided into five gears: beginning, front, middle, back, and end. Assuming the total length of the video is 35 seconds, each gear is approximately 7 seconds. Assuming that the material usage position is between 12 seconds and 15.3 seconds, the average is 13.65. Calculate and round down the material center position / segment interval, and calculate and determine the specific gear to which it belongs based on the center position of the material clip.

[0145] S522: Determine standard parameters based on the material position, wherein the standard parameters at least include a standard material play count, a standard material click rate, and a standard material loss rate;

[0146] Specifically, after determining the specific gear of the material, the standard parameters are determined according to the material position gear, wherein the standard parameters include at least the standard material play number, the standard material click rate and the standard material loss rate; and each gear corresponds to different standard parameters. For example, reference indicators for 5 different positions are defined respectively, R[0].ctr represents the standard click rate of the beginning material, and R[4].play represents the standard play number of the ending material. The standard parameters provide a benchmark for subsequent evaluation, making the evaluation process more accurate and targeted.

[0147] S523: Calculate the standard parameters and the evaluation parameters to determine the evaluation result of the single material. Specifically, the evaluation parameters of the material are compared and analyzed with the standard parameters to calculate the evaluation result of the single use of the material. The evaluation result includes the score of each indicator, taking into account the performance of the playback volume, click-through rate and churn rate, and finally obtains a comprehensive score. The calculation formula is:

[0148]

[0149] Where ε=0.000001, x.score represents the comprehensive score, x.play represents the number of original video material plays, x.ctr represents the average click-through rate of the original video material per second, x.lr represents the average loss rate of the original video material per second, x.pos represents the original video material position, R[x.pos].play represents the standard number of plays of the original video material position, R[x.pos].ctr represents the standard click-through rate of the original video material position, R[x.pos].lr represents the standard loss rate of the original video material position, and the symbol This score represents the value truncated based on the lower limit a and upper limit b. This score reflects the overall effectiveness of the creative in a single use, providing advertisers with a specific evaluation basis.

[0150] S53: If the original video material is used multiple times, a weighted average calculation is performed on the single material evaluation results corresponding to each original video material according to the number of real-time material playbacks to determine multiple material evaluation results.

[0151] Specifically, when a material is used multiple times, a weighted average of the evaluation results of each use of the material is calculated. The weighting is based on the logarithmic number of plays of each use, ensuring that uses with a large number of plays have a higher weight in the comprehensive evaluation. The calculation formula is as follows:

[0152]

[0153] Here, X[i].play represents the number of plays of the original video material corresponding to its i-th use; X[i].score represents the evaluation result of the original video material corresponding to its i-th use. This approach provides a more comprehensive evaluation result, reflecting the overall performance of the material over multiple uses. This weighted average calculation method improves evaluation accuracy and helps advertisers better understand the comprehensive effectiveness of the material in different delivery scenarios.

[0154] After S5, the method further includes:

[0155] S61: Obtaining the advertisement type and delivery platform for the target video highlights;

[0156] Specifically, the target video highlights are obtained for advertising delivery, and the platform and ad type targeted are determined. For example, for traditional media like television, factors such as channel switching time and audience viewing habits are considered when evaluating creatives. For streaming platforms like short video platforms, user behavior data, such as the time and number of comments sent, interactive behaviors such as likes and reposts, and the platform's content recommendation algorithm, are analyzed during creative evaluation. Regarding ad types, Vlog ads typically focus on natural content placement and the creator's personal style, while review ads emphasize product performance and audience trust. In this step, this data will be collected and analyzed to provide a foundation for subsequent advertising strategy development, ensuring that the ad content and delivery methods match the characteristics of the platform and ad type, thereby improving advertising effectiveness and user acceptance.

[0157] S62: performing a secondary correction on the material evaluation result according to the advertisement type and delivery platform, and using the corrected evaluation result as the delivery effect analysis result.

[0158] Specifically, after obtaining the characteristics of the ad type and delivery platform, the preliminary material evaluation results are subjected to a secondary correction. This process combines the original evaluation results with the platform's user behavior patterns and ad interaction data through an in-depth analysis of the characteristics of the ad type and delivery platform. For example, in Vlog ads, platform users prefer natural and personalized content, which will lower the score of rigid ads and increase the score of content-based ads. In evaluation ads, more emphasis will be placed on the clarity and credibility of product information to ensure the best delivery effect on specific platforms. In addition, for streaming platforms, the focus will be on user interactive behavior data, such as barrage and likes, while in traditional media, more attention will be paid to the ad broadcast time period and the audience's viewing habits. Through this series of corrections, the final evaluation results will be more accurate and can better guide subsequent advertising delivery and effect evaluation, ensuring that the ads achieve the expected communication effect.

