Image processing method and device, storage medium and electronic equipment

Abstract

The invention discloses an image processing method and device, a storage medium and electronic equipment, and relates to the field of financial science and technology. The method comprises the following steps: acquiring a target image in a transaction process; segmenting the target image into L image areas, and determining M candidate areas from the L image areas; extracting N-dimensional features of each candidate region, and forming an N * M-dimensional feature matrix by using all the features of the M candidate regions; determining the probability that each candidate region in the M candidate regions belongs to K image types according to the feature matrix through K classifiers; and determining a target type of the target image according to the probability that each candidate area belongs to the K image types, and determining a transaction result according to the target type of the target image. The technical problem that in the prior art, due to the fact that the complexity of the image authenticity verification process of the transaction process is high, the image auditing speed of the transaction process is low, and the transaction auditing efficiency is low is solved.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and specifically, to an image processing method, device, storage medium and electronic device. Background Art

[0002] With the acceleration of scientific and technological progress and the rapid development of information technology, images and videos have become an indispensable part of daily life. However, with the continuous development of image processing technology and AI generation technology, text information tampering and image information tampering have become more and more common. For example, AI generation models can be used to create false images to forge or spread false information, posing a huge threat to personal and property safety.

[0003] At the same time, financial services are often accompanied by a large amount of image information, such as identity information and certification materials uploaded by customers, which are all at risk of false information and difficult to identify manually. How to quickly and accurately identify key information and the authenticity of information is the focus and difficulty of transaction review.

[0004] In the existing business process, users need to upload image material information or video material information to perform business operations. Traders log in to the corresponding system, view the image or video materials, obtain key information from them, perform identification and information verification, and complete the review steps. However, the image authenticity verification process in the existing business process is highly complex, and manual processing has great limitations. The authenticity of the image information cannot be identified, and the image review speed of the transaction process is slow, resulting in technical problems such as low transaction review efficiency.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] The main purpose of the present application is to provide an image processing method, device, storage medium and electronic device, so as to at least solve the technical problem in the prior art that the image review speed of the transaction process is slow due to the high complexity of the image authenticity verification process of the transaction process, thereby causing low transaction review efficiency.

[0007] In order to achieve the above-mentioned purpose, according to one aspect of an embodiment of the present application, an image processing method is provided, including: acquiring a target image in a transaction process; dividing the target image into L image areas, and determining M candidate areas from the L image areas, wherein L is an integer greater than 1, M is an integer less than or equal to L and greater than 1, and the probability that the M candidate areas include the same object is greater than a preset threshold; extracting N-dimensional features of each candidate area, and forming all features of the M candidate areas into a feature matrix of N*M dimensions, wherein N is an integer greater than 1; determining the probability that each of the M candidate areas belongs to K image types according to the feature matrix through K classifiers, wherein K is an integer greater than 1, and the K classifiers correspond one-to-one to the K image types, and the image type is used to distinguish whether the candidate area is a real shot image; determining the target type of the target image according to the probability that each candidate area belongs to the K image types, and determining the transaction result according to the target type of the target image.

[0008] Optionally, the probability that each of the M candidate areas belongs to the K image types is determined based on the feature matrix by K classifiers, including: multiplying the feature matrix of N*M dimensions by the weight matrix composed of K classifiers to obtain a probability distribution matrix of N*K dimensions, wherein the weight matrix is ​​a matrix of M*K dimensions; and representing the probability that each of the M candidate areas belongs to the K image types by the probability distribution matrix.

[0009] Optionally, the target type of the target image is determined according to the probability that each candidate area belongs to K image types, including: calculating the intersection-and-union ratio of any two candidate areas among the M candidate areas, wherein the intersection-and-union ratio of any two candidate areas represents the ratio of the overlapping area area of ​​the two candidate areas to the total area of ​​the two candidate areas; taking the candidate area with the highest probability of belonging to the i-th image type as the first area corresponding to the i-th image type, and taking the candidate area among the M candidate areas whose intersection-and-union ratio with the first area is less than a first threshold as the second area corresponding to the i-th image type, wherein i is a positive integer less than or equal to K; and determining the target type of the target image according to the first area and the second area corresponding to each image type.

[0010] Optionally, the target type of the target image is determined according to the first area and the second area corresponding to each image type, including: determining the offset and scaling ratio of the first area in the target direction, and the offset and scaling ratio of the second area in the target direction through a regressor; adjusting the first area to a first target area according to the offset and scaling ratio of the first area in the target direction; adjusting the second area to a second target area according to the offset and scaling ratio of the second area in the target direction; and determining the target type of the target image according to the first target area and the second target area.

[0011] Optionally, determining the target type of the target image based on the first target area and the second target area includes: determining the probability that the target image belongs to each image type among K image types based on the probability value of the first target area corresponding to each image type belonging to the image type, the number of second target areas corresponding to each image type, and the probability value of each second target area belonging to the image type; and taking the image type to which the target image belongs with the highest probability as the target type of the target image.

[0012] Optionally, obtaining a target image in the transaction process includes: obtaining user data and transaction data involved in the transaction process from a target file by accessing a target storage path; and using an image in the user data and transaction data that needs to be authenticated as the target image.

[0013] Optionally, the training process of the classifier includes the following steps: obtaining training images of K image types, wherein the training images of the K image types include at least images taken actually in different business scenarios and images taken not actually; dividing the training images into a plurality of sub-region images; extracting image features of each sub-region image; and iteratively training a neural network according to the image features of each sub-region image of the training images of each image type to obtain a classifier corresponding to the image type.

[0014] Optionally, extracting image features of each sub-region image includes: performing feature annotation on abnormal sub-image regions of training images that are not actually shot, wherein the abnormal sub-image regions are image regions constructed using an image synthesis method, and the feature annotation is used to generate a feature file including coordinate information, a synthesis method, and a saving path of the abnormal sub-image regions.

