A video identification method based on video frame features

Abstract

The present invention discloses a video identification method based on video frame features, which relates to the field of video classification technology. The method comprises step S100, obtaining an identification video and extracting processing features by frame, wherein the processing features include color comparison data, marker value comparison data and portrait data; step S200, analyzing and processing the color comparison data and the portrait data to generate color statistical labels; step S300, analyzing and processing the marker value comparison data and the portrait data to generate marker value labels; step S400, combining and analyzing the color statistical labels and the marker value labels to generate label comparison objects, wherein the label comparison objects include label high difference objects, label medium difference objects and label low difference objects; step S500, performing label proportion weight analysis on the label comparison objects, substituting the label proportion comparison threshold into the label proportion comparison threshold for comparison, and integrating the analysis to generate real video targets, refined identification targets and forged video targets.

Description

Technical Field

[0001] The present invention relates to the field of video classification, and in particular to a video identification method based on video frame features. Background Art

[0002] After searching, the publication number CN113034430B was used to identify the video authenticity verification method and system based on time watermark change analysis. This comparative document analyzed the watermarks in a large number of videos and found that for time watermarks, they change once per second, and the change position is relatively fixed, which is the last digit of the time watermark second. Video image detection is used to identify whether the seconds at the end of adjacent frames have changed, and the clock change pattern is calculated, thereby achieving the purpose of video authenticity verification.

[0003] The existing technology has the following deficiencies:

[0004] 1. Some existing technologies may rely on only a single feature for identification, such as voiceprint, watermark, or video quality. This reliance on a single feature may result in inaccurate and unreliable identification results.

[0005] 2. Existing technologies may have limitations when dealing with different types of videos and scenes. In video authenticity verification, they may only focus on specific features, such as audio features and inter-frame correlation, while ignoring other important features, such as image features and motion features.

[0006] In order to solve the above-mentioned problems, a video identification method based on video frame features is proposed. Summary of the Invention

[0007] The purpose of the present invention is to provide a video identification method based on video frame features to solve the shortcomings of the background technology.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] The video identification method based on video frame features comprises the following steps:

[0010] Step S100, obtaining the identification video and extracting processing features by frame, wherein the processing features include color comparison data, marker value comparison data and portrait data;

[0011] Step S200: Analyze and process the color comparison data and the portrait data to generate a color statistical label;

[0012] Step S300 , analyzing and processing the marker value comparison data and the portrait data to generate a marker value label;

[0013] Step S400 , combining and analyzing the color statistical labels and the marker numerical labels to generate label comparison objects, wherein the label comparison objects include label high difference objects, label medium difference objects, and label low difference objects;

[0014] Step S500 , performing label ratio weight analysis on the label comparison object to generate a real video target, a refined identification target, and a forged video target.

[0015] In a preferred embodiment, the identification video is divided into m identification intervals according to a single frame image, where m is an integer greater than 0;

[0016] The color comparison data includes basic color data and secondary color data, the basic color data is the basic color quantity value, and the secondary color data is the secondary color quantity value.

[0017] In a preferred embodiment, the marker value comparison data is specifically marker value data, and the marker value data is the number of markers entered into the marker analysis model.

[0018] In a preferred embodiment, the portrait data includes a portrait offset value and a portrait lens ratio. The portrait offset value is based on the lower neck end of the person as a reference point, specifically the difference in the lateral offset displacement of the portrait. The portrait lens ratio is the proportion of the person's image in the image.

[0019] In a preferred embodiment, the step of generating color statistical labels specifically includes:

[0020] The color statistical label includes a first similarity label, a first slight difference label, and a first height difference label;

[0021] Obtain the basic color data z, secondary color data c, portrait offset value p, and portrait lens ratio v in the frame image of the kth frame of the detection video, and obtain the first color image ratio G(k) in the kth frame through formula analysis. Always greater than 0;

[0022] The same method is used to obtain the color comparison data and portrait data in the frame image of the k+n frame, where n is the frame number difference, n= , the second color image ratio in the k+n frame is obtained by formula analysis (k+n), by obtaining the first color image ratio G(k) and the second color image ratio The difference between (k+n) is processed by formula to obtain the color statistical difference α;

[0023] α>0, substitute the color statistical difference α into the color comparison thresholds α1 and α2 for comparison analysis. The color comparison threshold α1 is less than the color comparison threshold α2. When the color statistical difference α is greater than 0 and less than the color comparison threshold α1, the first similarity label is generated for the frame images between the kth frame and the k+nth frame;

[0024] When the color statistical difference α is greater than the color comparison threshold α1 and less than the color comparison threshold α2, the first slight difference label is generated for the frame image between the kth frame and the k+nth frame; when the color statistical difference α is greater than the color comparison threshold α2, the first high difference label is generated for the frame image between the kth frame and the k+nth frame.

