基于静态情绪先验和时空多视图的面部视频情绪评估方法

By employing static emotion priors and spatiotemporal multi-view methods, the problems of lack of frame-level labels and pose changes in facial video emotion assessment are solved, achieving high-precision and robust facial video emotion assessment.

CN121963273BActive Publication Date: 2026-07-17SOUTH CHINA UNIV OF TECH

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SOUTH CHINA UNIV OF TECH
Filing Date
2026-01-16
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing facial video emotion assessment methods lack frame-level emotion labels, cannot fully learn the evolution of emotions in subtle temporal sequences, ignore the correlation between fine local facial regions and multi-scale temporal sequences, and cannot handle facial pose changes and occlusion, resulting in unstable feature extraction and degraded assessment performance.

Method used

We employ a method based on static emotion priors and spatiotemporal multi-views, extracting global and local features through feature enhancement, knowledge transfer, and spatial attention. We then combine multi-head attention for interaction, perform progressive temporal modeling and joint loss optimization, and achieve end-to-end training.

Benefits of technology

It improves the accuracy and robustness of facial video emotion assessment, reduces the cost of technology implementation, and enhances the model's generalization ability and real-time application performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121963273B_ABST
    Figure CN121963273B_ABST
Patent Text Reader

Abstract

本发明公开了一种基于静态情绪先验和时空多视图的面部视频情绪评估方法。首先提取待评估视频和静态情绪图集的通用面部特征并进行增强,得到情绪特征与状态特征。通过最大均值差异损失实现从静态图集到视频的情绪知识迁移,获得视频全局情绪特征。接着,将长程状态特征划分局部区域,利用空间注意力提取局部状态特征,再通过多头注意力机制进行全局‑局部和局部‑局部交互,生成帧级状态特征。最后基于帧级状态特征得到全局变量,实现对面部视频的情绪评估。本发明能鲁棒、高效地提取多尺度时空特征,提升情绪评估精度。
Need to check novelty before this filing date? Find Prior Art