Financial business background environment visual risk control detection method based on image analysis

Abstract

The invention relates to a financial business background environment visual risk control detection method based on image analysis. The method comprises the following steps: acquiring continuous image frame data, and extracting a feature point data set from the continuous image frame data; wherein the feature point data set comprises target feature points of each frame of image frame data; performing feature point change analysis on the obtained feature point data set to obtain an image analysis result; acquiring a background change analysis result according to the image analysis result and a preset judgment mode; wherein the background change analysis result comprises at least one of significance judgment information and abnormal mark information. According to the scheme provided by the invention, the progressive change and periodic interference in the background environment can be effectively and systematically captured, so that the capability of judging the abnormal change of the background environment is improved, and the problem of misjudgment caused by neglecting the time sequence relevance is effectively solved.

Description

Technical Field

[0001] This application relates to the technical field of financial business data processing, and particularly to a visual risk control detection method for the background environment of financial business based on image analysis. Background Art

[0002] In the current field of financial business risk control, the monitoring of changes in the background environment of loan customers is mainly based on static image feature comparison technology.

[0003] In the related art, environmental image data collected at a single time point is often used, and consistency verification is carried out by extracting visual features in the spatial dimension (such as object layout, scene identification, lighting conditions); however, in the above verification method, since the feature sampling at a single time point cannot construct a time series dimension analysis model, it is difficult to effectively identify the temporal features of progressive scene transformations (such as gradual adjustment of interior decoration, gradual change of day and night light), and it is easy to produce misjudgments due to the failure of feature point matching, affecting the accuracy of risk control detection. Summary of the Invention

[0004] To solve or partially solve the problems existing in the related art, this application provides a visual risk control detection method for the background environment of financial business based on image analysis, which can effectively and systematically capture progressive changes and periodic interferences in the background environment, thereby improving the ability to judge abnormal changes in the background environment and effectively solving the problem of misjudgment caused by ignoring temporal relevance.

[0005] The first aspect of this application provides a visual risk control detection method for the background environment of financial business based on image analysis, including: Obtain continuous image frame data, and extract a feature point data set from the continuous image frame data; wherein, the feature point data set includes: target feature points of each frame of the image frame data; Perform feature point change analysis on the obtained feature point data set to obtain an image analysis result; According to the image analysis result and a preset judgment method, obtain a background change analysis result; wherein, the background change analysis result includes at least one of significance judgment information and abnormal marking information.

[0006] In some embodiments, the obtaining continuous image frame data and extracting a feature point data set from the continuous image frame data includes: Obtain continuous image frame data, and extract feature points from each frame of the image frame data by using the same preset feature extraction method to obtain a feature point data set; wherein, the preset feature extraction method includes at least one of scale-invariant feature transform, speeded-up robust features, and fast rotation invariant features.

[0007] In some embodiments, the feature point change analysis of the obtained feature point data set includes: Determine the reference frame feature points and the comparison frame feature points matching the reference frame feature points from the feature point data set, and compare the comparison frame feature points with the reference frame feature points; Perform data analysis based on the obtained comparison result to obtain an image analysis result.

[0008] In some embodiments, the comparing the comparison frame feature points with the reference frame feature points includes: Calculate the feature point vector parameters of the reference frame feature points and the comparison frame feature points, and compare the comparison frame feature points with the reference frame feature points for the feature point vector parameters; and / or Calculate the feature point intensity parameters of the reference frame feature points and the comparison frame feature points, and compare the comparison frame feature points with the reference frame feature points for the feature point intensity parameters.

[0009] In some embodiments, the performing data analysis based on the obtained comparison result includes: When the obtained comparison result includes the comparison result of the feature point vector parameters and / or the comparison result of the feature point intensity parameters, after normalizing the comparison result of the feature point vector parameters and / or the comparison result of the feature point intensity parameters, construct a feature change curve with time as the horizontal axis.

[0010] In some embodiments, before performing data analysis based on the obtained comparison result to obtain an image analysis result, it further includes: Calculate the feature point distribution parameters of the reference frame feature points and the comparison frame feature points, and compare the comparison frame feature points with the reference frame feature points for the feature point distribution parameters; The method further includes: When the obtained comparison result includes the comparison result of the feature point distribution parameters, determine the image geometric change result according to the comparison result of the feature point distribution parameters.

[0011] In some embodiments, the obtaining the background change analysis result according to the image analysis result and a preset judgment method includes: Calculate the preset type parameters of the feature change curve to obtain change curve parameters; Compare the change curve parameters with a preset dynamic threshold and perform a significance evaluation to determine significance judgment information; When the significance judgment information includes abnormal information, the image frame data associated with the abnormal change is input into a preset image classification model for abnormal classification and recognition; According to the significance judgment information and the type of abnormal change, an analysis result of background change is output.