[0025] In a preferred embodiment, the steps of generating a marker numerical label are:

[0026] The marker numerical labels include a second similarity label, a second slight difference label, and a second high difference label;

[0027] Obtain the marker numerical data d, portrait offset value p, and portrait shot ratio v in the frame image of the kth frame of the detection video, and obtain the first marker image ratio H(k) in the kth frame through formula analysis. Always greater than 0;

[0028] The same method is used to obtain the marker numerical comparison data and portrait in the frame image of the k+nth frame, where n is the frame number difference, n= , formula analysis is performed to obtain the second marker image ratio H(k+n) in the k+nth frame, and the marker statistical difference β is obtained by formula processing the difference between the obtained first marker image ratio H(k) and the second marker image ratio H(k+n), β>0;

[0029] The marker statistical difference β is substituted into the marker comparison thresholds β1 and β2 for comparison analysis. The marker comparison threshold β1 is less than the marker comparison threshold β2. When the marker statistical difference β is greater than 0 and less than the marker comparison threshold β1, the second similarity label is generated for the frame images between the kth frame and the k+nth frame; when the marker statistical difference β is greater than the marker comparison threshold β1 and less than the marker comparison threshold β2, the second slight difference label is generated for the frame images between the kth frame and the k+nth frame; when the marker statistical difference β is greater than the marker comparison threshold β2, the second high difference label is generated for the frame images between the kth frame and the k+nth frame.

[0030] In a preferred embodiment, the combined analysis step of the color statistical label and the marker numerical label comprises:

[0031] When the frame images between the kth frame and the k+nth frame simultaneously have the first similar label and the second similar label, the first slight difference label and the second similar label, or the second slight difference label and the first similar label, the frame images between the kth frame and the k+nth frame are subjected to label low-difference object generation. When the frame images between the kth frame and the k+nth frame simultaneously have the first slight difference label and the second slight difference label, the first similar label and the second high-difference label, or the second similar label and the first high-difference label, the frame images between the kth frame and the k+nth frame are subjected to label medium-difference object generation. When the frame images between the kth frame and the k+nth frame simultaneously have the first slight difference label and the second high-difference label, the first high-difference label and the second slight difference label, or the first high-difference label and the second high-difference label, the frame images between the kth frame and the k+nth frame are subjected to label high-difference object generation.

[0032] In a preferred embodiment, the steps of generating a real video target, refining the identification target, and forging the video target include:

[0033] Assume that among m identification intervals, the number of identification intervals containing high-difference objects is x, the number of identification intervals containing medium-difference objects is y, and the number of identification intervals containing low-difference objects is s. Use the weight analysis formula to formally analyze the high-difference objects x, medium-difference objects y, and low-difference objects s to obtain the identification factor W.

[0034] Set the identification factor comparison thresholds W1 and W2, and substitute the generated identification factor W into the identification factor comparison thresholds W1 and W2 for analysis, where W1 < W2;

[0035] If the identification factor W is greater than the identification factor comparison threshold W2, then the real video target is generated for the identification video;

[0036] If the identification factor W is greater than the identification factor comparison threshold W1 and less than the identification factor comparison threshold W2, then a refined identification target is generated for the identification video;

[0037] If the identification factor W is greater than 0 and less than the identification factor comparison threshold W1, a forged video target is generated for the identification video.

[0038] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0039] 1. By collecting multiple features such as the number of colors, number of markers, and human movement in the video frame, richer information can be obtained for identification. This type of comprehensive analysis can increase the accuracy and reliability of identification.

[0040] 2. The number of colors and markers reflects the visual information in the video frame. The authenticity of the video can be judged by counting and analyzing these features. Compared with the traditional single-feature method, comprehensive consideration of multiple visual features can provide a more comprehensive identification basis.

