Image forgery recognition method, device, computer equipment and storage medium

Abstract

The embodiments of this application belong to the field of artificial intelligence and are applied in the field of image detection. It involves an image forgery recognition method, including pre-training and validation training of a vehicle inspection picture recognition model, and performing verification during the pre-training and validation training phases respectively. Obtain the group of vehicle accident determination pictures uploaded by the current user, input it into the trained vehicle inspection picture recognition model for image forgery recognition, and perform risk control processing on the pictures through the model output results. Use the vehicle inspection picture recognition model to process the photos uploaded by users, and identify whether the picture set includes forged pictures by analyzing the proportion of the main vehicle area in the photos. On the other hand, use the detection method of artificial intelligence to identify forged pictures, reduce the amount of manual review, and improve the efficiency and accuracy of picture review.

Description

Technical Field

[0001] This application relates to the technical fields of artificial intelligence and image detection, and particularly to an image forgery recognition method, device, computer device, and storage medium. Background Art

[0002] The traditional method for auditing vehicle accident pictures is to take on-site photos and conduct manual audits. In recent years, with the development of artificial intelligence technology, the auditing of vehicle accident pictures has gradually been upgraded to users taking and uploading their own photos, and then conducting artificial intelligence audits. However, in this process, there are often users who forge photos of off-site vehicle inspections. Off-site vehicle inspection means that a vehicle is inspected in a place other than the place of purchase. When performing the vehicle inspection task, users are required to upload multiple photos, and each photo needs to contain information about the main vehicle and surrounding environment such as landmark buildings.

[0003] One way of forging is that when taking photos, users deliberately block the environmental information around the photos in other directions and only retain the information of the landmark building in the front-facing photo. And this photo with key information is often forged and altered in various ways. How to analyze the photo set uploaded by users to identify whether the photo set includes forged pictures has become a recognition problem. Summary of the Invention

[0004] The purpose of the embodiments of this application is to propose an image forgery recognition method, device, computer device, and storage medium, so as to analyze the photo set uploaded by users to identify whether the photo set includes forged pictures. On the other hand, use the detection method of artificial intelligence to identify forged pictures, reduce the amount of manual auditing, and improve the efficiency and accuracy of photo auditing.

[0005] To solve the above technical problems, the embodiments of this application provide an image forgery recognition method, which adopts the following technical solutions:

[0006] An image forgery recognition method includes the following steps:

[0007] Collect several pictures containing car images from a public dataset as a training photo set and input it into a vehicle inspection photo recognition model;

[0008] Use the feature extraction sub-model in the vehicle inspection photo recognition model to extract features from the pictures to be inspected in the training photo set, and obtain the feature values corresponding to each pixel point in the pictures to be inspected;

[0009] Based on the feature values and the preset probability map sub-model in the vehicle inspection photo recognition model, perform class prediction on each pixel point in the picture to be inspected, obtain each pixel point with a predicted class of car, and perform binarization processing on each pixel point in the picture to be inspected based on the predicted class;

[0010] Based on the binarization result, calculate the proportion of each pixel point with the predicted category of car in the corresponding picture to be inspected.

[0011] Based on the binarization result and the proportion, perform model verification on the car inspection picture recognition model to obtain the pre-trained car inspection picture recognition model.

[0012] Obtain the group of car accident identification pictures uploaded by the current user, input them into the pre-trained car inspection picture recognition model, and obtain the target proportion of each picture in the group of identification pictures. Among them, the group of identification pictures all includes: at least one accident picture taken from each azimuth of the vehicle's left front, left rear, right front, right rear, and directly in front when identifying a car accident.

[0013] Based on the target proportion and the preset risk control threshold, determine whether there are false pictures in the group of identification pictures.

[0014] Further, the step of performing model verification on the car inspection picture recognition model based on the binarization result and the proportion specifically includes:

[0015] Calculate the actual occupancy ratio of the number of pictures after binarization in the picture to be inspected.

[0016] Judge whether the actual occupancy ratio is less than the preset occupancy ratio.

[0017] If it is less, continue to perform iterative update training on the car inspection picture recognition model until the actual occupancy ratio is greater than or equal to the preset occupancy ratio, and complete the first verification.

[0018] Calculate the error value between the proportion and the preset proportion of its corresponding picture to be inspected.

[0019] Obtain the number of pictures within the allowable error range of the error value, and calculate the actual occupancy ratio of the number of pictures in the picture to be inspected.

[0020] Judge whether the actual occupancy ratio is less than the preset occupancy ratio.

[0021] If it is less, continue to perform iterative update training on the car inspection picture recognition model until the actual occupancy ratio is greater than or equal to the preset occupancy ratio, and complete the second verification.

[0022] Further, after the step of obtaining the pre-trained car inspection picture recognition model, the method further includes:

[0023] Collect several groups of verified picture groups that have completed vehicle accident identification, and successively input each picture in the verified picture groups as a picture to be inspected into the pre-trained vehicle inspection picture recognition model;

[0024] Based on the pre-trained vehicle inspection picture recognition model, perform feature extraction, class prediction, and binarization processing on the picture to be inspected, and obtain the proportion of each pixel point with the predicted class of vehicle in the corresponding picture to be inspected;

[0025] Based on the proportion, perform model verification on the pre-trained vehicle inspection picture recognition model until the verification of the pre-trained model is completed.