[0012] In some embodiments, the method further includes: according to the analysis result of background change, a reminder message and / or a preset service control instruction are displayed on a preset interface.

[0013] A second aspect of the present application provides a visual risk control detection system for the financial business background environment based on image analysis, including A data acquisition module, configured to acquire continuous image frame data, and extract a feature point data set from the continuous image frame data; wherein, the feature point data set includes: target feature points of each frame of the image frame data; A feature analysis module, configured to perform feature point change analysis on the acquired feature point data set to obtain an image analysis result; A result analysis module, configured to obtain an analysis result of background change according to the image analysis result and a preset judgment method; wherein, the analysis result of background change includes at least one of significance judgment information and abnormal marking information.

[0014] A third aspect of the present application provides an electronic device, including: A processor; and A memory, on which executable code is stored, and when the executable code is executed by the processor, the processor is caused to execute the method as described above.

[0015] A fourth aspect of the present application provides a computer-readable storage medium, on which executable code is stored, and when the executable code is executed by a processor of an electronic device, the processor is caused to execute the method as described above.

[0016] The technical solution provided by the present application may include the following beneficial effects: The technical solution of the present application can effectively and systematically capture the gradual changes and periodic interferences in the background environment by collecting continuous image frame data and performing feature point acquisition and change analysis, thereby improving the ability to judge abnormal changes in the background environment. And through the output significance judgment information and abnormal marking information, the degree and risk category of environmental changes can be synchronously reflected. Compared with the single-time point feature matching method, the problem of misjudgment caused by ignoring the temporal correlation can be effectively solved, providing multi-dimensional dynamic risk control indicators for the verification of the financial business background.

[0017] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. Description of the Drawings

[0018] The above and other objects, features, and advantages of the present application will become more apparent by describing the exemplary embodiments of the present application in more detail in conjunction with the accompanying drawings, wherein, in the exemplary embodiments of the present application, the same reference numerals generally represent the same components.

[0019] Figure 1 is a schematic flowchart of a visual risk control detection method for a financial business background environment based on image analysis shown in an embodiment of the present application; Figure 2 is another schematic flowchart of a visual risk control detection method for a financial business background environment based on image analysis shown in an embodiment of the present application; Figure 3 is another schematic flowchart of a visual risk control detection method for a financial business background environment based on image analysis shown in an embodiment of the present application; Figure 4 is a schematic structural diagram of a visual risk control detection system for a financial business background environment based on image analysis shown in an embodiment of the present application; Figure 5 is a schematic structural diagram of an electronic device shown in an embodiment of the present application. Detailed Description of the Embodiments

[0020] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0021] The terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. The singular forms "a", "the", and "said" used in the present application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0022] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of this application, the meaning of "a plurality" is two or more, unless otherwise specifically defined.

[0023] In the related art, environmental image data collected at a single time point is often used, and consistency verification is carried out by extracting visual features in the spatial dimension (such as object layout, scene identification, lighting conditions); however, in the above verification method, since the feature sampling at a single time point cannot construct a time series dimension analysis model, it is difficult to effectively identify the time series features of progressive scene transformations (such as gradual adjustment of interior decoration, gradual change of day and night light), and it is easy to produce misjudgment due to the failure of feature point matching, affecting the accuracy of risk control detection.

[0024] In view of the above problems, an embodiment of this application provides a visual risk control detection method for the financial business background environment based on image analysis, which can effectively and systematically capture progressive changes and periodic interferences in the background environment, thereby improving the ability to judge abnormal changes in the background environment and effectively solving the problem of misjudgment caused by ignoring time series relevance.

[0025] The technical solutions of the embodiments of this application are described in detail below with reference to the accompanying drawings.

[0026] Figure 1 is a schematic flowchart of a visual risk control detection method for the financial business background environment based on image analysis shown in the embodiments of this application.

[0027] See Figure 1 , the visual risk control detection method for the financial business background environment based on image analysis of this application includes: S110, obtaining continuous image frame data, and extracting a feature point data set from the continuous image frame data; wherein, the feature point data set includes target feature points of each frame of image frame data.

[0028] In this step, continuous image frame data is obtained from the device side where the financial business is handled, and feature points are extracted from the continuous image frame data to form a feature point data set.

[0029] Among them, the device side can be an intelligent service side with video capture capabilities. Among them, the intelligent service side can include, but is not limited to: intelligent teller machines (such as ATMs, VTMs, intelligent counters), self-service business handling terminals (such as self-service account opening machines, mobile business expansion terminals), mobile terminal devices (such as tablets and smartphones dedicated to financial institutions), and remote video teller systems (such as intelligent terminals supporting two-way video interaction). The above intelligent service sides can continuously collect image frame data of the business handling environment in real time through built-in or external high-definition cameras, infrared sensors, or multi-spectral imaging modules. It should be understood that the intelligent service side collects continuous image frame data in real time.