[0041] 3. The motion characteristics of the human body in a video can provide more contextual information. By analyzing parameters such as the image ratio and lateral offset of the human figure, the authenticity of the video can be determined. This analysis method has certain advantages in identifying synthesized, tampered, or forged videos.

[0042] In summary, the present invention adopts frame image processing of video frames, combined with a method of color number, marker number and human motion numerical value, aiming to improve the accuracy and reliability of video authenticity identification by comprehensively utilizing multiple feature information, and more comprehensively consider the content and features in the video, thereby increasing the credibility of the identification. Such a method can make full use of the visual information in the video, provide a richer and multi-dimensional basis for judgment, and help to accurately distinguish real videos from fake videos. At the same time, the method can also provide a more intuitive and explainable way to determine the authenticity of the video, making it more operational and understandable in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0044] Figure 1 The present invention is a flow chart of a video identification method based on video frame features. DETAILED DESCRIPTION

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] See also Figure 1 As shown, the video identification method based on video frame features described in this embodiment includes the following steps:

[0047] Step S100, obtaining the identification video and extracting processing features by frame, wherein the processing features include color comparison data, marker value comparison data and portrait data;

[0048] Step S200: Analyze and process the color comparison data and the portrait data to generate a color statistical label;

[0049] Step S300 , analyzing and processing the marker value comparison data and the portrait data to generate a marker value label;

[0050] Step S400 , combining and analyzing the color statistical labels and the marker numerical labels to generate label comparison objects, wherein the label comparison objects include label high difference objects, label medium difference objects, and label low difference objects;

[0051] Step S500 , performing label ratio weight analysis on the label comparison object to generate a real video target, a refined identification target, and a forged video target.

[0052] The present invention is mainly based on the setting of frame images with human portraits in the same background. The identification video is transmitted and obtained by the video owner. The identification video is divided into m identification intervals according to the single frame image, where m is an integer greater than 0.

[0053] The color comparison data includes basic color data and secondary color data, the basic color data is the basic color quantity value, and the secondary color data is the secondary color quantity value;

[0054] The marker value comparison data is specifically marker value data, and the marker value data is the number of markers entered into the marker analysis model;

[0055] Portrait data includes portrait offset value and portrait lens ratio. The portrait offset value is the difference in the horizontal offset displacement of the portrait, and the portrait lens ratio is the proportion of the person in the image.

[0056] It should be noted that: the basic colors include but are not limited to red, yellow and blue, and the secondary colors include but are not limited to dark red, orange, sky blue and lavender;

[0057] The identification video is frame-processed, partitioned and feature extracted, specifically using a video frame image analysis model in machine learning, which includes a color quantity statistical model, a marker quantity statistical model and an offset angle analysis model.

[0058] Taking the marker quantity statistical model as an example, the model establishment steps are described:

[0059] In this model, the marker is the target of action;

[0060] Data collection: Determine the types of markers that need to be identified and classified, and collect the relevant video frame image data to ensure that the dataset contains marker samples of multiple categories and covers different angles, lighting conditions and changes.

[0061] Data preprocessing: Preprocess the collected video frame images, such as image enhancement and resizing, to improve the effect of subsequent feature extraction.

[0062] Feature extraction: Use computer vision techniques to extract key appearance features from preprocessed images.

[0063] Commonly used features include texture features (such as gray-level co-occurrence matrix, local binary pattern) and shape features (such as boundary descriptors, contour features). The appropriate feature extraction method is selected according to the characteristics of the marker.

[0064] Model design: Based on the extracted features, design an appropriate classification or recognition model. You can use machine learning algorithms such as support vector machines or deep learning models such as convolutional neural networks for training and classification. Depending on the task requirements, you can choose a multi-category classification model or a target detection model.

[0065] Model training and evaluation: Use the collected dataset to train the model, divide the dataset into training set and validation set, use appropriate training algorithms and optimization methods to train the model, and use the validation set to evaluate and tune the model to obtain good classification performance and generalization ability.

[0066] Marker recognition and classification: Use the trained model to identify and classify markers in new video frames. Based on the output of the model, each marker is assigned to the corresponding category. Decisions can be made using thresholds or probabilities.

[0067] Marker statistics and output: Based on the identification and classification results, the markers are counted and counted, and statistical information such as the number and proportion of markers in each category can be calculated.

[0068] Relevant statistical information can be output, such as the location information, area or length of each marker.

