Author(s)

Mohan Yu

Corresponding Author

Mohan Yu

Abstract

Knowledge of one’s first language, which is greatly indicative of their cultural background, is a valuable asset in today’s diverse society. Different languages have unique phonetic combinations, causing native speakers to use facial muscles in distinct patterns. This paper proposes the spatial feature attention-enhanced facial action unit detection model SFAE-Net to detect facial movements by quantifying Facial Action Units (AUs). SFAE-Net consists of two modules: the face key-point-assisted region learning module (LRL) and the multi-scale region learning module (MSL). LRL uses key points to focus on AU regions, while MSL captures multi-scale features to improve generalizability. Experiments show SFAE-Net achieves an F1-score of 62.7% and accuracy of 79.5%, outperforming state-of-the-art models. The paper also provides an instance of analyzing first language effects on facial muscles using the model.

Keywords

Linguistics, Facial Features, Deep Learning, Key Point Detection, Attention Mechanisms