[0026] Further, before the step of successively inputting each picture in the verified picture groups as a picture to be inspected into the pre-trained vehicle inspection picture recognition model, the method further includes:

[0027] Carry out group difference numbering on the several groups of verified picture groups that have completed vehicle accident identification;

[0028] After the step of performing feature extraction and class prediction on the picture to be inspected based on the pre-trained vehicle inspection picture recognition model, and obtaining the proportion of each pixel point with the predicted class of vehicle in the corresponding picture to be inspected, the method further includes:

[0029] Group and organize the obtained proportions based on the group difference numbering;

[0030] Obtain the proportions within the same group, and obtain the coefficient of variation of the current picture group based on the proportions;

[0031] Compare the coefficient of variation of each verified picture group with a preset variation threshold to determine the number of successfully verified groups of each verified picture group;

[0032] Calculate the actual occupancy ratio of the number of successfully verified groups of the verified picture groups in the total verified picture groups;

[0033] Use the backpropagation algorithm to perform fitting operations to obtain the error value between the actual occupancy ratio and the preset occupancy ratio;

[0034] Judge whether the error value is within the allowable model prediction error;

[0035] If not, continue to perform iterative update training on the vehicle inspection picture recognition model until the error value is within the allowable model prediction error, and complete the third verification.

[0036] Further, the preset risk control threshold includes a first risk control threshold and a second risk control threshold. The step of determining whether there are fake pictures in the identified picture group based on the target ratio and the preset risk control threshold specifically includes:

[0037] Obtain the coefficient of variation of the current picture group based on the target ratio, and judge the relationship between the coefficient of variation and the first risk control threshold and the second risk control threshold, where the first risk control threshold is greater than the second risk control threshold;

[0038] If the coefficient of variation is greater than the first risk control threshold, there are fake pictures in the identified picture group;

[0039] If the coefficient of variation is greater than the second risk control threshold and less than the first risk control threshold, send the identified picture group for manual review;

[0040] If the coefficient of variation is less than the second risk control threshold, there are no fake pictures in the identified picture group.

[0041] Further, the step of obtaining the coefficient of variation of the current picture group based on the ratio specifically includes:

[0042] Based on the preset algorithm formula: Obtain the coefficient of variation, where std() represents the standard deviation function, mean() represents the mean function, p1, p2, …, p n-1 , p n represents the ratio corresponding to each picture in the identified picture group, and n is a positive integer representing the number of pictures in the identified picture group.

[0043] To solve the above technical problems, an image forgery recognition device is further provided in an embodiment of the present application, which adopts the following technical solutions:

[0044] An image forgery recognition device includes:

[0045] A first input module, configured to collect several pictures containing car images from a public dataset as a training picture set and input it into a vehicle inspection picture recognition model;

[0046] A feature extraction module, configured to use a feature extraction sub-model in the vehicle inspection picture recognition model to extract features of the pictures to be inspected in the training picture set, and obtain the feature values corresponding to each pixel point in the pictures to be inspected;

[0047] A category prediction module, configured to perform category prediction on each pixel point in the picture to be inspected based on the feature value and a preset probability map sub-model in the vehicle inspection picture recognition model, obtain each pixel point with a predicted category of car, and perform binarization processing on each pixel point in the picture to be inspected based on the predicted category;

[0048] A calculation module, configured to calculate, based on the binarization processing result, the proportion of each pixel point with the predicted category of vehicle in the corresponding vehicle inspection picture.

[0049] A pre-training verification module, configured to perform model verification on the vehicle inspection picture recognition model based on the binarization processing result and the proportion, and obtain a pre-trained vehicle inspection picture recognition model.

[0050] A testing module, configured to obtain a group of vehicle accident determination pictures uploaded by the current user, input the pictures into the pre-trained vehicle inspection picture recognition model, and obtain the target proportion of each picture in the determination picture group, where the determination picture group includes at least one accident picture taken from each of the left front, left rear, right front, right rear, and front of the vehicle when determining a vehicle accident.

[0051] A judgment module, configured to determine whether there are fake pictures in the determination picture group based on the target proportion and a preset risk control threshold.

[0052] Furthermore, the image forgery recognition device further includes: a verification training module, and the verification training module includes a second input sub-module, a verification training sub-module, and a verification and calibration sub-module, where

[0053] The second input sub-module is configured to collect several groups of verification picture groups that have completed vehicle accident determination, and sequentially input each picture in the verification picture group into the pre-trained vehicle inspection picture recognition model as a picture to be inspected.

[0054] The verification training sub-module is configured to perform feature extraction, category prediction, and binarization processing on the picture to be inspected based on the pre-trained vehicle inspection picture recognition model, and obtain the proportion of each pixel point with the predicted category of vehicle in the corresponding picture to be inspected.

[0055] The verification and calibration sub-module is configured to perform model verification on the pre-trained vehicle inspection picture recognition model based on the proportion until the verification of the pre-trained model is completed.

[0056] Compared with the prior art, the embodiments of the present application mainly have the following beneficial effects:

[0057] The image forgery recognition method described in the embodiments of the present application pre-trains and validates the vehicle inspection picture recognition model, and performs verification during the pre-training and validation training phases respectively. The vehicle accident determination picture group uploaded by the current user is obtained and input into the trained vehicle inspection picture recognition model for image forgery recognition. The pictures are subject to risk control processing based on the model output results. The vehicle inspection picture recognition model is used to process the photos uploaded by the user. By analyzing the proportion of the area of the main vehicle in the photos, it is identified whether the picture set includes forged pictures. On the other hand, artificial intelligence detection methods are used to identify forged pictures, reducing the amount of manual review and improving the efficiency and accuracy of picture review. Brief Description of the Drawings

[0058] To more clearly illustrate the solutions in the present application, the following will briefly introduce the drawings required for the description of the embodiments of the present application. Obviously, the following-described drawings are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0059] Figure 1 is an exemplary system architecture diagram to which the present application can be applied;