[0030] Among them, the target feature points can be feature points that meet preset conditions. For example, the target feature points can be the edges of objects in the picture. Another example is that the target feature points can be texture feature points with high distinctiveness and resistance to light interference, such as wall decoration textures, device surface markings, etc., which are stable structures not easily affected by minor deformations. By presetting the constraint conditions for feature point collection, the stability and pertinence of the feature point collection process can be effectively improved.

[0031] S120. Perform feature point change analysis on the obtained feature point data set to obtain an image analysis result.

[0032] In this step, perform feature point change analysis on the obtained feature point data set, where the feature point change situations that can be analyzed include, but are not limited to, displacement changes, intensity changes, distribution changes, etc. among the feature points, to obtain an image analysis result for judging whether the background in the continuous image frame data has changed.

[0033] S130. Obtain a background change analysis result according to the image analysis result and a preset judgment method; among them, the background change analysis result includes at least one of the following information: significance judgment information and abnormal marking type.

[0034] In this step, through the preset judgment method, based on the image analysis result, perform background change judgment, and then obtain the background change analysis result. Among them, the background change analysis result can include at least one of the significance judgment information and the abnormal marking information, so as to judge whether the background has changed based on at least one of the above information.

[0035] In this embodiment, the visual risk control detection method for the financial business background environment based on image analysis of the present application can effectively and systematically capture the progressive changes and periodic interferences in the background environment by collecting continuous image frame data and performing feature point collection and change analysis, thereby improving the ability to judge abnormal changes in the background environment. Moreover, through the output significant judgment information and abnormal type marking, the degree of environmental change and risk category can be simultaneously reflected. Compared with the single-time-point feature matching method, it can effectively solve the misjudgment problem caused by ignoring the temporal correlation, and provide multi-dimensional dynamic risk control indicators for the verification of the financial business background.

[0036] Figure 2 It is another flowchart of the visual risk control detection method for the financial business background environment based on image analysis shown in the embodiment of the present application.

[0037] See Figure 2 , the visual risk control detection method for the financial business background environment based on image analysis of the present application includes: S210, obtain continuous image frame data, extract feature points from each frame of image frame data by using the same preset feature extraction method, and obtain a feature point data set; wherein, the preset feature extraction method includes at least one of scale-invariant feature transform, speeded-up robust features, and fast rotation invariant features.

[0038] In this step, feature points are extracted from each positive image frame data by using the same preset feature extraction method, and feature points that meet the same extraction conditions are extracted from each frame of image frame data to establish a highly correlated feature point data set.

[0039] Among them, scale-invariant feature transform (SIFT) detects key points in the image by constructing a Gaussian difference pyramid, and extracts features with rotation and scale invariance in the scale space. The most important thing of scale-invariant feature transform is to calculate the gradient direction histogram in the neighborhood of the key points to generate a 128-dimensional description vector, which is suitable for feature matching in complex scenes; speeded-up robust features (SURF) approximate the key points by using the Hessian matrix, accelerates the calculation by using the integral image, and generates a 64-dimensional description vector through the Haar wavelet response. Compared with scale-invariant feature transform, speeded-up robust features significantly improve the operation efficiency while maintaining rotation and scale invariance, and are suitable for scenarios with high real-time requirements. Fast rotation invariant features (ORB) combine improved FAST corner detection and BRIEF descriptor, achieve rotation invariance through direction compensation, and generate a 32-dimensional binary description vector. Fast rotation invariant features mainly use Hamming distance matching, with extremely high calculation efficiency, and are suitable for mobile devices or resource-constrained environments.

[0040] It should be understood that the feature points obtained through Scale-Invariant Feature Transform (SIFT), Speeded-Up Robust Features (SURF), and Fast Rotation-Invariant Feature (FREAK) may vary, that is, by using these three feature extraction methods, different types of feature points can be extracted. Among them, SIFT can detect blob-like structures, such as regions with rich texture (such as leaves, fabrics), circular markings, etc., and has a weak response to edges; SURF can focus on blob detection, but has a stronger response to the edge-corner transition region, such as door and window frames, regular geometric patterns; FREAK can mainly capture high-curvature corners (such as table corners, intersections of text strokes) and is insensitive to smooth regions.

[0041] As an example, taking the same office scene image as an example, the types of feature points obtained by SIFT, SURF, and FREAK are described as follows: Among them, SIFT may extract feature points at the center of the pattern of the wall decoration painting; SURF may extract feature points at the junction of the picture frame edge and the wall; FREAK extracts feature points at the sharp corners of the office desk and the edges of the keyboard keys.

[0042] It should be noted that the method of the present application can obtain corresponding feature points by using at least one of SIFT, SURF, and FREAK for the same frame of image frame data. That is to say, the present application can use the following combination methods for feature point extraction for the same frame of image frame data: 1. SIFT or SURF or FREAK; 2. SIFT + SURF, or SIFT + FREAK, or SURF + FREAK; 3. SIFT + SURF + FREAK.