[0015] In order to achieve the above-mentioned purpose, according to another aspect of an embodiment of the present application, an image processing device is further provided, including: an acquisition unit, which acquires a target image in a transaction process; a processing unit, which divides the target image into L image regions, and determines M candidate regions from the L image regions, wherein L is an integer greater than 1, M is an integer less than or equal to L and greater than 1, and the probability that the M candidate regions include the same object is greater than a preset threshold; an extraction unit, which extracts N-dimensional features of each candidate region, and combines all features of the M candidate regions into a feature matrix of N*M dimensions, wherein N is an integer greater than 1; a first determination unit, which determines the probability that each of the M candidate regions belongs to K image types according to the feature matrix through K classifiers, wherein K is an integer greater than 1, the K classifiers correspond one-to-one to the K image types, and the image type is used to distinguish whether the candidate region is a real shot image; a second determination unit, which determines the target type of the target image according to the probability that each candidate region belongs to the K image types, and determines the transaction result according to the target type of the target image.

[0016] According to another aspect of an embodiment of the present application, there is further provided an electronic device, comprising: a memory storing an executable program; and a processor for running the program, wherein the above-mentioned image processing method is executed when the program is running.

[0017] According to another aspect of an embodiment of the present application, a computer-readable storage medium is further provided, the computer-readable storage medium including a stored executable program, wherein when the executable program runs, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned image processing method.

[0018] According to another aspect of an embodiment of the present application, a computer program product is also provided, including computer instructions, which implement the steps of the above-mentioned image processing method when executed by a processor.

[0019] In an embodiment of the present application, a target image in a transaction process is first acquired, and then the target image is divided into L image regions, and M candidate regions are determined from the L image regions, wherein L is an integer greater than 1, and M is an integer less than or equal to L and greater than 1, and the probability that the M candidate regions include the same object is greater than a preset threshold, and then N-dimensional features of each candidate region are extracted, and all features of the M candidate regions are combined into a feature matrix of N*M dimensions, wherein N is an integer greater than 1, and then K classifiers are used to determine the probability that each of the M candidate regions belongs to K image types according to the feature matrix, wherein K is an integer greater than 1, and the K classifiers correspond one-to-one to the K image types, and the image type is used to distinguish whether the candidate region is a real shot image, and finally, according to the probability that each candidate region belongs to the K image types, the target type of the target image is determined, and the transaction result is determined according to the target type of the target image.

[0020] From the above content, it can be seen that the present application automatically obtains target images from business scenarios of financial transactions and performs standardized preprocessing on them, including resizing, color correction, etc., to eliminate the potential impact of image quality on subsequent analysis. At the same time, it ensures that the images input into the training model have a uniform format and size, allowing the model to more focused on analyzing the intrinsic characteristics of the image.

[0021] Secondly, use image segmentation algorithms, such as Selective Search, to subdivide the preprocessed image into multiple candidate areas. Subsequently, the defect data in these candidate areas are carefully annotated manually or semi-automatically to generate a Pascal format XML file containing image path, defect type, and coordinate information. These files and images together constitute a training set for training deep learning models. Next, a pre-trained deep convolutional neural network (such as VGG) is used to perform deep N-dimensional feature extraction on each candidate area to generate feature vectors. These feature vectors further form an N*M dimensional feature matrix, providing rich data resources for subsequent classification and identification.

[0022] In addition, by integrating K classifiers, each of which specializes in one image type, the present application can quantitatively evaluate the possibility that each candidate region belongs to K image types based on the feature matrix. The evaluation process not only considers the surface information of the image, but also deeply explores the potential features of the image, making the authenticity identification more accurate and comprehensive.

[0023] It can be seen that through the technical solution of the present application, the purpose of using deep learning and RPA technology to extract key features that are helpful in identifying authenticity from complex images is achieved, thereby achieving the technical effect of efficient and rapid review of key image information in financial transactions, and further solving the technical problem of low transaction review efficiency caused by the high complexity of the image authenticity verification process of the transaction process in the prior art, which leads to slow image review of the transaction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0025] Figure 1 A hardware structure block diagram of a computer terminal for implementing an image processing method is shown;

[0026] Figure 2 is a flowchart of an optional image processing method according to an embodiment of the present application;

[0027] Figure 3 is a feature classification representation diagram of an optional probability distribution matrix according to an embodiment of the present application;

[0028] Figure 4 It is a representation diagram of an optional non-maximum suppression and an overlapping suggestion box according to an embodiment of the present application;

[0029] Figure 5 is an optional flow chart for identifying the authenticity of a transaction image according to an embodiment of the present application;

[0030] Figure 6 It is an optional representation diagram of using a regressor model training according to an embodiment of the present application;

[0031] Figure 7 is a flowchart of an optional RPA automation process method according to an embodiment of the present application;

[0032] Figure 8 is a schematic diagram of an image processing device provided according to an embodiment of the present application;

[0033] Fig. 9 It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0035] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0036] It should also be noted that the collected information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this application are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data are in compliance with relevant laws, regulations and standards, necessary confidentiality measures are taken, and public order and good customs are not violated, and corresponding operation entrances are provided for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions to provide users with corresponding operation entrances for users to choose to agree or refuse the results of automated decision-making; if the user chooses to refuse, the expert decision-making process will be entered.

[0037] According to an embodiment of the present application, a method embodiment of an image processing method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0038] It should be noted that an image processing system can be used as the execution subject of the image processing method of the embodiment of the present application. It is understandable that the image processing method provided in the embodiment of the present application can also be executed by other systems or devices, and the embodiment of the present application does not specifically limit this.

