CN-116152509-B - Image-based risk mining method, device, storage medium and equipment

CN116152509BCN 116152509 BCN116152509 BCN 116152509BCN-116152509-B

Abstract

The invention provides a risk mining method, device, storage medium and equipment based on images. The method comprises the steps of obtaining a plurality of existing service orders, wherein each service order comprises user images, extracting background image features of the user images in each existing service order to obtain an image feature library, clustering the background image features in the image feature library to obtain a plurality of image feature classes, each image feature class corresponds to one background image feature set, filtering error classes and/or risk-free classes in the plurality of image feature classes to obtain risk classes, and forming the risk image feature library. The categories in the obtained risk image feature library have risk and are consistent with the business requirements of anti-fraud risk identification. The categories in the risk image feature library are screened, so that the risk is achieved, and the categories are not excessively fragmented.

Inventors

WANG CHUN
MENG FANYU
CHENG ANMIN
FU DENGGUO
ZHOU XUNYI

Assignees

北京中关村科金技术有限公司

Dates

Publication Date: 20260512
Application Date: 20220908

Claims (10)

1. An image-based risk mining method, comprising: Acquiring a plurality of existing service orders, wherein the service orders comprise user images; Extracting background image features of user images in all existing service orders to obtain an image feature library; clustering the background image features in the image feature library to obtain a plurality of image feature classes, wherein each image feature class corresponds to a background image feature set; filtering out error classes and/or risk-free classes in the plurality of image feature classes to obtain risk classes, and forming a risk image feature library; the method further comprises the steps of: clustering background image features in the image feature library by using a clustering center of risk classes obtained according to a plurality of existing service orders as seeds and adopting a clustering algorithm of specified class quantity to obtain an updated background image feature set corresponding to the risk classes so as to form an updated risk image feature library, wherein the specified class quantity is the same as the risk class quantity in the risk image feature library, or Acquiring a plurality of risk service orders; extracting background image features of the user images in the risk service orders to obtain a risk image feature library; clustering the background image features in the risk image feature library to obtain at least one risk class; clustering background image features in the image feature library by using a clustering center of risk classes obtained according to a plurality of risk service orders as seeds and adopting a clustering algorithm of specified class quantity to obtain an updated background image feature set corresponding to the risk classes so as to form an updated risk image feature library, wherein the specified class quantity is the same as the risk class quantity in the risk image feature library.
2. The image-based risk mining method according to claim 1, wherein the extracting background image features of the user image in each existing service order to obtain an image feature library comprises: Carrying out image segmentation on the user image in each existing service order, and extracting a corresponding background image; determining whether each extracted background image meets preset characteristic information conditions; and extracting background image features by utilizing a pre-trained image feature extraction model aiming at the background image meeting the preset feature information condition to obtain an image feature library.
3. The image-based risk mining method according to claim 2, wherein the predetermined feature information condition includes that an area ratio of the background is greater than a first predetermined threshold and/or that texture features of the background image meet a predetermined requirement.
4. The image-based risk mining method of claim 1, wherein the clustering of background image features in the image feature library using a clustering algorithm of a specified number of categories comprises: Determining a most similar cluster center for each background image feature in the image feature library; Taking out background image features with the similarity with the most similar cluster center meeting a third preset threshold from the image feature library, and putting the background image features into a candidate list of risk classes corresponding to the most similar cluster center; For each risk class, calculating a temporary clustering center of the risk class according to the background image feature set of the risk class and the background image features in the candidate list; Calculating the similarity between the background image features in the candidate list of the risk class and the temporary clustering center; updating background image features of which the similarity with the temporary clustering center in the candidate list meets a fourth preset threshold to the risk class and updating the clustering center of the risk class, and replacing the background image features of which the similarity with the temporary clustering center in the candidate list does not meet the fourth preset threshold with the image feature library; and circularly executing the clustering process until all risk classes are not changed or the preset circulation times are reached.
5. The image-based risk mining method of claim 1, further comprising: acquiring a new service order; Extracting background image characteristics of the user image in the new service order; matching the extracted background image features with the background image features in the risk image feature library to obtain N background image features most similar to the extracted background image features; And counting the risk classes to which the N background image features belong, and determining the most risk class to which the N background image features belong as the risk class corresponding to the background image features of the user image in the new service order.
6. The image-based risk mining method of claim 5, wherein the matching the extracted background image features with the background image features in the risk image feature library to obtain N background image features most similar to the extracted background image features comprises: calculating the similarity between the extracted background image features and each background image feature in the risk image feature library to obtain the most similar first M background image features; And filtering out the background image features with insufficient similarity of a preset measurement threshold value from the M background image features to obtain the final N most similar background image features.
7. The image-based risk mining method according to claim 5, wherein when there are a plurality of user images in the new service order, the matching the extracted background image features with the background image features in the risk image feature library to obtain N background image features most similar to the extracted background image features includes: respectively matching the background image characteristics extracted according to each user image in the plurality of user images with the background image characteristics in the risk image characteristic library to obtain N most similar background image characteristics corresponding to the background image characteristics extracted by each user image; And integrating N most similar background image features corresponding to the background image features extracted according to the user images, sequencing, and taking the first N background image features as the N most similar background image features.
8. An image-based risk mining apparatus, comprising: the system comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring a plurality of existing service orders, and the service orders comprise user images; the extraction module is used for extracting background image features of the user images in all the existing service orders to obtain an image feature library; the clustering module is used for clustering the background image features in the image feature library to obtain a plurality of image feature classes, and each image feature class corresponds to one background image feature set; the database building module is used for filtering error classes and/or risk-free classes in the plurality of image feature classes to obtain risk classes and form a risk image feature database; The mining module is used for clustering background image features in the image feature library by using a clustering algorithm with the number of specified categories by taking a clustering center of risk categories obtained according to a plurality of existing service orders as a seed to obtain an updated background image feature set corresponding to the risk categories so as to form an updated risk image feature library, wherein the number of the specified categories is the same as the number of risk categories in the risk image feature library, or The mining module is used for acquiring a plurality of risk service orders; extracting background image features of the user images in the risk service orders to obtain a risk image feature library; clustering the background image features in the risk image feature library to obtain at least one risk class; clustering background image features in the image feature library by using a clustering center of risk classes obtained according to a plurality of risk service orders as seeds and adopting a clustering algorithm of specified class quantity to obtain an updated background image feature set corresponding to the risk classes so as to form an updated risk image feature library, wherein the specified class quantity is the same as the risk class quantity in the risk image feature library.
9. A computer readable storage medium, having stored thereon a computer program which, when executed by one or more processors, implements the method of any of claims 1 to 7.
10. A computer device comprising a memory and one or more processors, the memory having stored thereon a computer program which, when executed by the one or more processors, implements the method of any of claims 1 to 7.