[0043] Of course, the present application can also achieve more types of feature point extraction by adding or replacing other feature point extraction methods, which is not limited here.

[0044] When the method of the present application obtains the feature point dataset, the feature points are extracted from each frame of image frame data by using the same extraction method, and different types of feature extraction methods are used for a single frame of image frame data. In this way, the data richness and integrity of the obtained feature point dataset can be ensured, and the accuracy of subsequent feature point change analysis for the feature point dataset can be effectively improved.

[0045] Among them, the extraction of feature points from continuous image frame data can be achieved through a pre-trained feature extraction model. Among them, the pre-trained feature extraction model can use a pre-trained deep convolutional neural network as the feature extraction model, such as HRNet, SuperPoint, which is not limited here. Further, the pre-trained feature extraction model can sample and extract feature points from the edge region of the object in the picture. It can be understood that in the application scenario of forgery scene recognition, in an artificial forgery scene (such as virtual background implantation), the edges of virtual objects generally have phenomena such as blurring, jaggedness or discontinuity. Sampling with the object edge as the guide can effectively enhance the sensitivity to the recognition of forgery scenes.

[0046] It should be understood that the pre-trained feature extraction model can use at least one of the aforementioned scale-invariant feature transform, speeded-up robust features, and fast rotation-invariant features for feature extraction, or it can also be a combination of algorithms such as the Canny edge detection algorithm or a pre-trained edge detection model (such as HED) and the above at least one feature extraction method to achieve the extraction of the strong edge contour of the object in the image.

[0047] S220, determine the reference frame feature points and the comparison frame feature points that match the reference frame feature points from the feature point dataset.

[0048] In this step, the reference frame feature points serving as the comparison benchmark and the comparison frame feature points for comparison calculation are determined from the obtained feature point dataset.

[0049] It should be understood that each frame of image frame data can determine the feature points belonging to the same frame of image frame data. Among them, the reference frame feature points can be the feature points extracted from the image frame data that is the first in time order among the continuous image frame data, and the comparison frame feature points are the feature points corresponding to the other image frame data except the first in the sorting.

[0050] S230, compare the comparison frame feature points with the reference frame feature points to obtain an image analysis result.

[0051] In this step, through data analysis of the comparison result between the obtained comparison frame feature points and the reference frame feature points, an image analysis result that can be used to judge background changes is obtained.

[0052] It should be understood that the process of comparing the comparison frame feature points with the reference frame feature points is to compare according to the feature points corresponding to different frames of image frame data.

[0053] Among them, the image frame data corresponding to the reference frame feature points is defined as the reference image, and the image frame data corresponding to the comparison frame feature points is defined as the comparison image. When comparing the comparison frame feature points with the reference frame feature points, it can be along the time axis direction. Sequentially, the comparison frame feature points corresponding to the comparison images distributed along the time axis direction are compared with the reference frame feature points corresponding to the reference image.

[0054] Among them, the image analysis result can include one or more of change curve parameters, significance judgment information, and image geometric change results.

[0055] S240. According to the image analysis result and the preset judgment method, obtain the background change analysis result; among them, the background change analysis result includes at least one of significance judgment information and abnormal marking information.

[0056] In this step, through the preset judgment method, based on the image analysis result, background change judgment is performed, and then the background change analysis result is obtained. Among them, the background change analysis result can include at least one of significance judgment information and abnormal marking information. Based on this, it can be determined whether the background has changed according to at least one of the above information.

[0057] In this embodiment, the visual risk control detection method for the financial business background environment based on image analysis of the present application extracts feature points by using the same preset feature extraction method for the acquired continuous image frame data, and different types of feature extraction methods are used for single-frame image frame data to effectively make the data richness and integrity of the acquired feature point data set, and can effectively improve the accuracy of subsequent feature point change analysis for the feature point data set; by first determining the reference frame feature points and the comparison frame feature points, obtaining the image analysis result according to the comparison result of comparing the comparison frame feature points with the reference frame feature points, and using the feature point change trend and change intensity represented by the image analysis result for evaluation, the background change analysis result can be obtained, which can effectively improve the accuracy of judging background abnormal changes.

[0058] Figure 3 It is another flow schematic diagram of the visual risk control detection method for the financial business background environment based on image analysis shown in the embodiment of the present application.

[0059] See Figure 3 This visual risk control detection method for the financial business background environment based on image analysis of the present application includes: S310. Acquire continuous image frame data, and extract feature points from each frame of image frame data by using the same preset feature extraction method to obtain a feature point data set, where the preset feature extraction method includes at least one of scale-invariant feature transform, speeded-up robust features, and fast rotation-invariant features.