[0039] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing an image processing method. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0040] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuit may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be fully or partially integrated into any of the other components in the computer terminal 10 (or mobile device). As in the image processing method involved in the embodiments of the present application, the data processing circuit serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0041] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the image processing method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned image processing method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0042] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0043] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0044] Under the above operating environment, this application provides Figure 2 The image processing method shown. Figure 2 FIG. 1 is a flowchart of an optional image processing method according to an embodiment of the present application. Figure 2 As shown, the method comprises the following steps:

[0045] Step S201, obtaining a target image during a transaction.

[0046] Optionally, the image processing system accurately acquires the target image in the transaction process. The acquisition process is mainly a sensitive capture of transaction data, ensuring that the extracted image information is directly associated with the physical evidence of the transaction.

[0047] Optionally, the image processing system needs to access transaction record files stored in a specific path, such as a database, cloud storage or local storage. Then, all transaction entries containing image data are identified and extracted from these files, and the images are used as the basis for the next step of analysis.

[0048] Step S202: divide the target image into L image regions, and determine M candidate regions from the L image regions.

[0049] In step S202, L is an integer greater than 1, M is an integer less than or equal to L and greater than 1, and the probability that the M candidate regions include the same object is greater than a preset threshold.

[0050] Optionally, in the preparatory stage of image processing, the image processing system uses image segmentation technology to divide the acquired target image into regions, in order to divide the image into L non-overlapping image sub-regions, ensuring that each sub-region can fully reflect the local characteristics of the image.

[0051] Optionally, the L value is set to an integer greater than 1, which means that the image will be decomposed into at least two or more regions to facilitate subsequent feature extraction and analysis from multiple angles, thereby improving the accuracy and efficiency of intelligent recognition.

[0052] Optionally, the image processing system selects M candidate regions from L image regions. The M value is defined as an integer less than or equal to L and greater than 1, which means that even after the image is segmented, the image processing system will still focus on analyzing a small number (M) of regions that are considered most likely to contain the same object, and the probability of these regions being identified exceeds a preset threshold.

[0053] Optionally, the selection of M candidate regions reflects the reasonable quantity control of image regions by the image processing system. The numerical relationship between L and M ensures the comprehensiveness and pertinence of the analysis, avoids the problem of over-segmentation or under-analysis, and thus lays a dual foundation of efficiency and accuracy in the process of region segmentation and candidate region screening.

[0054] Step S203: extract N-dimensional features of each candidate region, and combine all features of the M candidate regions into a feature matrix of N*M dimensions.

[0055] In step S203, N is an integer greater than 1.

[0056] Optionally, the image processing system performs fine feature extraction on the M candidate regions in order to capture their intrinsic representation from the image's texture, shape, color, and semantic levels. The feature extraction process utilizes pre-trained convolutional neural networks (CNNs), especially high-performance models such as VGG, to convert each candidate region into an N-dimensional feature vector, where N is an integer greater than 1, representing the dimension of the feature space. The purpose of feature extraction is to convert the visual information of the image into a machine-processable mathematical description, thereby providing a solid data foundation for subsequent classification and identification.

[0057] Optionally, after extracting the features of the candidate regions, the image processing system constructs a highly structured N*M dimensional feature matrix. The matrix is ​​composed of N-dimensional feature vectors of M candidate regions arranged horizontally, each vector represents a quantitative description of the local features of the image, and the matrix as a whole constitutes a collective expression of the global features of the image.

[0058] Step S204 , determining the probability that each of the M candidate regions belongs to K image types according to the feature matrix using K classifiers.

[0059] In step S204, K is an integer greater than 1, and the K classifiers correspond one-to-one to the K image types, and the image type is used to distinguish whether the candidate region is a real image.

[0060] Optionally, the image processing system determines the probability that each of the M candidate regions belongs to K image types based on the feature matrix through K classifiers. The core of this process is to use K classifiers, where K is an integer greater than 1, to ensure a detailed probability evaluation of each possibility of the M candidate regions.

[0061] Optionally, the classification method is based on the N-dimensional features of each candidate region, using a pre-trained SVM (Support Vector Machine) classifier to interact with the feature matrix to generate a matrix representing the probability that the M candidate regions belong to K image types.

[0062] Step S205, determining the target type of the target image according to the probability that each candidate region belongs to the K image types, and determining the transaction result according to the target type of the target image.

[0063] Optionally, the image processing system determines the overall attribution type of the target image based on feature analysis and probability evaluation of the M candidate regions. This process takes the probability matrix output by the K classifiers for the M candidate regions as input and calculates the average or aggregate probability of each image type for the target image.

[0064] Optionally, the image processing system determines a target type based on the probability as the most likely classification label of the target image. The target type not only reflects the core features of the target image, but is also directly related to the authenticity of the image, and is the key basis for subsequent transaction review and judgment.

[0065] It can be known from the contents of step S201 to step S205 that in the embodiment of the present application, firstly, a target image in the transaction process is acquired, then the target image is divided into L image regions, and M candidate regions are determined from the L image regions, wherein L is an integer greater than 1, M is an integer less than or equal to L and greater than 1, and the probability that the M candidate regions include the same object is greater than a preset threshold, and then the N-dimensional features of each candidate region are extracted, and all the features of the M candidate regions are combined into a feature matrix of N*M dimensions, wherein N is an integer greater than 1, and then the probability that each of the M candidate regions belongs to K image types is determined according to the feature matrix by K classifiers, wherein K is an integer greater than 1, and the K classifiers correspond one-to-one to the K image types, and the image type is used to distinguish whether the candidate region is a real shot image, and finally, the target type of the target image is determined according to the probability that each candidate region belongs to the K image types, and the transaction result is determined according to the target type of the target image.

[0066] From the above content, it can be seen that the present application automatically obtains target images from business scenarios of financial transactions and performs standardized preprocessing on them, including resizing, color correction, etc., to eliminate the potential impact of image quality on subsequent analysis. At the same time, it ensures that the images input into the training model have a uniform format and size, allowing the model to more focused on analyzing the intrinsic characteristics of the image.

