A target object counterfeit detection method based on a multi-modal large model

By employing a cross-attention mechanism in a multimodal large model, deep fusion of spatial and frequency domain features is achieved, addressing the insufficient face forgery detection capabilities in existing technologies and improving the detection's ability to distinguish forgeries and its generalization effect.

CN122176767APending Publication Date: 2026-06-09SHENZHEN JIUNIU YIMAO INTELLIGENT IOT TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SHENZHEN JIUNIU YIMAO INTELLIGENT IOT TECH CO LTD
Filing Date
2026-01-28
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

Existing technologies fail to effectively integrate spatial and frequency domain features in face forgery detection, resulting in insufficient detection capabilities.

Method used

A multimodal large model is adopted, and the spatial domain feature tensor and frequency domain feature tensor are mapped to query sequence and key value sequence respectively through cross attention mechanism to generate context feature vector, thereby realizing deep fusion of the two at the semantic level and dynamically and adaptively extracting information.

Benefits of technology

It improves the ability to detect and distinguish fake faces, enhances the recognition effect of fake faces, and has stronger generalization ability and dynamic semantic content perception ability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122176767A_ABST
    Figure CN122176767A_ABST
Patent Text Reader

Abstract

The application relates to the technical field of target object forgery detection, and discloses a target object forgery detection method based on a multimodal large model, an electronic device and a computer readable storage medium, the method comprising the following steps: acquiring a target RGB image, extracting a spatial domain feature of the target RGB image, and obtaining a spatial domain feature tensor; converting the target RGB image into a frequency domain representation image, extracting a frequency domain feature of the frequency domain representation image, and obtaining a frequency domain feature tensor; taking one of the spatial domain feature tensor and the frequency domain feature tensor as a query feature tensor and taking the other as a key-value pair feature tensor; mapping the query feature tensor into a query sequence and mapping the key-value pair feature tensor into a key sequence and a value sequence respectively; generating a context feature vector through a preset cross attention mechanism based on the query sequence, the key sequence and the value sequence; and classifying according to the context feature vector to obtain a classification result. In the foregoing manner, the application improves the detection capability of target object forgery.
Need to check novelty before this filing date? Find Prior Art