[0060] In this step, the same preset feature extraction method is used to extract feature points from each positive image frame data, and feature points that meet the same extraction conditions are extracted from each frame of image frame data to establish a feature point data set with high correlation.

[0061] S320. Determine the reference frame feature points and the comparison frame feature points that match the reference frame feature points from the feature point data set.

[0062] In this step, the reference frame feature points serving as the comparison benchmark and the comparison frame feature points for comparison calculation are determined from the obtained feature point data set.

[0063] S330. Calculate the feature point vector parameters between the reference frame feature points and the comparison frame feature points, and calculate the feature point intensity parameters between the reference frame feature points and the comparison frame feature points.

[0064] In this step, after determining the reference frame feature points and the comparison frame feature points in the feature point data set, by comparing the comparison frame feature points with the reference frame feature points, calculate the feature point vector parameters between the comparison frame feature points and the reference frame feature points, and calculate the feature point vector parameters between the comparison frame feature points and the reference frame feature points.

[0065] Among them, the feature point vector parameter corresponds to the displacement vector of the feature point, and the feature point intensity parameter corresponds to the change intensity of the feature point. It should be understood that the feature point vector parameter and the feature point intensity parameter are obtained by separately comparing the comparison frame feature points matched with the reference frame feature points of the reference image. For example, the displacement vector and the change intensity between the feature points corresponding to the two matched image frame data can be obtained by taking the difference after comparison.

[0066] S340. After normalizing the feature point vector parameters and the feature point intensity parameters, construct a feature change curve with time as the horizontal axis.

[0067] In this step, after normalizing the obtained feature point vector parameters and the feature point intensity parameters, construct a feature change curve corresponding to the displacement vector and the change intensity of the corresponding feature points with time as the horizontal axis.

[0068] It should be understood that the constructed feature change curve can correspond to the feature point vector parameter and the feature point intensity parameter respectively, and different feature change curves are constructed according to the feature point vector parameter and the feature point intensity parameter respectively.

[0069] S350. Calculate the preset type parameters of the feature change curve to obtain the change curve parameters.

[0070] In this step, according to the obtained characteristic change curve, the preset parameters in the characteristic change curve are calculated to obtain the change curve parameters that meet the preset requirements.

[0071] Among them, the preset type parameters may include but are not limited to at least one of the mean, standard deviation, and peak frequency. It should be understood that the change curve parameters include at least one of the mean, standard deviation, and peak frequency.

[0072] S360. Compare the change curve parameters with the preset dynamic threshold to determine the significance judgment information.

[0073] In this step, the calculated change curve parameters are compared with the preset dynamic threshold set in advance to determine whether the change of the characteristic points corresponding to the change curve parameters is a normal change or an abnormal change, and the significance judgment information is determined according to the comparison result.

[0074] Among them, the significance judgment information may include: dynamic threshold, significance judgment information, and abnormal judgment result.

[0075] As an example, the following uses X t to represent the change curve, μ represents the mean of the curve, σ represents the standard deviation of the curve, and describes the setting process of the dynamic threshold and the process of judging whether the change of the characteristic points corresponding to the change curve parameters is a normal change or an abnormal change through the dynamic threshold: The dynamic threshold formula is as follows:

[0076] Among them, k is a hyperparameter; among them, k can be set to different values according to different type parameters of the characteristic points corresponding to the change curve, for example, set to 2; When if X t ≤ Threshold, it means that the change of the characteristic points corresponding to the change curve parameters is a normal change, otherwise it means an abnormal change, that is, it means that there is an abnormality in the background change. When it is judged that there is an abnormality in the background change, an abnormal judgment result information indicating the existence of an abnormality is generated in the significance judgment information.

[0077] S370. When it is determined that there is an abnormality in the abnormal judgment result in the significance judgment information, the image frame data associated with the abnormal change is input into the preset image classification model for abnormal classification recognition to obtain the abnormal marking information.

[0078] In this step, when it can be determined according to the abnormal judgment result in the significance judgment information that there is an abnormality in the background change in the current continuous image frame data, the image frame data associated with the abnormal change is input into the pre-trained preset image classification model for automatic abnormal classification recognition, and the preset image classification model outputs the abnormal marking information.

[0079] Among them, the preset image classification model can be implemented using a convolutional neural network in related technologies. For example, the preset image classification model is a CNN-RNN hybrid architecture, which will not be elaborated here.

[0080] Among them, the abnormal marking information can characterize the type of background abnormal change. Among them, the type of abnormal change can include but is not limited to: scene forgery, abnormal change, and fraud behavior. Among them, scene forgery can refer to the type of abnormal change where the background is replaced with a false scene (such as green screen technology); abnormal change can refer to the type of abnormal change where the environment is changed to try to cover up the real background; fraud behavior can refer to the type of abnormal change where a consistent background is forged by adjusting light, scenery, etc.