[0067] Secondly, use image segmentation algorithms, such as Selective Search, to subdivide the preprocessed image into multiple candidate areas. Subsequently, the defect data in these candidate areas are carefully annotated manually or semi-automatically to generate a Pascal format XML file containing image path, defect type, and coordinate information. These files and images together constitute a training set for training deep learning models. Next, a pre-trained deep convolutional neural network (such as VGG) is used to perform deep N-dimensional feature extraction on each candidate area to generate feature vectors. These feature vectors further form an N*M dimensional feature matrix, providing rich data resources for subsequent classification and identification.

[0068] In addition, by integrating K classifiers, each of which specializes in one image type, the present application can quantitatively evaluate the possibility that each candidate region belongs to K image types based on the feature matrix. The evaluation process not only considers the surface information of the image, but also deeply explores the potential features of the image, making the authenticity identification more accurate and comprehensive.

[0069] It can be seen that through the technical solution of the present application, the purpose of using deep learning and RPA technology to extract key features that are helpful in identifying authenticity from complex images is achieved, thereby achieving the technical effect of efficient and rapid review of key image information in financial transactions, and further solving the technical problem of low transaction review efficiency caused by the high complexity of the image authenticity verification process of the transaction process in the prior art, which leads to slow image review of the transaction process.

[0070] In an optional embodiment, the image processing system first multiplies the N*M dimensional feature matrix with the weight matrix composed of K classifiers to obtain an N*K dimensional probability distribution matrix, where the weight matrix is ​​an M*K dimensional matrix, and then uses the probability distribution matrix to represent the probability that each of the M candidate areas belongs to the K image types.

[0071] Optionally, the image processing system performs a scientific operation on the N-dimensional feature vector set generated by the M candidate regions, that is, the N*M-dimensional feature matrix, and the M*K-dimensional weight matrix composed of K image classifiers. This operation is essentially a dot product operation of the feature matrix and the specific weights of the classifier, which aims to quantify the strength of the association between each candidate region and the K image types. Through this calculation process, an N*K-dimensional probability distribution matrix is ​​generated, in which each row represents a candidate region and each column corresponds to one of the K image types. Each element in the matrix represents the probability that the candidate region belongs to a specific image type, providing a quantitative basis for subsequent image authenticity judgment.

[0072] Optionally, using the obtained probability distribution matrix, the image processing system can characterize the possibility that each of the M candidate regions belongs to K image types. The probability distribution matrix not only reflects the similarity between the image content and the known image types, but also implies the judgment logic of the deep learning model for the authenticity of the image. Figure 3 As shown, among the 4096-dimensional features, the probability that the first feature belongs to the first type is 0.12 (the probability that the first feature belongs to other types is not shown); the probability that the 4096th feature belongs to the kth type is 0.35 (the probability that the 4096th feature belongs to other types is not shown). Figure 3 0.89 and 0.01 are also probability values.

[0073] Optionally, by analyzing the distribution probability of each candidate region on different image types, the image processing system can identify the type that is most likely to reflect the authenticity of the image, and based on this information, comprehensively evaluate the overall authenticity of the image, thereby improving the accuracy of image discrimination.

[0074] From the above content, it can be seen that the image processing system obtains a probability distribution matrix by multiplying the feature matrix with the weight matrix composed of the classifier, and then uses the probability distribution matrix to represent the probability that each of the M candidate areas belongs to the K image types, thereby realizing the intelligent extraction and classification of image features, so that the authenticity of image information can be accurately identified in complex and changeable financial transaction scenarios.

[0075] In an optional embodiment, the image processing system first calculates the intersection-and-union ratio of any two candidate areas among the M candidate areas, wherein the intersection-and-union ratio of any two candidate areas represents the ratio of the overlapping area of ​​the two candidate areas to the total area of ​​the two candidate areas, and then takes the candidate area with the highest probability of belonging to the i-th image type as the first area corresponding to the i-th image type, and takes the candidate area among the M candidate areas whose intersection-and-union ratio with the first area is less than a first threshold as the second area corresponding to the i-th image type, wherein i is a positive integer less than or equal to K, and then determines the target type of the target image based on the first area and the second area corresponding to each image type.

[0076] Optionally, for the N*K dimensional probability distribution matrix of the M candidate regions, the degree of spatial overlap between the candidate regions is measured by calculating the Intersection over Union (IoU) of any two candidate regions.

[0077] Optionally, the IoU index, i.e. (A∩B) / (AUB), is the ratio of the overlapping area of ​​AB to the total area of ​​AB. Intuitively speaking, IoU represents the ratio of AB overlap. The larger the IoU, the greater the overlap ratio of AB, that is, the more similar A and B are. This calculation process provides a key quantitative indicator for non-maximum suppression (NMS), which aims to filter out redundant candidate areas and retain the most representative image features.

[0078] Optionally, for K types of images, the image processing system automatically identifies and assigns each candidate region to the corresponding image type. First, the first region is determined. For the i-th image type (i≤K), the candidate region with the highest probability and representativeness is identified and defined as the first region. This region is considered to be the most characteristic part of the image type. Then the second region is screened, and from the M candidate regions, the regions whose IoU ratio with the first region is less than a preset first threshold are further screened and defined as the second region. Although these regions overlap with the first region, the degree of overlap is low, so independent feature information is still retained, which is crucial for the overall judgment of the image type.

[0079] Optionally, by defining and screening the first region and the second region, the image processing system can comprehensively consider the distribution of characteristic regions under each image type, and then determine the comprehensive type of the target image. For each image type i, the image processing system evaluates the characteristic performance of the first region and the second region, combines their position information and probability weights in the N*K probability distribution matrix, and finally determines the characteristic combination of the target image under this type, ensuring the accuracy and comprehensiveness of the image type judgment.