[0069] The color quantity statistical model changes the target to color according to the above method, while the offset angle analysis model changes the target to the offset position value and the proportion of portrait shots according to the above method.

[0070] The steps to generate color statistics labels include:

[0071] The color statistical label includes a first similarity label, a first slight difference label, and a first height difference label;

[0072] Obtain the basic color data z, secondary color data c, portrait offset value p, and portrait lens ratio v in the frame image of the kth frame of the detection video, and obtain the first color image ratio G(k) in the kth frame through formula analysis. The specific formula is:

[0073]

[0074] in, Always greater than 0, e is a calculation constant;

[0075] The same method is used to obtain the color comparison data and portrait data in the frame image of the k+n frame, where n is the frame number difference, n= , the second color image ratio in the k+n frame is obtained by formula analysis (k+n), by obtaining the first color image ratio G(k) and the second color image ratio The difference between (k+n) is processed by formula to obtain the color statistical difference α. The specific formula of α is:

[0076] α (k+n)- ;

[0077] α>0, substitute the color statistical difference α into the color comparison thresholds α1 and α2 for comparison analysis. The color comparison threshold α1 is less than the color comparison threshold α2. When the color statistical difference α is greater than 0 and less than the color comparison threshold α1, the first similarity label is generated for the frame images between the kth frame and the k+nth frame;

[0078] When the color statistical difference α is greater than the color comparison threshold α1 and less than the color comparison threshold α2, the first slight difference label is generated for the frame image between the kth frame and the k+nth frame; when the color statistical difference α is greater than the color comparison threshold α2, the first high difference label is generated for the frame image between the kth frame and the k+nth frame.

[0079] The steps to generate marker numerical labels are:

[0080] The marker numerical labels include a second similarity label, a second slight difference label, and a second high difference label;

[0081] Obtain the marker numerical data d, portrait offset value p, and portrait shot ratio v in the frame image of the kth frame of the detection video, and obtain the first marker image ratio H(k) in the kth frame through formula analysis. The specific formula is:

[0082]

[0083] in, Always greater than 0, j is the error correction constant;

[0084] The same method is used to obtain the marker numerical comparison data and portrait in the frame image of the k+nth frame, where n is the frame number difference, n= , the second marker image ratio H(k+n) in the k+nth frame is obtained by formula analysis, and the difference between the obtained first marker image ratio H(k) and the second marker image ratio H(k+n) is used to obtain the marker statistical difference β. The specific formula of β is:

[0085] β (k+n)- ;

[0086] β>0, substitute the marker statistical difference β into the marker comparison threshold β1 and β2 for comparison analysis, the marker comparison threshold β1 is less than the marker comparison threshold β2, when the marker statistical difference β is greater than 0 and less than the marker comparison threshold β1, then generate the second similarity label for the frame images between the kth frame and the k+nth frame; when the marker statistical difference β is greater than the marker comparison threshold β1 and less than the marker comparison threshold β2, then generate the second slight difference label for the frame images between the kth frame and the k+nth frame; when the marker statistical difference β is greater than the marker comparison threshold β2, then generate the second high difference label for the frame images between the kth frame and the k+nth frame;

[0087] The combined analysis steps for color statistical labels and marker numerical labels include:

[0088] The label comparison objects include label high difference objects, label medium difference objects and label low difference objects. When the frame images between the kth frame and the k+nth frame simultaneously have the first similar label and the second similar label, the first slight difference label and the second similar label, or the second slight difference label and the first similar label, the label low difference objects are generated for the frame images between the kth frame and the k+nth frame. When the frame images between the kth frame and the k+nth frame simultaneously have the first slight difference label and the second slight difference label, the first similar label and the second high difference label, or the second similar label and the first high difference label, the label medium difference objects are generated for the frame images between the kth frame and the k+nth frame. When the frame images between the kth frame and the k+nth frame simultaneously have the first slight difference label and the second high difference label, the first high difference label and the second slight difference label, or the first high difference label and the second high difference label, the label high difference objects are generated for the frame images between the kth frame and the k+nth frame.

[0089] It should be noted that compared to high-difference objects, low-difference objects indicate that, under the same background, although there is a change in the movement of the person, the overall color and number of markers vary less. This dynamic analysis function enables dynamic analysis and processing of the identification video without human influence.