[0081] Among them, it is also possible to calculate the feature point distribution parameter between the reference frame feature points and the comparison frame feature points by comparing the comparison frame feature points with the reference frame feature points. Among them, the feature point distribution parameter can include but is not limited to: the change in the distribution density of feature points. Among them, the geometric change between images is determined through the feature point distribution parameter, for example, determining whether there is deformation, rotation, or translation between images.

[0082] Among them, when the saliency judgment information includes abnormal information, it is also possible to input the image frame data associated with the abnormal change and the feature point distribution parameter into the preset image classification model for abnormal classification recognition to obtain abnormal change information. By inputting the image frame data associated with the abnormal change and the feature point distribution parameter into the preset image classification model for abnormal classification recognition together, the geometric change result between images can be obtained using the feature point distribution parameter, which further effectively improves the recognition accuracy of the preset image classification model for abnormal classification recognition. Of course, the preset image classification model can also combine the environmental temperature parameter synchronously collected together with the continuous image frame data for abnormal classification recognition. For example, the environmental temperature parameter corresponding to the continuous image frame data is obtained through the thermal imaging information acquisition technology in related technologies, so that the preset image classification model can use more-dimensional data for abnormal classification recognition.

[0083] S380. Output the background change analysis result according to the saliency judgment information and the abnormal change information.

[0084] In this step, according to the obtained saliency judgment information and the abnormal change information output by the model, the background change analysis result is output.

[0085] Among them, the background change analysis result can be output in the format of a dictionary. For example, the background change analysis result is output in the form of a dictionary: { "Change threshold": 70, "Saliency score": 78, "Whether abnormal": true, "Abnormal flag": "Scene forgery" } Of course, the Beijing change analysis can also be output in other ways, such as in text form, which is not restricted here.

[0086] In some embodiments, after obtaining the background change analysis result, a reminder message can also be displayed on a preset interface.

[0087] Among them, the reminder message can correspond to the background change analysis result.

[0088] As an example, the abnormal flag type and significance score threshold in the background change analysis result are used to generate a hierarchical alarm message and display it in real time on the business handling display interface. For high-risk events (such as a significance score exceeding 70 points and marked as "Scene forgery"), the interface will trigger a full-screen red mask warning. At the same time, a dynamic pop-up window will be suspended at the top of the operation window. The pop-up window will embed an abnormal type icon (such as the dynamic effect of a virtual shield breaking corresponding to the forged scene), a risk level label (such as "Severely abnormal"), and a highlighted contour box of the abnormal area (rendered in real time through a semantic segmentation algorithm), and synchronously play a preset voice warning content (such as "Environmental abnormality detected, please immediately abort the business"); for low-risk events (such as a score lower than 70 points), the interface will display the content "Background detection is normal, please continue with business handling".

[0089] In some embodiments, after obtaining the background change analysis result, a preset business control instruction can also be triggered. Among them, the preset business control instruction can include but is not limited to: data interception instructions, security verification instructions, data retention instructions, and device control instructions. Among them, the data interception instruction is used to automatically pause or terminate the current owner process, such as automatically pausing the current transaction process, freezing the electronic signature operation permission, and sending a transaction suspension instruction code to the financial core system (such as the HTTP 451 status code extension protocol); the security verification instruction is used to start business security verification, such as starting a biometric secondary verification process (invoking a live detection interface) or triggering a request for remote video customer service intervention; the data retention instruction is used to save the current business handling data and send it to a specified platform for storage, such as uploading the image frame data determined to have background abnormal changes to the business server; the device control instruction is used to control the acquisition device to adjust the data acquisition action, such as controlling the camera focus adjustment of the acquisition device to obtain a wider range of image frame data.

[0090] In this embodiment, the visual risk control detection method for the financial business background environment based on image analysis of the present application performs joint analysis based on multi-dimensional data such as feature point vector parameters, intensity parameters, and distribution parameters. Moreover, by combining normalized feature curve modeling and adaptive dynamic threshold evaluation, it can effectively distinguish real scene structure changes from instantaneous interferences, accurately identify background forgery behaviors such as scene forgery, abnormal changes, and fraud behaviors, and can effectively improve the accuracy and reliability of abnormal detection in the financial business background environment.

[0091] Corresponding to the foregoing embodiment of the application function implementation method, the present application also provides a visual risk control detection system for the financial business background environment based on image analysis, an electronic device, and corresponding embodiments.

[0092] Figure 4 It is a schematic structural diagram of the visual risk control detection system for the financial business background environment based on image analysis shown in the embodiment of the present application.

[0093] See Figure 4 , the visual risk control detection system 400 for the financial business background environment based on image analysis of the present application includes a data acquisition module 410, a feature analysis module 420, and a result analysis module 430.

[0094] The data acquisition module 410 is used to obtain continuous image frame data and extract a feature point data set from the continuous image frame data; wherein, the feature point data set includes: target feature points of each frame of image frame data.