[0080] Alternatively, if Figure 4 As shown in the figure, the specific implementation method of non-maximum suppression to remove overlapping suggestion boxes is: Step 1: Define the IoU index, that is, (A∩B) / (AUB), which is the ratio of the overlapping area of ​​AB to the total area of ​​AB. Intuitively speaking, IoU represents the ratio of AB overlap. The larger the IoU, the greater the proportion of the overlapping part of AB, that is, the more similar A and B are. Step 2: Find the area with the highest probability among the candidate areas in each category, calculate the IoU values ​​of other areas and this area, and delete all candidate areas with IoU values ​​greater than the threshold. In this way, only a few candidate areas with low overlap rates can be retained, and duplicate areas can be removed.

[0081] From the above content, we can see that the image processing system determines the comprehensive type of the target image by defining and screening the first area and the second area, which reflects the refinement and extraction of image features by the deep learning model. It also scientifically analyzes the results of multiple classifiers through the setting of IoU ratio and threshold, and realizes the intelligent classification and allocation of candidate areas.

[0082] In an optional embodiment, the image processing system first determines the offset and scaling ratio of the first region in the target direction and the offset and scaling ratio of the second region in the target direction through a regressor, and then adjusts the first region to a first target region according to the offset and scaling ratio of the first region in the target direction, and then adjusts the second region to a second target region according to the offset and scaling ratio of the second region in the target direction, and finally determines the target type of the target image according to the first target region and the second target region.

[0083] Optionally, the image processing system uses a regressor to adjust the position and size of the first region and the second region initially identified by the deep learning model, aiming to accurately locate and refine the region contour according to the attributes of each region in the probability distribution matrix, thereby improving the accuracy of image type recognition. Through the regressor, the image processing system can calculate the offset and scaling of each region in the target direction. The determination of these parameters is based on the prediction results of the deep learning model, and the best matching degree after the region adjustment is ensured through the least squares method or similar optimization algorithm.

[0084] Optionally, the model training process includes using manually annotated data as training set data, selecting a suitable regression model to initialize model parameters, and then inputting the data into the model for model regression training. The least squares method can be used to solve the linear regression problem to obtain an actual area generated by the training set data.

[0085] Optionally, the image processing system converts the first region into a more accurate first target region according to the offset and scaling ratio of the first region in the target direction calculated by the regressor. By applying the above offset and scaling ratio, the boundary of the first region is adjusted to ensure that it more accurately matches the geometric features of the specific type of information contained in the image. The converted first target region not only retains the key features of the original region, but also further optimizes its position and size, thereby improving the accuracy and reliability of image type recognition.

[0086] Optionally, for a second region whose IoU ratio with the first region is less than a first threshold, the image processing system achieves precise adjustment from the second region to the second target region through an offset and scaling ratio determined by a regressor, thereby ensuring that the second target region is more accurately positioned in the image, while optimizing its geometric dimensions to better reflect the characteristic performance of one of the K image types.

[0087] Optionally, by combining the prediction of the first target region and the second target region, the image processing system can comprehensively analyze the characteristic information contained in the target image and determine the target type of the target image. This fully considers the position and size of each region in the image and the probability association with K types of image types, ensuring the accuracy and comprehensiveness of the image category judgment.

[0088] From the above content, it can be seen that the image processing system determines the target type of the target image by adjusting the first target area and the second target area, and fully considers the position, size and probabilistic association of each area in the image with K image types, thereby ensuring the accuracy and comprehensiveness of image category judgment.

[0089] In an optional embodiment, the image processing system first determines the probability that the target image belongs to each image type among K image types based on the probability value of the first target area corresponding to each image type belonging to the image type, the number of second target areas corresponding to each image type, and the probability value of each second target area belonging to the image type, and then uses the image type to which the target image belongs with the highest probability as the target type of the target image.

[0090] Optionally, the image processing system combines the probability attribution value of the first target area, the number of second target areas, and the probability contribution of each second target area to the image type to which it belongs, and calculates the probability that the target image belongs to each of the K image types. This process fully considers the recognition results of the key feature areas in the image and the comprehensive information of the non-key areas, ensuring the comprehensiveness and accuracy of the image type attribution judgment.

[0091] Optionally, after calculating the probability value of the target image belonging to each of the K types of images, the image processing system selects the type with the largest probability value of the target image belonging to the K types of images as its target type. Through this process, the image processing system can finally determine the type of the target image, ensuring the scientificity and accuracy of the image review process.

[0092] From the above content, it can be seen that the image processing system uses the probability value of the first target area of ​​each image type belonging to the image type, the number of second target areas corresponding to each image type, and the probability value of each second target area belonging to the image type to accurately and efficiently determine the type of the target image.

[0093] In an optional embodiment, the image processing system first obtains the user data and transaction data involved in the transaction process from the target file by accessing the target storage path, and then uses the image in the user data and transaction data whose authenticity needs to be verified as the target image.

[0094] Optionally, by directly accessing the target storage path, the image processing system can efficiently read and obtain comprehensive data in the transaction process, including user data and transaction data. This not only reflects the precise positioning of the data source, but also ensures the efficiency and accuracy of data acquisition, providing a solid data foundation for subsequent image identification and business audits.

[0095] Optionally, for image information that needs to be authenticated, the image processing system uses these images as target images for model input, aiming to quickly locate those images that may have a significant impact on business audit results from a large amount of transaction data.

[0096] From the above content, it can be seen that by accessing the target storage path, obtaining the user data and transaction data involved in the transaction process from the target file, and then using the images in the user data and transaction data that need to be authenticated as the target image, data acquisition in the automated audit process can be achieved, and effective positioning and processing of key image information can be achieved.

[0097] In an optional embodiment, the training process of the classifier includes the following steps: first, obtaining training images of K image types, wherein the training images of the K image types include at least images that are actually taken in different business scenarios and images that are not actually taken; then, dividing the training images into a plurality of sub-region images; and then extracting image features of each sub-region image; and finally, iteratively training the neural network according to the image features of each sub-region image of the training images of each image type to obtain a classifier corresponding to the image type.