[0060] Figure 2 is a flowchart of an embodiment of the image forgery recognition method according to the present application;

[0061] Figure 3 is Figure 2 a flowchart of a specific implementation of the first verification of the vehicle inspection picture recognition model during the pre-training process in step 205 shown;

[0062] Figure 4 is Figure 2 a flowchart of a specific implementation of the second verification of the vehicle inspection picture recognition model during the pre-training process in step 205 shown;

[0063] Figure 5 is a flowchart of a specific implementation of the verification training of the vehicle inspection picture recognition model that has completed pre-training in the embodiments of the present application;

[0064] Figure 6 is Figure 5 a flowchart of a specific implementation of step 503 shown;

[0065] Figure 7 is Figure 2 a flowchart of a specific implementation of step 207 shown;

[0066] Figure 8 is a schematic structural diagram of an embodiment of the image forgery recognition device according to the present application;

[0067] Figure 9 It is a schematic structural diagram of a specific implementation manner of the verification training module shown in this embodiment;

[0068] Figure 10 It is a schematic structural diagram of an embodiment of a computer device according to the present application. Specific implementation manner

[0069] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs; the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above drawings are intended to cover non-exclusive inclusion. The terms "first", "second", etc. in the specification and claims of this application or the above drawings are used to distinguish different objects and not to describe a specific order.

[0070] Referring to "embodiment" herein means that a specific feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0071] In order to enable those skilled in the art of this technology to better understand the solution of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0072] As Figure 1 shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0073] Users can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications may be installed on the terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0074] The terminal devices 101, 102, and 103 can be various electronic devices with a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, desktop computers, and so on.

[0075] The server 105 can be a server that provides various services, such as a background server that supports the pages displayed on the terminal devices 101, 102, and 103.

[0076] It should be noted that the image forgery recognition method provided by the embodiments of the present application is generally executed by the server / terminal device. Correspondingly, the image forgery recognition device is generally set in the server / terminal device.

[0077] It should be understood that Figure 1 the numbers of the terminal devices, networks, and servers in

[0078] Continue to refer to Figure 2 , which shows a flowchart of an embodiment of the image forgery recognition method according to the present application. The image forgery recognition method includes the following steps:

[0079] Step 201, collect several pictures containing car images from the public dataset as the training atlas and input them into the vehicle inspection picture recognition model.

[0080] In this embodiment, the vehicle inspection picture recognition model is jointly composed of a DeepLab V3 feature extraction sub-model and a fully connected CRFs probability map sub-model.

[0081] In this embodiment, the public dataset can directly use the pictures in the public dataset PASCAL VOC 2012 as the training atlas. PASCAL VOC 2012 contains pictures with 20 labels, which are: person, bird, cat, cow, dog, horse, sheep, airplane, bicycle, boat, bus, car, motorcycle, train, bottle, chair, table, green plant, sofa, TV. Since our goal is to identify cars, only the pictures containing cars are screened, and there are 255 pictures in total. Each picture has a description file that gives the label corresponding to each pixel point in the picture.

[0082] Pre-training is performed using images in the public dataset PASCAL VOC 2012. Since it already comes with car labels, there is no need for developers to manually label the labels of each pixel corresponding to the cars in the images, reducing the upfront labeling work and improving development efficiency.

[0083] Step 202: Use the feature extraction sub-model in the vehicle inspection image recognition model to extract features from the images to be inspected in the training set, and obtain the feature values corresponding to each pixel in the images to be inspected.

[0084] In this embodiment, the DeepLab V3 feature extraction sub-model is used to obtain the feature values of the input image in the form of an artificial neural network. The specific acquisition method includes: after the image is input into the vehicle inspection image recognition model, first, 3 3*3 convolutional layers and 1 max-pooling layer are used to obtain an initial feature map; then, the initial feature map is used as the input and input into the enhancement layer to obtain multi-dimensional features. The enhancement layer contains 5 parallel feature transformations, namely 1 3*3 convolutional layer, 3 atrous convolutional layers, and 1 global average pooling layer, to obtain 5 multi-dimensional feature maps; secondly, these 5 feature maps are fused through the fusion layer, and the concat() function is used for fusion; finally, it is output through the output layer. The output layer contains 1 3*3 convolution for transitioning the fusion result of the concat() function, 1 sampling layer that uses bilinear interpolation to match the size of the output feature map with the input image, and 1 1*1 convolution to ensure that the number of output channels is the same as the number of output feature maps.

[0085] In this embodiment, the enhancement layer can directly adopt the Atrous Spatial Pyramid Pooling (ASPP) module for image-level feature enhancement, that is, the spatial atrous pooling convolutional pyramid, to capture more-dimensional image information and facilitate the extraction of image feature values.

[0086] Step 203: Based on the feature values and the preset probability map sub-model in the vehicle inspection image recognition model, perform class prediction on each pixel in the image to be inspected, obtain the pixels with the predicted class of car, and perform binarization processing on each pixel in the image to be inspected based on the predicted class.

[0087] In this embodiment, after performing class prediction on each pixel in the image to be inspected based on the feature values and the preset probability map sub-model in the vehicle inspection image recognition model and obtaining the pixels with the predicted class of car, the pixels with the predicted class of car are marked with a distinct color, and the pixels with the predicted class not being a car are marked with white.

[0088] By adopting the Conditional Random Fields (CRFs) algorithm, eigenvalue is obtained to predict the pixel points with the category of car in the picture to be tested, and the prediction result is obtained. The picture to be detected is binarized according to the prediction result, which is convenient for obtaining the proportion of each pixel point with the category of car in the corresponding picture to be tested subsequently.