[0095] In some embodiments, the data acquisition module 410 can be used to obtain continuous image frame data, extract feature points from each frame of image frame data using the same preset feature extraction method, and obtain a feature point data set. Wherein, the preset feature extraction method includes at least one of scale-invariant feature transform, speeded-up robust features, and fast rotation invariant features.

[0096] The feature analysis module 420 is used to perform feature point change analysis on the obtained feature point data set to obtain an image analysis result.

[0097] In some embodiments, the feature analysis module 420 can be used to determine reference frame feature points and comparison frame feature points that match the reference frame feature points from the feature point data set, compare the comparison frame feature points with the reference frame feature points; and perform data analysis based on the obtained comparison result to obtain an image analysis result.

[0098] In some embodiments, the feature analysis module 420 may also be used to calculate the feature point vector parameters of the reference frame feature points and the comparison frame feature points, and compare the feature point vector parameters between the comparison frame feature points and the reference frame feature points; and / or calculate the feature point intensity parameters of the reference frame feature points and the comparison frame feature points, and compare the feature point intensity parameters between the comparison frame feature points and the reference frame feature points.

[0099] In some embodiments, when the obtained comparison results include the comparison results of the feature point vector parameters and / or the comparison results of the feature point intensity parameters, the feature analysis module 420 may also be used to normalize the comparison results of the feature point vector parameters and / or the comparison results of the feature point intensity parameters, and then construct a feature change curve with time as the horizontal axis.

[0100] In some embodiments, the feature analysis module 420 may also be used to calculate the feature point distribution parameters of the reference frame feature points and the comparison frame feature points, and compare the feature point distribution parameters between the comparison frame feature points and the reference frame feature points; when the obtained comparison results include the comparison results of the feature point distribution parameters, determine the image geometric change results according to the comparison results of the feature point distribution parameters.

[0101] The result analysis module 430 is used to obtain the background change analysis results according to the image analysis results and the preset judgment method; wherein, the background change analysis results include at least one of the significance judgment information and the abnormal marking information.

[0102] In some embodiments, the result analysis module 430 may also be used to calculate the preset type parameters of the feature change curve to obtain the change curve parameters; compare the change curve parameters with the preset dynamic threshold and perform significance evaluation to determine the significance judgment information; when the significance judgment information includes abnormal information, input the image frame data associated with the abnormal change into the preset image classification model for abnormal classification and recognition; output the background change analysis results according to the significance judgment information and the abnormal change type.

[0103] In some embodiments, the system further includes a result response module 440, and the result response module 440 is used to display reminder information and / or trigger a preset service control instruction on a preset interface according to the background change analysis results.

[0104] In this embodiment, the visual risk control detection method for the financial business background environment based on image analysis of the present application can effectively and systematically capture the progressive changes and periodic interferences in the background environment by collecting continuous image frame data and performing feature point collection and change analysis, thereby improving the ability to judge abnormal changes in the background environment. Moreover, through the output of the significance judgment information and the abnormal type marking, the degree of environmental change and the risk category can be simultaneously reflected. Compared with the single-time-point feature matching method, the problem of misjudgment caused by ignoring the temporal correlation can be effectively solved, providing multi-dimensional dynamic risk control indicators for the verification of the financial business background.

[0105] Regarding the device in the above embodiment, the specific manners in which each module performs operations have been described in detail in the embodiment related to the method, and will not be elaborated here.

[0106] Figure 5 It is a schematic structural diagram of an electronic device shown in an embodiment of the present application.

[0107] See Figure 5 , the electronic device 1000 includes a memory 1010 and a processor 1020.

[0108] The processor 1020 may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0109] The memory 1010 may include various types of storage units, such as system memory, read-only memory (ROM), and permanent storage devices. Among them, the ROM may store static data or instructions required by the processor 1020 or other modules of the computer. The permanent storage device may be a readable and writable storage device. The permanent storage device may be a non-volatile storage device that does not lose the stored instructions and data even when the computer is powered off. In some embodiments, the permanent storage device employs a mass storage device (such as a magnetic or optical disk, flash memory) as the permanent storage device. In some other embodiments, the permanent storage device may be a removable storage device (such as a floppy disk, optical drive). The system memory may be a readable and writable storage device or a volatile readable and writable storage device, such as dynamic random access memory. The system memory may store some or all of the instructions and data required by the processor during operation. In addition, the memory 1010 may include any combination of computer-readable storage media, including various types of semiconductor storage chips (such as DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), and magnetic disks and / or optical disks may also be used. In some embodiments, the memory 1010 may include a removable storage device that is readable and / or writable, such as a compact disc (CD), read-only digital versatile disc (such as DVD-ROM, dual-layer DVD-ROM), read-only Blu-ray disc, super density disc, flash memory card (such as SD card, min SD card, Micro-SD card, etc.), magnetic floppy disk, etc. The computer-readable storage medium does not include carrier waves and instantaneous electronic signals transmitted wirelessly or wired.

