.png){kind=icon}
.png){kind=icon}
Return
A motion flow guided MicroNet framework for micro expression recognition
.png){kind=icon}
PRE
AI
DOI：10.1016/j.jvcir.2026.104765.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The micro-expression recognition (MER) has gained high attention in real-world applications: human–computer inter action, depression estimation, virtual reality, etc. However, MER systems still face difficulty in capturing the subtle information with spatial appearance of micro expressions. This paper proposes an efficient motion flow-guided MicroNet framework for MER, comprising a motion flow generator (MFGen) and an avalanche feature (AFeat) block. The MFGen extracts temporal changes in expressive regions by analyzing pixel motion intensity across frames, while the AFeat block captures spatiotemporal features through a multi-lateral complementary feature (MCFeat) block, which elicits coarse and deep edge responses from multi-scale receptive fields. The MicroNet estimates momentary variations and learns affective appearance features of micro-expressions. Its efficacy is evaluated through experiments on two setups using six datasets: CASME-I, CASME-II, CAS(ME) 2, SAMM, SMIC, and COMPOSITE, with validation schemes demonstrating generalization and robustness. Eight ablation experiments further validate the role of each module.
Keywords:
Micro-expression recognition
Motion flow
Spatiotemporal features
Avalanche feature block
Multi-lateral complementary feature
Journal
IF:
3.1
Papers:
414
Citations:
5.6K