Graph network learns how emotions relate to sharpen facial expression recognition

Facial emotion recognition systems still stumble on a deceptively simple problem: some emotions simply look alike.

To build that graph, the authors extract Local Binary Pattern features from the principal component of morphological profiles, and the resulting graph feature matrix is processed by a Graph Convolutional Network to learn emotion relationship representations.

Subject of Research: A graph convolutional network that models relationships between emotion categories to improve facial emotion recognition.

Article Title: ERGCN: emotion relationship graph convolutional network with cross-attention fusion for facial emotion recognition Article References: Imani, M., & Kabir, E. (2026).

ERGCN: emotion relationship graph convolutional network with cross-attention fusion for facial emotion recognition.