[0089] In the embodiment of the present application, the step of binarizing each pixel point in the picture to be tested according to the predicted category specifically includes: after obtaining the category prediction result of the probability map submodel, setting the pixel points with the predicted category of car and the pixel points of non-car in the picture to be tested as 1 and 0 respectively, or representing them with different colors.

[0090] Step 204: Calculate the proportion of each pixel point with the predicted category of car in the corresponding picture to be tested based on the binarization result.

[0091] In this embodiment, the step of calculating the proportion of each pixel point with the predicted category of car in the corresponding picture to be tested based on the binarization result specifically includes: obtaining the ratio of the number of pixel points with the predicted category of car to the number of all pixel points in the picture to be tested, and using the ratio as the proportion; or calculating the area ratio of the different colors after binarization in the picture to be tested, and using the area ratio as the proportion.

[0092] Step 205: Based on the binarization result and the proportion, perform model verification on the vehicle inspection picture recognition model to obtain the pre-trained vehicle inspection picture recognition model.

[0093] In this embodiment, the step of performing model verification on the vehicle inspection picture recognition model based on the binarization result and the proportion includes the first verification and the second verification.

[0094] Continue to refer to Figure 3 , Figure 3 Yes Figure 2 is a flowchart of a specific implementation manner of the first verification of the vehicle inspection picture recognition model during the pre-training process in step 205, including the steps:

[0095] Step 301: Calculate the actual occupancy ratio of the number of binarized pictures in the picture to be tested.

[0096] Step 302: Determine whether the actual occupancy ratio is less than the preset occupancy ratio.

[0097] Step 303: If so, continue to perform iterative update training on the vehicle inspection picture recognition model until the actual occupancy ratio is greater than or equal to the preset occupancy ratio, and complete the first verification.

[0098] Step 304, if not, the first verification is successful.

[0099] By adopting the first verification method, the vehicle inspection picture recognition model is optimized during the pre-training process, and the applicability of the model after the pre-training is completed is improved.

[0100] Continue to refer to Figure 4 , Figure 4 Yes Figure 2 is a flowchart of a specific implementation manner of the second verification of the vehicle inspection picture recognition model during the pre-training process in step 205 shown in

[0101] Step 401, calculate the error value between the ratio and the preset ratio of the corresponding picture to be inspected;

[0102] Step 402, obtain the number of pictures whose error values are within the allowable error range, and calculate the actual occupancy ratio of the number of pictures in the pictures to be inspected;

[0103] Step 403, determine whether the actual occupancy ratio is less than the preset occupancy ratio;

[0104] Step 404, if so, continue to perform iterative update training on the vehicle inspection picture recognition model until the actual occupancy ratio is greater than or equal to the preset occupancy ratio, and the second verification is completed;

[0105] Step 405, if not, the second verification is successful.

[0106] By adopting the second verification method, the vehicle inspection picture recognition model is optimized during the pre-training process, and the applicability of the model after the pre-training is completed is further improved.

[0107] In this embodiment, after the step of obtaining the pre-trained vehicle inspection picture recognition model, the method further includes: collecting several groups of verification picture groups that have completed vehicle accident identification, and sequentially using each picture in the verification picture group as a picture to be inspected and inputting it into the pre-trained vehicle inspection picture recognition model;

[0108] Based on the pre-trained vehicle inspection picture recognition model, feature extraction, class prediction and binarization processing are performed on the pictures to be inspected, and the ratio of the pixel points of each predicted class as a vehicle to the corresponding picture to be inspected is obtained; based on the ratio, the pre-trained vehicle inspection picture recognition model is model-verified until the verification of the pre-trained model is completed.

[0109] Continue to refer to Figure 5 , Figure 5 is a flowchart of a specific implementation manner of the verification training of the pre-trained vehicle inspection picture recognition model in the embodiment of the present application, including steps:

[0110] Step 501: Collect several groups of verification picture groups that have completed vehicle accident identification, and sequentially input each picture in the verification picture groups as a picture to be inspected into the pre-trained vehicle inspection picture recognition model.

[0111] Step 502: Based on the pre-trained vehicle inspection picture recognition model, perform feature extraction, class prediction, and binarization processing on the picture to be inspected, and obtain the proportion of each pixel point with the predicted class of vehicle in the corresponding picture to be inspected.

[0112] Step 503: Based on the proportion, perform model verification on the pre-trained vehicle inspection picture recognition model until the verification of the pre-trained model is completed.

[0113] In this embodiment, before the step of sequentially inputting each picture in the verification picture groups as a picture to be inspected into the pre-trained vehicle inspection picture recognition model, the method further includes: performing group difference numbering on the several groups of verification picture groups that have completed vehicle accident identification.

[0114] In this embodiment, after the step of performing feature extraction and class prediction on the picture to be inspected based on the pre-trained vehicle inspection picture recognition model, and obtaining the proportion of each pixel point with the predicted class of vehicle in the corresponding picture to be inspected, the method further includes: performing grouped arrangement on the obtained proportion based on the group difference numbering; obtaining the proportion within the same group, and obtaining the coefficient of variation of the current picture group based on the proportion; comparing the coefficient of variation of each verification picture group with a preset coefficient of variation threshold to determine the number of successfully verified groups of each verification picture group; calculating the actual occupancy ratio of the number of successfully verified groups of the verification picture groups in the total verification picture groups; performing fitting operation using the backpropagation algorithm to obtain the error value between the actual occupancy ratio and the preset occupancy ratio; determining whether the error value is within the allowable model prediction error; if not, continue to perform iterative update training on the vehicle inspection picture recognition model until the error value is within the allowable model prediction error, and complete the third verification.

