FACIAL MICRO-EXPRESSION EMOTION RECOGNITION USING DEEP LEARNING

Abstract

Facial micro-expressions are rapid, involuntary facial movements that reveal genuine emotions. This study proposes an efficient deep learning pipeline to detect and classify micro-expressions (happiness, sadness, anger, surprise, fear, and disgust) from static facial images. We introduce a lightweight, attention-enhanced Convolutional Neural Network (CNN), designed and trained from scratch for this fine-grained recognition task. The proposed model integrates a Convolutional Block Attention Module (CBAM) into a custom CNN backbone to improve its focus on discriminative facial regions. After determining that a transfer learning approach was ineffective, our model was successfully trained on a large composite dataset built from publicly available sources (CK+ and the Facial Expression Recognition dataset). Performance was validated on a stratified hold-out set by comparing the proposed attention-enhanced model against a baseline custom CNN. This work demonstrates a practical framework for achieving a superior balance of accuracy and efficiency for static-image Micro-Expression Recognition (MER).