[0098] Optionally, the image processing system obtains training images of K types of images mainly by selecting real image information in different business scenarios and business image information of different scenarios generated by AI as the image sets to be trained.

[0099] Optionally, after obtaining the training image, the image processing system uses an optional image segmentation algorithm, such as a selective search algorithm, to segment each training image into multiple sub-region images. The segmentation process not only considers the geometric structure and visual feature distribution of the image, but also ensures that each sub-region image can carry sufficient information for feature extraction and model training.

[0100] Optionally, for each generated sub-region image, the image processing system applies an optional feature extraction algorithm, such as a convolutional neural network (CNN), to extract multi-level visual features from the image.

[0101] Optionally, based on the sub-region image features of each image type, the image processing system iteratively trains the deep neural network to optimize the model parameters and improve the model's ability to recognize and classify image features. Through a large amount of training data and scientific training strategies, the neural network can learn the key features to distinguish K types of images, and finally generate a high-precision classifier corresponding to each image type.

[0102] From the above content, it can be seen that during the training process of the classifier, the image processing system achieves effective recognition and classification of K types of images through multi-source image data acquisition, scientific image segmentation process, application of feature extraction algorithm and iterative training of deep neural network.

[0103] In an optional embodiment, the image processing system performs feature annotation on abnormal sub-image areas of a training image that is not actually taken, wherein the abnormal sub-image area is an image area constructed using an image synthesis method, and the feature annotation is used to generate a feature file including coordinate information, synthesis method, and saving path of the abnormal sub-image area.

[0104] Optionally, the image processing system uses labeling software to calibrate the AI ​​defect part in the training image and generate a corresponding pascal format XML file. The generated XML file contains the image path, defect type and corresponding coordinate information.

[0105] Optionally, the training set consists of two parts: images and XML files corresponding to the images.

[0106] From the above content, it can be seen that the image processing system realizes effective recognition and recording of images by performing feature annotation on abnormal sub-image areas of training images that are not actually taken.

[0107] In an optional embodiment, Figure 5 An optional flow chart of identifying the authenticity of a transaction image according to an embodiment of the present application is shown, which can more clearly understand the process of identifying the authenticity of a transaction image in the image processing method:

[0108] 1. Input images: Select real image information under different business scenarios and business image information of different scenarios generated by AI as the image sets to be trained.

[0109] 2. Image preprocessing: Use the selective search algorithm to divide the image into small areas, and then merge the areas with a high probability of containing the same object as candidate areas for output. For example, if you want to generate 2,000 candidate areas, then normalize each area, that is, scale it to a fixed size.

[0110] 3. Label the defect data: Use labeling software to calibrate the AI ​​defect part in the training image and generate the corresponding pascal format XML file. The generated XML file contains the image path, defect type and corresponding coordinate information. The training set consists of two parts: the image and the XML file corresponding to the image.

[0111] 4. Feature extraction: Use CNN to extract features for each candidate region. Use the optional VGG pre-trained neural network and retrain the fully connected layer. Input the candidate region into the trained VGG network and wait for the fixed-dimensional feature output (4096 dimensions) to obtain a 2000*4096 feature matrix.

[0112] 5. Feature classification: Use SVM classifier to classify the output features of CNN. The data set is divided into real pictures and several categories of AI-generated pictures. Assuming there are 20 categories in total, 20 SVM classifiers are set. Multiply the 2000*4096 features with the 4096*20 weight matrix composed of 20 SVMs to obtain a 2000*20 dimensional matrix, which represents the probability that the 2000 candidate regions belong to the 20 categories respectively, so the sum of each row of the matrix is ​​1. For each column in the above 2000*20 dimensional matrix, that is, each category, non-maximum suppression is performed to propose overlapping suggestion boxes, and some candidate boxes with the highest probability in the category are obtained.

[0113] 6. Regression operation: Use the regressor to refine the position of the candidate region. The position of the candidate region obtained by the selective search algorithm may not be accurate, so 20 regressors are used to perform regression operations on the remaining suggestion boxes in the above 20 categories, and finally the corrected target region of each category is obtained.

[0114] Among them, Figure 6, frame 1 represents the candidate region, frame 3 represents the actual region (the manually marked region), and frame 2 represents the predicted region after regression. Here, the least squares method is used to solve the linear regression problem.

[0115] 7. Output results: The four parameters of the candidate region can be obtained through the regressor, namely: the x and y offsets of the candidate region, and the scaling factors of the height and width. These four parameters can be used to fine-tune the position of the candidate region to obtain the predicted region.

[0116] In an optional embodiment, Figure 7 A flowchart of an optional RPA automation process method according to an embodiment of the present application is shown, which can more clearly understand the RPA automation process:

[0117] 1. Customer data processing: The customer data to be processed is stored in a data file, which will serve as the entry point for RPA, i.e. which customers’ transaction review processes will be processed next.

[0118] 2. Data reading: Use the software automation function to specify the absolute path of the data file, which is the customer data file prepared in the first step of data processing. Then read the customer data required for processing the business process from the specified cell location in the file.

[0119] 3. Business data capture: After reading the data, you need to open the transaction review interface and get the results from the page. Use the data capture function to select the data to be captured on the specified page, then you can capture the data at the selected location and store it in a new data file at the specified path as the result of the above business process processing.

[0120] 4. Transaction image identification: The acquired image information is input into the image recognition system. If it is a real image, it will pass safely; if it is a forged image, the forgery point and risk factor will be output for secondary confirmation.

[0121] 5. Secondary confirmation by business personnel: Business personnel make a secondary judgment based on the key information captured by RPA and the judgment results in the image recognition system. If it passes, the transaction review is completed. If it fails, it is manually confirmed again.