[0110] Executable code is stored on the memory 1010, and when the executable code is processed by the processor 1020, it may cause the processor 1020 to execute some or all of the methods described above.

[0111] In addition, the method according to the present application may also be implemented as a computer program or a computer program product, which includes computer program code instructions for executing some or all of the steps of the above method of the present application.

[0112] Alternatively, the present application may also be implemented as a computer-readable storage medium (or non-transitory machine-readable storage medium or machine-readable storage medium), on which executable code (or computer program or computer instruction code) is stored. When the executable code (or computer program or computer instruction code) is executed by a processor of an electronic device (or a server, etc.), it causes the processor to execute some or all of the steps of the above method according to the present application.

[0113] The embodiments of the present application have been described above. The above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations are obvious to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The choice of terms used herein is intended to best explain the principles of the embodiments, the practical application, or the improvement of the technology in the market, or to enable other ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A visual risk control detection method for financial business background environment based on image analysis, characterized in that: include: Acquire continuous image frame data, and extract a feature point data set from the continuous image frame data; wherein the feature point data set includes: target feature points of the image frame data of each frame; Performing feature point change analysis on the acquired feature point data set to obtain an image analysis result; According to the image analysis result and a preset judgment method, a background change analysis result is obtained; wherein the background change analysis result includes: at least one of significance judgment information and abnormal marking information.

2. The method according to claim 1, characterized in that The acquiring of continuous image frame data and extracting a feature point data set from the continuous image frame data includes: Continuous image frame data are acquired, and feature points are extracted from the image frame data of each frame using the same preset feature extraction method to acquire a feature point data set; wherein the preset feature extraction method includes at least one of: scale-invariant feature transformation, accelerated robust feature, and fast rotation-invariant feature.

3. The method according to claim 1, characterized in that The step of performing feature point change analysis on the acquired feature point data set to obtain an image analysis result includes: Determine, from the feature point data set, a reference frame feature point and a comparison frame feature point that matches the reference frame feature point; The feature points of the comparison frame are compared with the feature points of the reference frame to obtain an image analysis result.

4. The method according to claim 3, characterized in that The comparing the feature points of the comparison frame with the feature points of the reference frame to obtain an image analysis result includes: Calculating feature point vector parameters between the reference frame feature points and the comparison frame feature points; and / or calculating feature point intensity parameters between the reference frame feature points and the comparison frame feature points; After normalizing the feature point vector parameters and / or the feature point intensity parameters, a feature change curve is constructed with time as the horizontal axis.

5. The method according to claim 4, characterized in that The obtaining of background change analysis results according to the image analysis results and a preset judgment method includes: Calculating preset type parameters of the characteristic change curve to obtain change curve parameters; Comparing the change curve parameter with a preset dynamic threshold to determine significance judgment information; the significance judgment information includes: a dynamic threshold, significance judgment information, and an abnormality judgment result; When the abnormality judgment result in the significance judgment information determines that an abnormality exists, the image frame data associated with the abnormal change is input into a preset image classification model for abnormality classification and recognition to obtain abnormality marking information; Outputting a background change analysis result according to the significance judgment information and the abnormal marking information.

6. The method according to claim 5, characterized in that The comparing the feature points of the comparison frame with the feature points of the reference frame to obtain an image analysis result further includes: Calculating feature point distribution parameters between the reference frame feature points and the comparison frame feature points; When the significance judgment information includes abnormal information, the image frame data associated with the abnormal change is input into a preset image classification model for abnormal classification and recognition to obtain abnormal change information, including: When the significance judgment information includes abnormality information, the image frame data associated with the abnormal change and the feature point distribution parameters are input into a preset image classification model for abnormality classification and recognition to obtain abnormal change information.

7. The method according to any one of claims 1 to 6, characterized in that: Also includes: According to the background change analysis result, a reminder message is displayed on a preset interface and / or a preset business control instruction is triggered.

8. A financial business background environment visual risk control detection system based on image analysis, characterized in that: include A data acquisition module is used to acquire continuous image frame data and extract a feature point data set from the continuous image frame data; wherein the feature point data set includes: target feature points of the image frame data of each frame; A feature analysis module, used to perform feature point change analysis on the acquired feature point data set to obtain an image analysis result; The result analysis module is used to obtain the background change analysis result according to the image analysis result and the preset judgment method; wherein the background change analysis result includes: at least one of the significance judgment information and the abnormal mark type.

9. An electronic device, characterized in that: include: processor; as well as A memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having executable code stored thereon, characterized in that: When the executable code is executed by a processor of an electronic device, the processor is caused to execute the method as claimed in any one of claims 1 to 7.

Images