[0115] Continue to refer to Figure 6 , Figure 6 Yes Figure 5 is a flowchart of a specific implementation manner of step 503 shown in

[0116] Step 601: Perform group difference numbering on the several groups of verification picture groups that have completed vehicle accident identification;

[0117] Step 602: Perform grouped arrangement on the obtained proportion based on the group difference numbering;

[0118] Step 603: Obtain the ratio within the same group, and obtain the coefficient of variation of the current picture group based on the ratio.

[0119] Step 604: Compare the coefficient of variation of each verification picture group with a preset variation threshold to determine the number of successfully verified groups in each verification picture group.

[0120] Step 605: Calculate the actual occupancy ratio of the number of successfully verified groups in the verification picture group in the total verification picture groups.

[0121] Step 606: Use the backpropagation algorithm to perform fitting operations to obtain the error value between the actual occupancy ratio and the preset occupancy ratio.

[0122] Step 607: Determine whether the error value is within the allowable model prediction error.

[0123] Step 608: If not, continue to iteratively update and train the vehicle inspection picture recognition model until the error value is within the allowable model prediction error, and complete the third verification.

[0124] Step 609: If so, the third verification is successful.

[0125] By adopting the method of the third verification, the vehicle inspection picture recognition model is verified and optimized after pre-training, and the applicability of the model after pre-training is improved again.

[0126] In this embodiment, the verification picture groups are selected from the pictures obtained on the property insurance company's own APP. Since the real pictures and forged pictures in the known pictures are used, the pictures obtained on the property insurance company's own APP, the real pictures and forged pictures in the known pictures are used as the verification picture set for verification, and the artificial screening is avoided by using the artificial intelligence method for reverse fitting, reducing the workload of the verification personnel.

[0127] In this embodiment, the step of obtaining the coefficient of variation of the current picture group based on the ratio specifically includes: based on a preset algorithm formula: Obtain the coefficient of variation, where std() represents the standard deviation function, mean() represents the mean function, p1, p2, …, p n-1 , p n represent the ratios corresponding to each picture in the verification picture group, and n is a positive integer representing the number of pictures in the verification picture group.

[0128] Step 206: Obtain the group of vehicle accident determination pictures uploaded by the current user, and input them into the pre-trained vehicle inspection picture recognition model to obtain the target ratio of each picture in the determination picture group, where the determination picture group includes at least one accident picture taken from each of the front left, rear left, front right, rear right, and directly in front of the vehicle when determining a vehicle accident.

[0129] Step 207: Determine whether there are false pictures in the identified picture group based on the target ratio and the preset risk control threshold.

[0130] In this embodiment, the preset risk control threshold includes a first risk control threshold and a second risk control threshold. The step of determining whether there are false pictures in the identified picture group based on the target ratio and the preset risk control threshold specifically includes: obtaining the coefficient of variation of the current picture group based on the target ratio, and judging the relationship between the coefficient of variation and the first risk control threshold and the second risk control threshold, where the first risk control threshold is greater than the second risk control threshold; if the coefficient of variation is greater than the first risk control threshold, there are false pictures in the identified picture group; if the coefficient of variation is greater than the second risk control threshold and less than the first risk control threshold, send the identified picture group for manual review; if the coefficient of variation is less than the second risk control threshold, there are no false pictures in the identified picture group.

[0131] Continue to refer to Figure 7 , Figure 7 Yes Figure 2 The flowchart of a specific implementation manner of step 207 shown in

[0132] Step 701: Obtain the coefficient of variation of the current picture group based on the target ratio, and judge the relationship between the coefficient of variation and the first risk control threshold and the second risk control threshold, where the first risk control threshold is greater than the second risk control threshold;

[0133] Step 702: If the coefficient of variation is greater than the first risk control threshold, there are false pictures in the identified picture group;

[0134] Step 703: If the coefficient of variation is greater than the second risk control threshold and less than the first risk control threshold, send the identified picture group for manual review;

[0135] Step 704: If the coefficient of variation is less than the second risk control threshold, there are no false pictures in the identified picture group.

[0136] During testing, since the authenticity of the unknown pictures is unknown, two risk control thresholds are introduced. If the coefficient of variation is greater than the first risk control threshold, it is directly determined that there are forged pictures in the test picture set. If the coefficient of variation is greater than the second risk control threshold and less than the first risk control threshold, risk control management is carried out and sent for manual review, ensuring more prudent review.

[0137] In this embodiment, the step of obtaining the coefficient of variation of the current picture group based on the ratio specifically includes: based on the preset algorithm formula: Obtain the coefficient of variation, where std() represents the standard deviation function, mean() represents the mean function, p1, p2, …, p n-1 , p n represents the ratio corresponding to each picture in the identified picture group, and n is a positive integer representing the number of pictures in the identified picture group.

[0138] In the image forgery recognition method according to the embodiment of the present application, through pre-training and verification training of the vehicle inspection picture recognition model, and performing verification respectively in the pre-training and verification training stages, obtain the vehicle accident identification picture group uploaded by the current user, input it into the trained vehicle inspection picture recognition model, perform image forgery recognition, and perform risk control processing on the pictures through the model output results. Use the vehicle inspection picture recognition model to process the photos uploaded by the user, and identify whether there are forged pictures in the picture set by analyzing the proportion of the main vehicle area in the photos. On the other hand, use the detection method of artificial intelligence to identify forged pictures, reduce the amount of manual review, and improve the efficiency and accuracy of picture review.

[0139] The embodiment of the present application can obtain and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results of theory, method, technology and application system.

[0140] Artificial intelligence basic technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, robotics, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0141] For example, in the embodiment of the present application, deep convolutional neural network and atrous spatial pyramid pooling in artificial intelligence are used for image processing.