[0159] Example 2

[0160] See Figure 8 Embodiment 2 of the present invention further provides an advertising material effect evaluation device based on weighted fusion analysis, the device comprising:

[0161] Material upload module, used to obtain the uploaded original video material;

[0162] A material segmentation module is used to perform scene switching detection on the original video material and segment the original video material into multiple scene video segments according to the scene switching detection result;

[0163] A blind watermark embedding module is used to embed the preset blind watermark information into each frame image in each scene video clip using image processing technology based on frequency domain transformation;

[0164] A blind watermark extraction module is used to extract blind watermark information from the target video collection obtained by editing the video clips of each scene after embedding the blind watermark information, and determine the material clip information in the target video collection based on the extracted blind watermark information;

[0165] The material evaluation module is used to evaluate each material segment based on the pre-collected delivery data information and the material segment information when the target video collection is used for advertising delivery, and output the material evaluation result.

[0166] Specifically, an advertising material effect evaluation device based on weighted fusion analysis provided by an embodiment of the present invention is adopted, and the device includes: a material uploading module, which is used to obtain the uploaded original video material; a material segmentation module, which is used to perform scene switching detection on the original video material, and divide the original video material into multiple scene video clips based on the scene switching detection result; a blind watermark embedding module, which is used to use the image processing technology based on frequency domain transformation to embed preset blind watermark information into each frame image in each scene video clip; a blind watermark extraction module, which is used to extract blind watermark information from the target video collection obtained by editing each scene video clip after embedding the blind watermark information, and determine the material clip information in the target video collection based on the extracted blind watermark information; a material evaluation module, which is used to evaluate each material clip based on the pre-collected delivery data information and the material clip information when the target video collection is used for advertising delivery, and output the material evaluation result. This device performs scene switching detection on the original video material, divides the original video material into multiple scene video clips, and uses image processing technology based on frequency domain transformation to embed blind watermark information into each frame image of each scene clip, ensuring that each material clip can be uniquely identified. In the edited target video collection, by extracting the blind watermark information, the position and usage of each original video material can be accurately identified and tracked. When advertising is delivered, the delivery effect of each original video material is accurately evaluated based on the pre-collected delivery data information and the extracted material clip information. This not only solves the problem of weak correlation between the finished advertisement and the original video material, but also provides accurate material performance data through refined evaluation, which helps advertisers optimize material usage and improve the efficiency and effectiveness of advertising delivery. This systematic and automated processing and analysis process reduces human intervention and subjective bias, and improves the accuracy and real-time nature of the evaluation.

[0167] Example 3

[0168] In addition, combined Figure 1The advertising material effect evaluation method based on weighted fusion analysis of the embodiment 1 of the present invention can be implemented by an electronic device. Figure 9 A schematic diagram of the hardware structure of an electronic device provided in Example 3 of the present invention is shown.

[0169] An electronic device may include a processor and a memory storing computer program instructions.

[0170] Specifically, the processor may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits for implementing the embodiments of the present invention.

[0171] The memory may include a large capacity memory for data or instructions. By way of example and not limitation, the memory may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory may include a removable or non-removable (or fixed) medium. Where appropriate, the memory may be inside or outside the data processing device. In a specific embodiment, the memory is a non-volatile solid-state memory. In a specific embodiment, the memory includes a read-only memory (ROM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or a flash memory, or a combination of two or more of these.

[0172] The processor reads and executes computer program instructions stored in the memory to implement any one of the advertising material effect evaluation methods based on weighted fusion analysis in the above embodiments.

[0173] In one example, the electronic device may further include a communication interface and a bus. Figure 9 As shown, the processor, memory, and communication interface are connected via a bus and communicate with each other.

[0174] The communication interface is mainly used to implement communication between the modules, devices, units and / or equipment in the embodiments of the present invention.

[0175] Bus comprises hardware, software or both, couples the parts of described equipment together.For example, and not limitation, bus can comprise accelerated graphics port (AGP) or other graphics bus, enhanced industry standard architecture (EISA) bus, front side bus (FSB), hypertransport (HT) interconnection, industry standard architecture (ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations.In suitable cases, bus can comprise one or more buses.Although the embodiment of the present invention describes and shows specific bus, the present invention considers any suitable bus or interconnection.

