Facial micro-expression recognition using stochastic graph convolutional network and dual transferred learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Micro-expression recognition (MER) is crucial for applications like lie detection but faces challenges with low accuracy compared to macro-expression recognition.
• Existing graph convolution networks (GCNs) for MER often use fixed structures, limiting their effectiveness.

Purpose of the Study:

• To propose a novel stochastic graph convolution network (SGCN) for micro-expression recognition (MER).
• To achieve state-of-the-art accuracy in MER by introducing randomness into graph structures and utilizing transfer learning.

Main Methods:

• A novel GCN architecture with a stochastic graph structure, where neighbors are randomly selected, was developed.
• The network employs a dual-branch approach: a spatial branch for images and a temporal branch for optical flow.
• Transfer learning was used, pre-training SGCNs on macro-expression datasets before applying them to MER tasks due to limited MER data.

Main Results:

• The proposed SGCN method achieved state-of-the-art performance on four well-known MER datasets.
• The stochastic graph structure demonstrated improved feature characterization and reduced computational complexity.
• Transfer learning effectively addressed the data scarcity issue in micro-expression datasets.

Conclusions:

• The novel SGCN with a stochastic graph structure and transfer learning significantly improves micro-expression recognition accuracy.
• This approach offers a promising direction for advancing MER technology in fields requiring subtle facial expression analysis.