[0142] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through computer-readable instructions. The computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (ROM), etc., or a random access memory (RAM), etc.

[0143] It should be understood that although the steps in the flowchart of the accompanying drawings are shown sequentially according to the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless otherwise clearly stated in this document, there is no strict order restriction for the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps.

[0144] Further referring to Figure 8 , as an implementation of the method shown above Figure 2 , this application provides an embodiment of an image forgery recognition device. This device embodiment corresponds to Figure 2 the method embodiment shown, and this device can be specifically applied to various electronic devices.

[0145] As shown in Figure 8 , the image forgery recognition device 800 described in this embodiment includes: a first input module 801, a feature extraction module 802, a class prediction module 803, a calculation module 804, a pre-training verification module 805, a test module 806, and a judgment module 807. Among them:

[0146] The first input module 801 is used to collect several pictures containing car images from a public dataset as a training atlas and input it into the vehicle inspection fake picture recognition model;

[0147] The feature extraction module 802 is used to extract features from the pictures to be inspected in the training atlas by using the feature extraction sub-model in the vehicle inspection fake picture recognition model to obtain the feature values corresponding to each pixel point in the pictures to be inspected;

[0148] The class prediction module 803 is used to perform class prediction on each pixel point in the pictures to be inspected based on the feature values and the preset probability map sub-model in the vehicle inspection fake picture recognition model, obtain the pixel points with the predicted class as a car, and perform binarization processing on each pixel point in the pictures to be inspected based on the predicted class;

[0149] The calculation module 804 is used to calculate the proportion of each pixel point with the predicted class as a car in the corresponding pictures to be inspected based on the binarization processing result;

[0150] The pre-training verification module 805 is used to perform model verification on the vehicle inspection fake picture recognition model based on the binarization processing result and the proportion to obtain the pre-trained vehicle inspection fake picture recognition model;

[0151] A test module 806, configured to obtain a set of pictures for identifying a car accident uploaded by a current user, input the set of pictures into the pre-trained model for identifying forged pictures of vehicle inspection, and obtain a target ratio of each picture in the set of pictures for identification, where the set of pictures for identification all includes: at least one accident picture taken from each of the left front, left rear, right front, right rear, and front of the vehicle when identifying a car accident;

[0152] A judgment module 807, configured to determine whether there are forged pictures in the set of pictures for identification based on the target ratio and a preset risk control threshold.

[0153] The image forgery identification device according to the embodiment of the present application obtains a set of pictures for identifying a car accident uploaded by a current user, inputs the set of pictures into a trained model for identifying pictures of vehicle inspection to perform image forgery identification, and performs risk control processing on the pictures through the output result of the model. The pictures uploaded by the user are processed by the model for identifying pictures of vehicle inspection, and by analyzing the proportion of the area of the main vehicle in the pictures, it is identified whether the picture set includes forged pictures. On the other hand, an artificial intelligence detection method is used to identify forged pictures, reducing the amount of manual review and improving the efficiency and accuracy of picture review.

[0154] Refer to Figure 9 , the image forgery identification device 800 described in this embodiment further includes: a verification and training module 808, Figure 9 is a structural diagram of a specific implementation manner of the verification and training module. The verification and training module 808 includes a second input sub-module 8081, a verification and training sub-module 8082, and a verification and calibration sub-module 8083, where:

[0155] Among them, the second input sub-module 8081 is configured to collect several sets of verification picture sets that have completed the identification of car accidents, and sequentially input each picture in the verification picture set into the pre-trained model for identifying forged pictures of vehicle inspection as a picture to be inspected;

[0156] The verification and training sub-module 8082 is configured to perform feature extraction, class prediction, and binarization processing on the picture to be inspected based on the pre-trained model for identifying forged pictures of vehicle inspection, and obtain the ratio of each pixel point with the predicted class as a car to the corresponding picture to be inspected;

[0157] The verification and calibration sub-module 8083 is configured to perform model calibration on the pre-trained model for identifying forged pictures of vehicle inspection based on the ratio until the calibration of the pre-trained model is completed.

[0158] The verification and training module in the image forgery identification device according to the embodiment of the present application further ensures the applicability and accuracy of the model by performing verification training and calibration on the pre-trained model for identifying pictures of vehicle inspection.

[0159] To solve the above technical problems, the embodiments of the present application further provide a computer device. Specifically, please refer to Figure 10 , Figure 10 which is the basic structural block diagram of the computer device in this embodiment.

[0160] The computer device 9 includes a memory 91, a processor 92, and a network interface 93 that are communicatively connected to each other through a system bus. It should be noted that only the computer device 9 with components 91-93 is shown in the figure, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented. Among them, those skilled in the art of the present technology can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0161] The computer device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The computer device can perform human-computer interaction with the user through a keyboard, a mouse, a remote control, a touchpad, or a voice control device, etc.

[0162] The memory 91 includes at least one type of readable storage medium, which includes flash memory, hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 91 may be an internal storage unit of the computer device 9, such as the hard disk or memory of the computer device 9. In other embodiments, the memory 91 may also be an external storage device of the computer device 9, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, FlashCard, etc. equipped on the computer device 9. Of course, the memory 91 may also include both the internal storage unit and the external storage device of the computer device 9. In this embodiment, the memory 91 is generally used to store the operating system and various application software installed on the computer device 9, such as computer-readable instructions of the image forgery recognition method. In addition, the memory 91 may also be used to temporarily store various data that have been output or will be output.

[0163] In some embodiments, the processor 92 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips. The processor 92 is generally used to control the overall operation of the computer device 9. In this embodiment, the processor 92 is used to run the computer-readable instructions stored in the memory 91 or process data, such as running the computer-readable instructions of the image forgery recognition method.

