Global-local combined features to detect pain intensity from facial expression images with attention mechanism

• DOI: 10.1016/j.jnlest.2024.100260
• Publication Date: 2024-05-22T01:08:42Z
• Authors: Jiang Wu; Yi Shi; Shun Yan; Hong-Mei Yan
• Keywords: Feature (linguistics); Expression (computer science)

The estimation of pain intensity is critical for medical diagnosis and treatment patients. With the development image monitoring technology artificial intelligence, automatic assessment based on facial expression behavioral analysis shows a potential value in clinical applications. This paper reports framework convolutional neural network with global local attention mechanism (GLA-CNN) effective detection at four-level thresholds using images. GLA-CNN includes two modules, namely (GANet) (LANet). LANet responsible extracting representative patch features faces, while GANet extracts whole to compensate ignored correlative between patches. In end, correlational subtle are fused final intensity. Experiments under UNBC-McMaster Shoulder Pain database demonstrate that outperforms other state-of-the-art methods. Additionally, visualization conducted present feature map GLA-CNN, intuitively showing it can extract not only but also ones. Our study demonstrates non-invasive feasible method, be employed as an auxiliary tool practice.