[0176] The electronic device provided in this embodiment performs scene switching detection on the original video material, divides the original video material into multiple scene video segments, and uses image processing technology based on frequency domain transformation to embed blind watermark information into each frame image of each scene segment, ensuring that each material segment can be uniquely identified. In the edited target video collection, by extracting the blind watermark information, the location and usage of each original video material can be accurately identified and tracked. When advertising is delivered, the delivery effect of each original video material is accurately evaluated based on the pre-collected delivery data information and the extracted material segment information. This not only solves the problem of weak correlation between the finished advertisement and the original video material, but also provides accurate material performance data through refined evaluation, which helps advertisers optimize material usage and improve the efficiency and effectiveness of advertising delivery. This systematic and automated processing and analysis process reduces manual intervention and subjective bias, and improves the accuracy and real-time performance of the evaluation.

[0177] Example 4

[0178] In addition, in conjunction with the advertising material effectiveness evaluation method based on weighted fusion analysis in the first embodiment, the fourth embodiment of the present invention may also be implemented by providing a computer-readable storage medium. The computer-readable storage medium stores computer program instructions; when executed by a processor, the computer program instructions implement any of the advertising material effectiveness evaluation methods based on weighted fusion analysis in the above embodiments.

[0179] The storage medium provided in this embodiment performs scene switching detection on the original video material, divides the original video material into multiple scene video clips, and uses image processing technology based on frequency domain transformation to embed blind watermark information into each frame image of each scene clip, ensuring that each material clip can be uniquely identified. In the edited target video collection, by extracting the blind watermark information, the location and usage of each original video material can be accurately identified and tracked. When advertising is delivered, the delivery effect of each original video material is accurately evaluated based on the pre-collected delivery data information and the extracted material clip information. This not only solves the problem of weak correlation between the finished advertisement and the original video material, but also provides accurate material performance data through refined evaluation, which helps advertisers optimize material usage and improve the efficiency and effectiveness of advertising delivery. This systematic and automated processing and analysis process reduces manual intervention and subjective bias, and improves the accuracy and real-time nature of the evaluation.

[0180] In summary, the embodiments of the present invention provide a method, apparatus, and device for evaluating the effectiveness of advertising materials based on weighted fusion analysis.

[0181] It should be understood that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted. In the above embodiments, several specific steps are described and illustrated as examples. However, the method of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art may make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present invention.

[0182] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant location, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0183] It should also be noted that the exemplary embodiments described herein describe methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the steps described above. In other words, the steps may be performed in the order described in the embodiments, or in a different order, or several steps may be performed simultaneously.

[0184] The above description is only a specific embodiment of the present invention. Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited to this. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention.

Claims

1. A method for evaluating the effect of advertising materials based on weighted fusion analysis, characterized in that: The method comprises: Performing scene switching detection and blind watermark information embedding on at least one original video material in an advertisement delivery scenario to obtain a target video collection; Extracting blind watermark information from the target video collection, and determining material segment information based on the extracted blind watermark information, wherein the material segment information at least includes the name and location of the original video material used in the target video collection; Based on the material segment information, pre-collected delivery data is statistically analyzed at the second level to determine real-time evaluation parameters related to the evaluation of the original video material and the number of times the original video material is used in the advertisement delivery, wherein the real-time evaluation parameters include at least: the number of real-time material plays, the real-time material click-through rate, and the real-time material loss rate; According to the number of uses, when the original video material is used multiple times, a weighted average calculation is performed on the single material evaluation results corresponding to each original video material according to the number of real-time material playbacks to determine multiple material evaluation results.

2. The advertising material effect evaluation method based on weighted fusion analysis according to claim 1 is characterized in that: The performing scene switching detection and blind watermark information embedding on at least one original video material in the advertisement delivery scenario to obtain a target video collection includes: Decomposing the original video material into multiple frames of original video images, and obtaining a color histogram corresponding to each original video image; Performing difference calculation on the color histograms of the original video images of adjacent frames to obtain the color histogram difference value; If the color histogram difference value is greater than a preset color histogram threshold, the adjacent frames of original video images are classified into two scene video segments; By using the image processing technology based on frequency domain transformation, the preset blind watermark information is embedded into each frame image in each scene video clip, and the video clips of each scene after the blind watermark information is embedded are edited to obtain the target video highlights.

3. The advertising material effect evaluation method based on weighted fusion analysis according to claim 2 is characterized in that: The method of using the image processing technology based on frequency domain transformation to embed the preset blind watermark information into each frame image in each scene video clip, and editing the scene video clips embedded with the blind watermark information to obtain the target video highlights includes: Perform color space conversion on each frame image in each scene video clip to obtain the target color channel; Perform discrete cosine transform on the target color channel to convert each frame image in each scene video clip into a scene frequency domain image; According to a preset key, the high-frequency area in the scene frequency domain image is modified and the preset blind watermark information is embedded in the scene frequency domain image.

