Article Details

Abstract

To address high-frequency fraud issues in health insurance, such as forging or tampering with medical documents and reports, this paper proposes an intelligent detection system integrating image tampering localization and multimodal large models. The system adopts a two-stage framework: first, an image tampering localization model performs pixel-level analysis on medical document images to locate tampering traces and generate corresponding masks; subsequently, the original images and masks are jointly fed into a multimodal large model to collaboratively analyze visual content, textual information, and the semantics of tampered regions for comprehensive fraud intent identification (e.g., key numerical tampering). Experiments on an internal dataset demonstrate that the system significantly outperforms single-modal methods in fraud detection performance, providing an efficient and automated solution for health insurance anti-fraud efforts.

Keywords

insurance fraud detection; image tampering localization; multimodal large model; system design

References

1. [1] Wang Weiguo Preliminary Study on the Participation of Commercial Insurance Companies in the Governance of Social Medical Insurance Fraud [J]. Journal of Zhengzhou University of Aeronautical Industry Management (Social Science Edition), 2025 (S1): 28-30
2. [2] Zhou Zili, Wang Kexin, Zhang Xin, etc Research on Typical Forms and Regulatory Measures of Medical Insurance Fraud: Based on a Study of 179 Typical Cases of Insurance Fraud Publicly Issued by the National Medical Security Administration [J]. Health Soft Science, 2025, 39 (2): 73-78
3. [3] Rombach R，Andreas B，Dominik L，et al. High-Resolution Image Synthesis With Latent Diffusion Models [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，2022：10684-10695.
4. [4] Wang Xian, Cui Caixia Research on Car Insurance Fraud Detection Based on GAN and Random Forest Fusion [J]. Information Technology and Informatization, 2025 (8): 103-107
5. [5] Wang Yuxuan Machine Learning Helps Anti Fraud Research in Health Insurance [D]. Taiyuan: Shanxi University of Finance and Economics, 2025
6. [6] Yang J X，Chen K，Ding K，et al. Auto Insurance Fraud Detection with Multimodal Learning [J].Data Intelligence，2023，5（2）：388-412.
7. [7] Su Y，Tan S Q，Huang J W. A Novel Universal Image Forensics Localization Model Based on Image Noise and Segment Anything Model [C]//IH&MMSec'24: Proceedings of the 2024 ACM Workshop on Information Hiding and Multimedia Security，2024：149-158.
8. [8] Kirillov A，Mintun E，Ravi N，et al. Segment Anything [C]//2023 IEEE/CVF International Conference on Computer Vision (ICCV)，2023：4015-4026.
9. [9] Vaswani A，Shazeer N，Parmar N，et al. Attention is All you Need [C]//NIPS'17: Proceedings of the 31st International Conference on Neural Information Processing Systems，2017：6000-6010.
10. [10] Guillaro，Fabrizio，Davide C，et al. TruFor: Leveraging All-Round Clues for Trustworthy Image Forgery Detection and Localization [C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，2023：20606-20615.

About the authors

Xu Xiaohan (1987-), female, Han ethnicity, native of Jingmen, Hubei Province, intermediate economist, master's student, research direction: insurance intelligent risk control.