Description

Image-based risk mining method, device, storage medium and equipment Technical Field The present invention relates to the field of computer technologies, and in particular, to an image-based risk mining method, apparatus, storage medium, and device. Background In the existing face identity verification process and other similar business scenes, the system guides and collects 1-N pieces of face image information of the user, and is used for judging whether the user is a true person (instead of a photo) and the user (consistent with the stated identity, instead of other people). In this procedure, the value of the background information (non-face or human body part) in the image is not effectively utilized. In addition to the face information in the image, the background information is also of great value. Many of the identity-spoofed or intermediated orders found in business appear aggregated from relatively fixed geographic locations and rooms. For the same intermediate-hosted order, the background of the user images representing these high-risk orders has similarity. Therefore, by mining orders with similar background parts in the image, unknown intermediate office points and/or fraudulent party places can be mined in an assisted mode, whether a new business order comes from a certain known intermediate office point and/or other party places can be judged online, and risks can be found in advance. In the related art, image information is classified by KNN classification (classification) instead of clustering (clustering) in the true sense, that is, the current image sample is essentially attributed to an existing class in a library by using the KNN classification technology, and if the attributed class cannot be found due to a similarity threshold and other reasons, the current image sample is constructed into a new class. However, such a category construction method has low association with the service, and when the search fails to find the attribution category, i.e. the new category is created, the category splitting is excessively fragmented, and meanwhile, the service risk cannot be fully mined because the association between the categories in the library cannot be discovered. In summary, there is a need in the art to solve the problem of how to effectively use the background information in the face image to mine out the business risk. Disclosure of Invention In order to solve the problems, the invention provides an image-based risk mining method, an image-based risk mining device, a storage medium and image-based risk mining equipment. In a first aspect, an embodiment of the present invention provides an image-based risk mining method, including: Acquiring a plurality of existing service orders, wherein the service orders comprise user images; Extracting background image features of user images in all existing service orders to obtain an image feature library; clustering the background image features in the image feature library to obtain a plurality of image feature classes, wherein each image feature class corresponds to a background image feature set; and filtering out error classes and/or risk-free classes in the plurality of image feature classes to obtain risk classes, thereby forming a risk image feature library. In some implementations, the extracting the background image features of the user image in each existing service order to obtain an image feature library includes: Carrying out image segmentation on the user image in each existing service order, and extracting a corresponding background image; determining whether each extracted background image meets preset characteristic information conditions; and extracting background image features by utilizing a pre-trained image feature extraction model aiming at the background image meeting the preset feature information condition to obtain an image feature library. In some implementations, the preset feature information condition includes that an area ratio of the background is greater than a first preset threshold value and/or that texture features of the background image meet a preset requirement. In some implementations, the method further comprises: Clustering background image features in the image feature library by using a clustering center of risk classes obtained according to a plurality of existing service orders as seeds and adopting a clustering algorithm of specified class quantity to obtain a background image feature set corresponding to updated risk classes so as to form an updated risk image feature library, wherein the specified class quantity is the same as the risk class quantity in the risk image feature library. In some implementations, the method further comprises: Acquiring a plurality of risk service orders; extracting background image features of the user images in the risk service orders to obtain a risk image feature library; clustering the background image features in the risk image feature library to obtain at least