4. The advertising material effect evaluation method based on weighted fusion analysis according to claim 2 is characterized in that: The preset color histogram threshold is determined by the following steps: Obtaining a current frame image from adjacent frames of original video images; Determining, based on the current frame image, a color histogram set of all frame images before the current frame image; The color histogram threshold is determined based on the color histogram set in combination with a preset smoothing parameter and an initial threshold.

5. The advertising material effect evaluation method based on weighted fusion analysis according to claim 3 is characterized in that: The modifying the high-frequency area in the scene frequency domain image according to the preset key and embedding the preset blind watermark information into the scene frequency domain image includes: According to the preset blind watermark information, obtain the template size of the area to be embedded; Selecting any target area in the scene frequency domain image according to the size of the template of the area to be embedded, wherein the size of the target area is the same as the size of the template of the area to be embedded; Performing frequency domain coefficient analysis on the target area, and calculating the square sum of all frequency domain coefficients in the target area as the regional energy value; Comparing the energy value of the region with a preset energy threshold, and if the energy value of the region is greater than the energy threshold, classifying the target region as a high-frequency region; If the energy value of the region is less than or equal to the energy threshold, classifying the target region as a low-frequency region; Repeatedly select any target area in the scene frequency domain image to perform frequency domain coefficient analysis until all areas in the scene frequency domain image are classified as high-frequency areas or low-frequency areas; Acquire multiple frames of scene frequency domain images, and select a scene frequency domain image as the current frame image; Using the optical flow method, the positions of each pixel in the high-frequency area of ​​the current frame image and the positions of each pixel in the previous frame image are analyzed to determine the optical flow field corresponding to each pixel; Calculate the mean of the optical flow field corresponding to each pixel point to determine the mean optical flow corresponding to the high-frequency area in the current frame image; When the optical flow mean is less than a preset optical flow threshold, modifying the high-frequency area in the current frame image and embedding the preset blind watermark information into the scene frequency domain image; Repeatedly select a scene frequency domain image as the current frame image to embed blind watermark information until the preset blind watermark information is embedded in all scene frequency domain images.

6. The advertising material effect evaluation method based on weighted fusion analysis according to claim 1 is characterized in that: When the original video material is used multiple times according to the number of times of use, a weighted average calculation is performed on the single material evaluation results corresponding to each original video material according to the number of real-time material playbacks, and after determining the multiple material evaluation results, the method further includes: Obtain the advertisement type and delivery platform for the target video highlights; The material evaluation result is corrected twice according to the advertisement type and delivery platform, and the corrected evaluation result is used as the delivery effect analysis result.

7. The advertising material effect evaluation method based on weighted fusion analysis according to any one of claims 1 to 6, characterized in that: The single material evaluation result is determined by the following steps: Obtaining the total duration of the target video collection, and determining a material position according to the original video material position and the total duration, wherein the material position is used to identify a relative time interval of the original video material in the target video collection; Determining standard parameters according to the material position gear, wherein the standard parameters at least include a standard material play count, a standard material click rate, and a standard material loss rate; The standard parameter and the evaluation parameter are calculated to determine the single material evaluation result.

8. An advertising material effect evaluation device based on weighted fusion analysis, characterized in that: The device comprises: A target video collection determination module is used to perform scene switching detection and blind watermark information embedding on at least one original video material in an advertisement delivery scenario to obtain a target video collection; a blind watermark information extraction module, configured to extract blind watermark information from a target video collection and determine material segment information based on the extracted blind watermark information, wherein the material segment information includes at least the name and location of the original video material used in the target video collection; a delivery data statistics module, configured to perform second-level statistics on the pre-collected delivery data based on the material clip information, and determine real-time evaluation parameters related to the evaluation of the original video material and the number of times the original video material is used in the advertisement delivery, wherein the real-time evaluation parameters include at least: the number of real-time material plays, the real-time material click-through rate, and the real-time material loss rate; The multiple-use material evaluation module is used to perform weighted average calculation on the single material evaluation results corresponding to each original video material according to the number of uses, when the original video material is used multiple times, based on the number of real-time material playbacks, to determine the multiple material evaluation results.

9. An electronic device, characterized in that: include: At least one processor, at least one memory, and computer program instructions stored in the memory, which implement the method according to any one of claims 1 to 7 when the computer program instructions are executed by the processor.

10. A storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images