[0090] A comprehensive analysis model can be used to generate the above-mentioned labels. The algorithm involved in the comprehensive analysis model is set based on color comparison data, marker numerical comparison data and portrait data. The visual features of images and video frames are extracted through multi-layer convolution and pooling operations in the convolutional neural network. The frame difference method is used to compare the differences between adjacent frames and extract dynamically changing features. The distribution of pixel values in the image is statistically analyzed through the image histogram to extract color features. The scale-invariant feature transformation method is used to detect key points in the image and extract local features. Through OpenPose: deep learning methods are used for human posture estimation to extract the position information of key points of the human body. The joint regression method based on deep learning: through models such as convolutional neural networks, the position coordinates of human joints are directly regressed.

[0091] The steps of generating real video targets, refining identification targets and forging video targets include:

[0092] Assume that in m identification intervals, the number of identification intervals containing high-difference objects is x, the number of identification intervals containing medium-difference objects is y, and the number of identification intervals containing low-difference objects is s. The identification factor W is obtained by formulating and analyzing the high-difference objects x, medium-difference objects y, and low-difference objects s using the weight analysis formula. The specific formula is:

[0093] W = + 1.0258

[0094] γ1, γ2, and γ3 are the weight factors of objects with high label difference, objects with medium label difference, and objects with low label difference, respectively. γ1, γ2, and γ3 are all greater than 0.

[0095] Set the identification factor comparison thresholds W1 and W2, and substitute the generated identification factor W into the identification factor comparison thresholds W1 and W2 for analysis, where W1 < W2;

[0096] If the identification factor W is greater than the identification factor comparison threshold W2, then the real video target is generated for the identification video;

[0097] If the identification factor W is greater than the identification factor comparison threshold W1 and less than the identification factor comparison threshold W2, then a refined identification target is generated for the identification video;

[0098] If the identification factor W is greater than 0 and less than the identification factor comparison threshold W1, a forged video target is generated for the identification video.

[0099] It should be noted that the weight algorithm involved is specifically weighted fusion. This method sets weights for different features and performs weighted fusion of each feature according to the weights. This step is also implemented in the comprehensive analysis model mentioned above.

[0100] The present invention uses frame image processing of video frames, combined with a method of color quantity, marker quantity and human motion numerical value, aiming to improve the accuracy and reliability of video authenticity identification by comprehensively utilizing multiple feature information, more comprehensively considering the content and features in the video, thereby increasing the credibility of the identification. Such a method can make full use of the visual information in the video, provide richer and multi-dimensional judgment basis, and help to accurately distinguish real videos from fake videos. At the same time, the method can also provide a more intuitive and explainable way to determine the authenticity of the video, making it more operational and understandable in practical applications.

[0101] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0102] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0103] In this application, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of a, b, or c" can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0104] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0105] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0106] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0107] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0108] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A video identification method based on video frame features, characterized in that: The method comprises the following steps: Step S100, obtaining the identification video and extracting processing features by frame, wherein the processing features include color comparison data, marker value comparison data and portrait data; The color comparison data includes basic color data and secondary color data, the basic color data is the basic color quantity value, and the secondary color data is the secondary color quantity value; The marker value comparison data is specifically marker value data, and the marker value data is the number of markers entered into the marker analysis model; The portrait data includes a portrait offset value and a portrait lens ratio. The portrait offset value is based on the lower neck of the person as a reference point and is specifically a lateral offset displacement difference of the portrait. The portrait lens ratio is the proportion of the person in the image. Step S200: Analyze and process the color comparison data and the portrait data to generate a color statistical label; Step S300 , analyzing and processing the marker value comparison data and the portrait data to generate a marker value label; Step S400 , combining and analyzing the color statistical labels and the marker numerical labels to generate label comparison objects, wherein the label comparison objects include label high difference objects, label medium difference objects, and label low difference objects; Step S500, performing label ratio weight analysis on the label comparison object to generate a real video target, a refined identification target, and a forged video target; The steps of generating real video targets, refining identification targets and forging video targets include: Assume that among m identification intervals, the number of identification intervals containing high-difference objects is x, the number of identification intervals containing medium-difference objects is y, and the number of identification intervals containing low-difference objects is s. Use the weight analysis formula to formally analyze the high-difference objects x, medium-difference objects y, and low-difference objects s to obtain the identification factor W. Set the identification factor comparison thresholds W1 and W2, and substitute the generated identification factor W into the identification factor comparison thresholds W1 and W2 for analysis, where W1 < W2; If the identification factor W is greater than the identification factor comparison threshold W2, then the real video target is generated for the identification video; If the identification factor W is greater than the identification factor comparison threshold W1 and less than the identification factor comparison threshold W2, then a refined identification target is generated for the identification video; If the identification factor W is greater than 0 and less than the identification factor comparison threshold W1, a forged video target is generated for the identification video.