[0122] According to another aspect of the embodiment of the present application, an image processing device is also provided, wherein: Figure 8 is a schematic diagram of an optional image processing device according to an embodiment of the present application, such as Figure 8 As shown, the image processing device includes: an acquisition unit 801, a processing unit 802, an extraction unit 803, a first determination unit 804, and a second determination unit 805.

[0123] An acquisition unit 801 acquires a target image in a transaction process; a processing unit 802 divides the target image into L image regions, and determines M candidate regions from the L image regions, wherein L is an integer greater than 1, M is an integer less than or equal to L and greater than 1, and the probability that the M candidate regions include the same object is greater than a preset threshold; an extraction unit 803 extracts N-dimensional features of each candidate region, and combines all features of the M candidate regions into a feature matrix of N*M dimensions, wherein N is an integer greater than 1; a first determination unit 804 determines the probability that each of the M candidate regions belongs to K types of image types according to the feature matrix through K classifiers, wherein K is an integer greater than 1, and the K classifiers correspond one-to-one to the K types of image types, and the image type is used to distinguish whether the candidate region is a real shot image; a second determination unit 805 determines the target type of the target image according to the probability that each candidate region belongs to the K types of image types, and determines the transaction result according to the target type of the target image.

[0124] Optionally, the first determination unit 804 includes: a first processing subunit and a second processing subunit. The first processing subunit is used to multiply the feature matrix of N*M dimensions by the weight matrix composed of K classifiers to obtain a probability distribution matrix of N*K dimensions, wherein the weight matrix is ​​a matrix of M*K dimensions; and the second processing subunit is used to represent the probability that each of the M candidate regions belongs to K types of image types through the probability distribution matrix.

[0125] Optionally, the second determination unit 805 includes: a first calculation subunit, a third processing subunit and a first determination subunit. The first calculation subunit is used to calculate the intersection-and-union ratio of any two candidate regions among the M candidate regions, wherein the intersection-and-union ratio of any two candidate regions represents the ratio of the overlapping area of ​​the two candidate regions to the total area of ​​the two candidate regions; the third processing subunit is used to take the candidate region with the highest probability of belonging to the i-th image type as the first region corresponding to the i-th image type, and take the candidate region among the M candidate regions whose intersection-and-union ratio with the first region is less than a first threshold as the second region corresponding to the i-th image type, wherein i is a positive integer less than or equal to K; the first determination subunit is used to determine the target type of the target image according to the first region and the second region corresponding to each image type.

[0126] Optionally, the first determination subunit includes: a first determination module, a first adjustment module, a second adjustment module and a second determination module. The first determination module is used to determine the offset and scaling ratio of the first region in the target direction, and the offset and scaling ratio of the second region in the target direction through a regressor; the first adjustment module is used to adjust the first region to the first target region according to the offset and scaling ratio of the first region in the target direction; the second adjustment module is used to adjust the second region to the second target region according to the offset and scaling ratio of the second region in the target direction; the second determination module is used to determine the target type of the target image according to the first target region and the second target region.

[0127] Optionally, the second determination module includes: a first determination submodule and a first processing submodule. The first determination submodule is used to determine the probability that the target image belongs to each image type among the K image types according to the probability value that the first target area corresponding to each image type belongs to the image type, the number of second target areas corresponding to each image type, and the probability value that each second target area belongs to the image type; and the first processing submodule is used to take the image type to which the target image has the highest probability of belonging as the target type of the target image.

[0128] Optionally, the acquisition unit 801 includes: a first acquisition subunit and a fourth processing subunit. The first acquisition subunit is used to acquire user data and transaction data involved in the transaction process from the target file by accessing the target storage path; the fourth processing subunit is used to use the image in the user data and transaction data that needs to be authenticated as the target image.

[0129] Optionally, the image processing device further includes: a second acquisition unit, a second processing unit, a second extraction unit, and a training unit. The second acquisition unit is used to acquire training images of K types of images, wherein the training images of the K types of images at least include images taken in real situations and images not taken in real situations; the second processing unit is used to divide the training images into a plurality of sub-region images; the second extraction unit is used to extract image features of each sub-region image; and the training unit is used to iteratively train the neural network according to the image features of each sub-region image of the training images of each image type, so as to obtain a classifier corresponding to the image type.

[0130] Optionally, the second extraction unit also includes: a fifth processing sub-unit, used to perform feature annotation on abnormal sub-image areas of the training image that is not actually taken, wherein the abnormal sub-image area is an image area constructed using an image synthesis method, and the feature annotation is used to generate a feature file including coordinate information, synthesis method and saving path of the abnormal sub-image area.

[0131] An embodiment of the present application may provide an electronic device, Fig. 9 is a structural block diagram of an electronic device according to an embodiment of the present application. Fig. 9 As shown, the electronic device may include: one or more ( Fig. 9 Only one is shown) a processor, a memory, a storage controller, and a peripheral interface, wherein the peripheral interface is connected to a radio frequency module, an audio module, and a display.

[0132] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the methods and devices in the embodiments of the present application, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0133] It can be understood by those skilled in the art that Fig. 9 The structure shown is for illustration only, and the electronic device may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (Mobile Internet Devices, MID), a PAD, or other terminal devices. Fig. 9 The structure of the electronic device is not limited. Fig. 9 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Fig. 9 Different configurations shown.

[0134] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0135] According to another aspect of the present application, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, wherein when the computer program is executed, the device where the computer-readable storage medium is located executes the above-mentioned image processing method.

[0136] According to another aspect of the present application, a computer program product is further provided, wherein the computer program product includes computer instructions, wherein when the computer instructions are executed, the device where the computer program product is located executes the above-mentioned image processing method.