[0164] The network interface 93 may include a wireless network interface or a wired network interface, and this network interface 93 is generally used to establish a communication connection between the computer device 9 and other electronic devices.

[0165] The computer device proposed in this embodiment belongs to the field of artificial intelligence technology. In this application, the vehicle inspection picture recognition model is pre-trained and verified, and checks are carried out respectively in the pre-training and verification training stages. The vehicle accident recognition picture group uploaded by the current user is obtained and input into the trained vehicle inspection picture recognition model for picture forgery recognition, and risk control processing is carried out on the pictures through the model output results. The vehicle inspection picture recognition model is used to process the photos uploaded by the user. By analyzing the proportion of the main vehicle area in the photos, it is identified whether the picture set includes forged pictures. On the other hand, the detection method of artificial intelligence is used to identify forged pictures, reducing the amount of manual review and improving the efficiency and accuracy of picture review.

[0166] This application also provides another implementation manner, that is, to provide a computer-readable storage medium, the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by a processor to enable the processor to execute the steps of the image forgery recognition method as described above.

[0167] The computer-readable storage medium proposed in this embodiment belongs to the field of artificial intelligence technology. In this application, the vehicle inspection picture recognition model is pre-trained and verified, and checks are carried out respectively in the pre-training and verification training stages. The vehicle accident recognition picture group uploaded by the current user is obtained and input into the trained vehicle inspection picture recognition model for picture forgery recognition, and risk control processing is carried out on the pictures through the model output results. The vehicle inspection picture recognition model is used to process the photos uploaded by the user. By analyzing the proportion of the main vehicle area in the photos, it is identified whether the picture set includes forged pictures. On the other hand, the detection method of artificial intelligence is used to identify forged pictures, reducing the amount of manual review and improving the efficiency and accuracy of picture review.

[0168] Through the description of the above implementation manners, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation manner. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disc), and includes several instructions to enable a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of this application.

[0169] Obviously, the embodiments described above are only a part of the embodiments of this application, rather than all of them. The preferred embodiments of this application are shown in the accompanying drawings, but they do not limit the patent scope of this application. This application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosed content of this application more thorough and comprehensive. Although this application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions described in the foregoing specific embodiments, or perform equivalent replacements on some of the technical features. Any equivalent structure that makes use of the content of the specification and drawings of this application, directly or indirectly applied in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. An image forgery recognition method, characterized in that, Including the following steps: Collect several pictures containing car images from a public dataset as a training picture set and input it into the vehicle inspection picture recognition model; Use the feature extraction sub-model in the vehicle inspection picture recognition model to extract features from the pictures to be inspected in the training picture set, and obtain the feature values corresponding to each pixel point in the pictures to be inspected; Based on the feature values and the preset probability map sub-model in the vehicle inspection picture recognition model, perform class prediction on each pixel point in the pictures to be inspected, obtain the pixel points with the predicted class as cars, and perform binarization processing on each pixel point in the pictures to be inspected based on the predicted class; Based on the binarization processing result, calculate the proportion of each pixel point with the predicted class as cars in the corresponding pictures to be inspected; Based on the binarization processing result and the proportion, perform model verification on the vehicle inspection picture recognition model to obtain the pre-trained vehicle inspection picture recognition model; Obtain the group of car accident determination pictures uploaded by the current user and input it into the pre-trained vehicle inspection picture recognition model to obtain the target proportion of each picture in the group of determination pictures. Among them, the group of determination pictures all includes: when determining a car accident, at least one accident picture taken from each of the left front, left rear, right front, right rear, and front of the vehicle; Based on the target proportion and the preset risk control thresholds, determine whether there are false pictures in the group of determination pictures. Among them, the preset risk control thresholds include a first risk control threshold and a second risk control threshold. The step of determining whether there are false pictures in the group of determination pictures based on the target proportion and the preset risk control thresholds specifically includes: Obtain the coefficient of variation of the current picture group based on the target proportion, and judge the relationship between the coefficient of variation and the first risk control threshold and the second risk control threshold. Among them, the first risk control threshold is greater than the second risk control threshold. The step of obtaining the coefficient of variation of the current picture group based on the target proportion specifically includes: Based on a preset algorithm formula: , obtain the coefficient of variation, where represents the standard deviation function, represents the mean function, represents the ratio corresponding to each picture in the identified picture group, is a positive integer representing the number of pictures in the identified picture group; If the coefficient of variation is greater than the first risk control threshold, there are false pictures in the group of determination pictures; If the coefficient of variation is greater than the second risk control threshold and less than the first risk control threshold, send the group of determination pictures to manual review; If the coefficient of variation is less than the second risk control threshold, there are no false pictures in the group of determination pictures.

2. The image forgery recognition method according to claim 1, wherein The step of performing model verification on the vehicle inspection picture recognition model based on the binarization processing result and the proportion specifically includes: Calculate the actual occupancy ratio of the number of pictures processed by binarization in the pictures to be inspected; Judge whether the actual occupancy ratio is less than the preset occupancy ratio; If it is less, continue to perform iterative update training on the vehicle inspection picture recognition model until the actual occupancy ratio is greater than or equal to the preset occupancy ratio to complete the first verification; Calculate the error value between the proportion and the preset proportion of its corresponding pictures to be inspected; Obtain the number of pictures within the allowable error range of the error value, and calculate the actual occupancy ratio of the number of pictures in the pictures to be inspected; Judge whether the actual occupancy ratio is less than the preset occupancy ratio; If it is less than, continue to iteratively update and train the vehicle inspection image recognition model until the actual occupancy ratio is greater than or equal to the preset occupancy ratio, and complete the second verification.