2. The video identification method based on video frame features according to claim 1, characterized in that: The identification video is divided into m identification intervals according to a single frame image, where m is an integer greater than O.

3. The video identification method based on video frame features according to claim 2, characterized in that: The steps to generate color statistics labels include: The color statistical label includes a first similarity label, a first slight difference label, and a first height difference label; Obtain the basic color data z, secondary color data c, portrait offset value p, and portrait lens ratio v in the frame image of the kth frame of the detection video, and obtain the first color image ratio G(k) in the kth frame through formula analysis. Always greater than 0; The same method is used to obtain the color comparison data and portrait data in the frame image of the k+n frame, where n is the frame number difference, n= , the second color image ratio in the k+n frame is obtained by formula analysis (k+n), by obtaining the first color image ratio G(k) and the second color image ratio The difference between (k+n) is processed by formula to obtain the color statistical difference α; α>0, substitute the color statistical difference α into the color comparison thresholds α1 and α2 for comparison analysis. The color comparison threshold α1 is less than the color comparison threshold α2. When the color statistical difference α is greater than 0 and less than the color comparison threshold α1, the first similarity label is generated for the frame images between the kth frame and the k+nth frame; When the color statistical difference α is greater than the color comparison threshold α1 and less than the color comparison threshold α2, the first slight difference label is generated for the frame image between the kth frame and the k+nth frame; when the color statistical difference α is greater than the color comparison threshold α2, the first high difference label is generated for the frame image between the kth frame and the k+nth frame.

4. The video identification method based on video frame features according to claim 3, characterized in that: The steps to generate marker numerical labels are: The marker numerical labels include a second similarity label, a second slight difference label, and a second high difference label; Obtain the marker numerical data d, portrait offset value p, and portrait shot ratio v in the frame image of the k-th frame of the detection video, and obtain the first marker image ratio H(k) in the k-th frame through formula analysis. Always greater than 0; The same method is used to obtain the marker numerical comparison data and portrait in the frame image of the k+nth frame, where n is the frame number difference, n= , formula analysis is performed to obtain the second marker image ratio H(k+n) in the k+nth frame, and the marker statistical difference β is obtained by formula processing the difference between the obtained first marker image ratio H(k) and the second marker image ratio H(k+n), β>0; The marker statistical difference β is substituted into the marker comparison thresholds β1 and β2 for comparison analysis. The marker comparison threshold β1 is less than the marker comparison threshold β2. When the marker statistical difference β is greater than 0 and less than the marker comparison threshold β1, the second similarity label is generated for the frame images between the kth frame and the k+nth frame; when the marker statistical difference β is greater than the marker comparison threshold β1 and less than the marker comparison threshold β2, the second slight difference label is generated for the frame images between the kth frame and the k+nth frame; when the marker statistical difference β is greater than the marker comparison threshold β2, the second high difference label is generated for the frame images between the kth frame and the k+nth frame.

5. The video identification method based on video frame features according to claim 4, characterized in that: The combined analysis steps for color statistical labels and marker numerical labels include: When the frame images between the kth frame and the k+nth frame simultaneously have the first similar label and the second similar label, the first slight difference label and the second similar label, or the second slight difference label and the first similar label, the frame images between the kth frame and the k+nth frame are subjected to label low-difference object generation. When the frame images between the kth frame and the k+nth frame simultaneously have the first slight difference label and the second slight difference label, the first similar label and the second high-difference label, or the second similar label and the first high-difference label, the frame images between the kth frame and the k+nth frame are subjected to label medium-difference object generation. When the frame images between the kth frame and the k+nth frame simultaneously have the first slight difference label and the second high-difference label, the first high-difference label and the second slight difference label, or the first high-difference label and the second high-difference label, the frame images between the kth frame and the k+nth frame are subjected to label high-difference object generation.

Images