[0137] The above-mentioned embodiments or examples disclosed in the present application are not exhaustive, but are only illustrative of some embodiments or examples, and are not intended to be specific limitations on the scope of protection disclosed in the present application. In the absence of contradiction, each step in a certain embodiment or example in the present application can be implemented as an independent example, and the steps can be combined arbitrarily. For example, the scheme after removing some steps in a certain embodiment or example can also be implemented as an independent example, and the order of the steps in a certain embodiment or example can be arbitrarily exchanged. In addition, the optional methods or optional examples in a certain embodiment or example can be combined arbitrarily; in addition, the various embodiments or examples can be combined arbitrarily, for example, some or all steps of different embodiments or examples can be combined arbitrarily, and a certain embodiment or example can be combined arbitrarily with the optional methods or optional examples of other embodiments or examples.

[0138] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0139] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0140] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0141] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0142] 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. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0143] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0144] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An image processing method, characterized in that: include: Acquire the target image during the transaction; Divide the target image into L image regions, and determine M candidate regions from the L image regions, where L is an integer greater than 1, M is an integer less than or equal to L and greater than 1, and the probability that the M candidate regions include the same object is greater than a preset threshold; Extract N-dimensional features of each candidate region, and combine all features of the M candidate regions into a feature matrix of N*M dimensions, where N is an integer greater than 1; Determine, by K classifiers according to the feature matrix, the probability that each of the M candidate regions belongs to K image types, wherein K is an integer greater than 1, and the K classifiers correspond one-to-one to the K image types, and the image type is used to distinguish whether the candidate region is a real image; According to the probability that each candidate area belongs to the K types of images, the target type of the target image is determined, and the transaction result is determined according to the target type of the target image.

2. The image processing method according to claim 1, characterized in that: Determining the probability that each of the M candidate regions belongs to K image types by K classifiers according to the feature matrix includes: Multiplying the feature matrix of N*M dimension by the weight matrix composed of K classifiers to obtain a probability distribution matrix of N*K dimension, wherein the weight matrix is ​​a matrix of M*K dimension; The probability distribution matrix is ​​used to represent the probability that each of the M candidate regions belongs to the K image types.

3. The image processing method according to claim 1, characterized in that: Determining the target type of the target image according to the probability that each candidate region belongs to the K types of image types, including: Calculating an intersection-and-union ratio of any two candidate regions among the M candidate regions, wherein the intersection-and-union ratio of any two candidate regions represents a ratio of an overlapping area of ​​the two candidate regions to a total area of ​​the two candidate regions; The candidate region with the highest probability of belonging to the i-th image type is used as the first region corresponding to the i-th image type, and the candidate region whose intersection-over-union ratio with the first region among the M candidate regions is less than a first threshold is used as the second region corresponding to the i-th image type, wherein i is a positive integer less than or equal to K; The target type of the target image is determined according to the first area and the second area corresponding to each image type.

4. The image processing method according to claim 3, characterized in that: Determining the target type of the target image according to the first area and the second area corresponding to each image type includes: Determine, by a regressor, an offset and a scaling ratio of the first region in a target direction, and an offset and a scaling ratio of the second region in the target direction; adjusting the first area to a first target area according to an offset of the first area in a target direction and a scaling ratio; adjusting the second area to a second target area according to the offset and scaling ratio of the second area in the target direction; An object type of the object image is determined according to the first object area and the second object area.

5. The image processing method according to claim 4, characterized in that: Determining the target type of the target image according to the first target area and the second target area includes: Determine the probability that the target image belongs to each of the K image types according to the probability value of the first target area corresponding to each image type belonging to the image type, the number of second target areas corresponding to each image type, and the probability value of each second target area belonging to the image type; The image type to which the target image belongs with the highest probability is used as the target type of the target image.

6. The image processing method according to claim 1, characterized in that: Obtain target images during the transaction, including: By accessing the target storage path, the user data and transaction data involved in the transaction process are obtained from the target file; The image whose authenticity needs to be authenticated among the user data and the transaction data is used as the target image.

7. The image processing method according to claim 1, characterized in that: The training process of the classifier includes the following steps: Acquire training images of K image types, wherein the training images of the K image types at least include images that are actually taken in different business scenarios and images that are not actually taken; Dividing the training image into a plurality of sub-region images; Extracting image features of each sub-region image; According to the image features of each sub-region image of the training image of each image type, the neural network is iteratively trained to obtain a classifier corresponding to the image type.

8. The image processing method according to claim 7, characterized in that: Extract image features of each sub-region image, including: Feature annotation is performed on the abnormal sub-image area of ​​the non-realistic training image, wherein the abnormal sub-image area is an image area constructed using an image synthesis method, and the feature annotation is used to generate a feature file including coordinate information, synthesis method and saving path of the abnormal sub-image area.

9. An image processing device, characterized in that: include: An acquisition unit, which acquires a target image during a transaction; a processing unit, dividing the target image into L image regions, and determining M candidate regions from the L image regions, wherein L is an integer greater than 1, M is an integer less than or equal to L and greater than 1, and a probability that the M candidate regions include the same object is greater than a preset threshold; An extraction unit extracts N-dimensional features of each candidate region, and combines all features of the M candidate regions into a feature matrix of N*M dimensions, where N is an integer greater than 1; A first determining unit, determining, by K classifiers according to the feature matrix, a probability that each of the M candidate regions belongs to K image types, wherein K is an integer greater than 1, the K classifiers correspond one-to-one to the K image types, and the image type is used to distinguish whether the candidate region is a real image; The second determination unit determines the target type of the target image according to the probability that each candidate area belongs to the K types of images, and determines the transaction result according to the target type of the target image.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is executed, the device where the computer-readable storage medium is located is controlled to execute the image processing method according to any one of claims 1 to 8.

11. An electronic device, characterized in that: include: A memory storing an executable program; A processor is used to run the program, wherein the program executes the image processing method according to any one of claims 1 to 8 when running.

12. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the image processing method according to any one of claims 1 to 8 are implemented.

Images