面向多模态特征融合的人脸活体检测方法及系统

CN122116492BActive Publication Date: 2026-07-17TIBET UNIVERSITY FOR NATIONALITIES

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
TIBET UNIVERSITY FOR NATIONALITIES
Filing Date
2026-04-24
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing single-modal face liveness detection technologies are not robust enough in the face of complex and ever-changing deception methods, and are difficult to effectively distinguish between real liveness and fake attacks. Furthermore, existing multimodal fusion methods fail to make full use of the complementarity and semantic affinity between modalities.

Method used

A multimodal feature fusion method is adopted, which simultaneously acquires RGB, infrared and depth images, uses Gabor filter kernels for texture enhancement and pixel-level registration, combines shallow convolutional networks and dual-branch processing to construct cross-modal affinity matrix and feature transfer mechanism, performs channel and spatial attention weighting, and finally uses a binary classifier for liveness detection.

Benefits of technology

It significantly improves the accuracy and generalization ability of multimodal face liveness detection, maintains stable feature representation under non-ideal imaging conditions, reduces computational burden and improves discrimination efficiency, and enhances the ability to distinguish complex attacks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122116492B_ABST
    Figure CN122116492B_ABST
Patent Text Reader

Abstract

本发明提供一种面向多模态特征融合的人脸活体检测方法及系统,涉及人脸活体检测技术领域。具体方法包括:同步获取RGB、红外和深度图像,对RGB图像进行Gabor纹理增强并完成与另外两种模态的像素级配准;将三种模态数据分别送入浅层卷积网络提取基础特征;对各模态基础特征实施双分支处理,通过轻量化卷积压缩提炼,并通过多尺度最大池化筛选凸显差异区域,经过融合得到降维特征;基于余弦相似度构建跨模态亲和矩阵,实现模态间的特征迁移与补强;对补强特征图进行通道与空间注意力加权后拼接融合,经分类器输出活体或非活体判别结果。本发明通过跨模态特征迁移与双分支特征降维机制,有效提升了复杂攻击场景下人脸活体检测的准确性与稳定性。
Need to check novelty before this filing date? Find Prior Art