Micro-expression recognition method and system based on context awareness and multi-modal routing

By employing context-aware and multimodal routing methods, the problems of context decoupling and multimodal fusion in micro-expression recognition technology are solved, achieving high-accuracy micro-expression recognition in complex scenarios and improving the model's recognition ability and robustness.

CN122435653APending Publication Date: 2026-07-21JILIN UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
JILIN UNIVERSITY
Filing Date
2026-04-22
Publication Date
2026-07-21

AI Technical Summary

Technical Problem

Existing micro-expression recognition technology is limited by "decontextualized single-modal visual analysis", which cannot effectively handle the semantic asymmetry of contextual sources. It faces the computational disaster and overfitting risk brought about by long sequence modal features, and blind multimodal fusion is prone to negative transfer effect, resulting in low recognition accuracy in complex real-world scenarios.

Method used

We employ a context-aware and multimodal routing approach, acquiring visual, audio, and text data to perform heterogeneous feature extraction and cross-modal fusion. By utilizing context type labels and contextual cross-attention mechanisms, we clarify the decoupled interaction triggers and achieve sentiment classification using fine-grained cross-modal self-attention mechanisms and adaptive gating fusion techniques.

Benefits of technology

It significantly improves the accuracy of micro-expression recognition in highly ambiguous natural interactive contexts, corrects visual paradigm defects, enhances the inference logic of micro-expression-induced intentions, reduces computational burden and overfitting risk, and reduces negative transfer effects.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122435653A_ABST
    Figure CN122435653A_ABST
Patent Text Reader

Abstract

The application relates to the field of artificial intelligence and emotion computing technology, and particularly discloses a micro-expression recognition method and system based on context perception and multi-modal routing, which comprises the following steps: respectively performing heterogeneous feature extraction on a visual micro-motion video segment, an audio segment and transcription text of a current sentence, and an audio segment and transcription text of context data; performing cross-modal fusion and sequence compression; generating context type embedding according to a context type label; extracting context perception sentence features through a context cross-attention mechanism; obtaining visual classification confidence based on a visual micro-motion feature tensor, and comparing the visual classification confidence with a preset threshold to generate an emotion classification result. Through a layer focusing fusion network, the application realizes high-dimensional heterogeneous feature extraction and emotion inducement analysis on facial micro-deformation caused by a multi-modal stimulus context, so that accurate prediction of concealed micro-expression is realized.
Need to check novelty before this filing date? Find Prior Art