An online adaptive face deepfake detection method and system

By using feature mining based on counterfactual verification and dynamic prototype evolution components for evidence uncertainty, combined with a lightweight architecture, the lag and local overfitting problems of large parameter models are solved, achieving low-overhead and efficient face forgery detection.

CN122244920APending Publication Date: 2026-06-19QINGDAO HARBIN INSTITUTE OF TECHNOLOGY (WEIHAI)
0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
QINGDAO HARBIN INSTITUTE OF TECHNOLOGY (WEIHAI)
Filing Date
2026-03-17
Publication Date
2026-06-19

AI Technical Summary

⚠Technical Problem

Existing face forgery detection technologies rely on large parameter models, resulting in high computational consumption, delayed updates, susceptibility to local overfitting, and poor generalization ability, failing to meet the real-time and robustness requirements in complex dynamic environments.

⚗Method used

By employing a feature mining component based on counterfactual verification and a dynamic prototype evolution component based on evidence uncertainty, counterfactual samples are generated through frequency domain mixing and attention extraction. Combined with a lightweight frozen ViT base and LoRA adaptation component, adaptive updates and identification of novel forgery attacks are achieved.

🎯Benefits of technology

It reduces the consumption of computing resources, enables accurate identification and real-time updates of face forgery, and improves the model's ability to identify new types of attacks and its generalization performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
Patent Text Reader

Abstract

This application provides an online adaptive deepfake face detection method and system, which solves the technical problems of existing face forgery detection methods that rely on large parameter models, resulting in high computational consumption, slow updates, susceptibility to local overfitting, and poor generalization ability. It includes acquiring a face image to be detected and a reference real image; inputting these into a feature mining component based on counterfactual verification to obtain counterfactual samples; inputting the face image to be detected and the corresponding counterfactual samples into a dynamic prototype evolution component based on evidence uncertainty for evolution to obtain the face forgery detection result, and adaptively updating new forgery attacks. This application can be widely applied in the field of forged face recognition technology.
Need to check novelty before this filing date? Find Prior Art