3. The image forgery recognition method according to claim 1, wherein After the step of obtaining the pre-trained vehicle inspection image recognition model, the method further includes: Collect several groups of verification image groups that have completed vehicle accident identification, and sequentially input each image in the verification image group as a test image into the pre-trained vehicle inspection image recognition model; Based on the pre-trained vehicle inspection image recognition model, perform feature extraction, class prediction, and binarization processing on the test image to obtain the proportion of each pixel point with the predicted class of vehicle in the corresponding test image; Based on the proportion, perform model verification on the pre-trained vehicle inspection image recognition model until the verification of the pre-trained model is completed.

4. The image forgery recognition method according to claim 3, wherein, Before the step of sequentially inputting each image in the verification image group as a test image into the pre-trained vehicle inspection image recognition model, the method further includes: Carry out group difference numbering on the several groups of verification image groups that have completed vehicle accident identification; After the step of performing feature extraction and class prediction on the test image based on the pre-trained vehicle inspection image recognition model to obtain the proportion of each pixel point with the predicted class of vehicle in the corresponding test image, the method further includes: Group and organize the obtained proportions based on the group difference numbering; Obtain the proportions within the same group, and based on the proportions, obtain the coefficient of variation of the current image group; Compare the coefficient of variation of each verification image group with a preset variation threshold to determine the number of successfully verified groups of each verification image group; Calculate the actual occupancy ratio of the number of successfully verified groups of the verification image group in the total verification image group; Use the backpropagation algorithm to perform fitting operations to obtain the error value between the actual occupancy ratio and the preset occupancy ratio; Determine whether the error value is within the allowable model prediction error; If not, continue to iteratively update and train the vehicle inspection image recognition model until the error value is within the allowable model prediction error, and complete the third verification.

5. An image forgery recognition device, characterized in that, The image forgery recognition device is used to implement the steps of the image forgery recognition method according to any one of claims 1 to 4. The image forgery recognition device includes: A first input module, configured to collect several pictures containing vehicle images from a public dataset as a training atlas and input them into a vehicle inspection image recognition model; A feature extraction module, configured to use the feature extraction sub-model in the vehicle inspection image recognition model to perform feature extraction on the test images in the training atlas to obtain the feature value corresponding to each pixel point in the test images; A class prediction module, configured to perform class prediction on each pixel point in the test image based on the feature value and the preset probability map sub-model in the vehicle inspection image recognition model to obtain each pixel point with the predicted class of vehicle, and perform binarization processing on each pixel point in the test image based on the predicted class; A calculation module, configured to calculate the proportion of each pixel point with the predicted class of vehicle in the corresponding test image based on the binarization processing result; A pre-training verification module, configured to perform model verification on the vehicle inspection image recognition model based on the binarization processing result and the ratio, and obtain a pre-trained vehicle inspection image recognition model; A test module, configured to obtain a group of vehicle accident determination pictures uploaded by a current user, input the group of pictures into the pre-trained vehicle inspection image recognition model, and obtain the target ratio of each picture in the group of determination pictures, where the group of determination pictures all include: at least one accident picture taken from each azimuth of the vehicle's left front, left rear, right front, right rear, and due front when determining a vehicle accident; A judgment module, configured to determine whether there are fake pictures in the group of determination pictures based on the target ratio and a preset risk control threshold, where the preset risk control threshold includes a first risk control threshold and a second risk control threshold, and the step of determining whether there are fake pictures in the group of determination pictures based on the target ratio and the preset risk control threshold specifically includes: Obtaining the coefficient of variation of the current picture group based on the target ratio, and judging the relationship between the coefficient of variation and the first risk control threshold and the second risk control threshold, where the first risk control threshold is greater than the second risk control threshold, and the step of obtaining the coefficient of variation of the current picture group based on the target ratio specifically includes: Based on a preset algorithm formula: , the coefficient of variation is obtained, where represents the standard deviation function, represents the mean function, represents the ratio corresponding to each picture in the identified picture group, is a positive integer representing the number of pictures in the identified picture group; If the coefficient of variation is greater than the first risk control threshold, there are fake pictures in the group of determination pictures; If the coefficient of variation is greater than the second risk control threshold and less than the first risk control threshold, send the group of determination pictures for manual review; If the coefficient of variation is less than the second risk control threshold, there are no fake pictures in the group of determination pictures.

6. The image forgery recognition device according to claim 5, characterized in that The image forgery recognition device further includes: a verification training module, and the verification training module includes a second input sub-module, a verification training sub-module, and a verification and calibration sub-module, where, The second input sub-module is configured to collect several groups of verification picture groups that have completed vehicle accident determination, and sequentially input each picture in the verification picture group into the pre-trained vehicle inspection image recognition model as a picture to be inspected; The verification training sub-module is configured to perform feature extraction, class prediction, and binarization processing on the picture to be inspected based on the pre-trained vehicle inspection image recognition model, and obtain the ratio of the pixel points with the predicted class of vehicle to the corresponding picture to be inspected; The verification and calibration sub-module is configured to perform model verification on the pre-trained vehicle inspection image recognition model based on the ratio until the verification of the pre-trained model is completed.

7. A computer device, including a memory and a processor, where computer-readable instructions are stored in the memory, and when the processor executes the computer-readable instructions, the steps of the image forgery recognition method according to any one of claims 1 to 4 are implemented.

8. A computer-readable storage medium, characterized in that, Computer-readable instructions are stored on a computer-readable storage medium, and when the computer-readable instructions are executed by a processor, the steps of the image forgery recognition method according to any one of claims 1 to 4 are implemented.

Images