Facial Emotion Recognition Based on Resnet-18 Model

Abstract

This paper introduces an efficient computational method capable of real-time facial expression analysis, extracting features and classifying seven distinct emotional states from static images. To bridge the gap between traditional recognition and robust automated analysis, we propose an optimized deep learning framework based on the ResNet18 architecture. The model employs a hybrid transfer learning approach and enhances generalization through strategic data augmentation. Furthermore, the architecture leverages the skip connections inherent in ResNet to maintain feature integrity. Trained on the CK+ dataset, the model achieved a state-of-the-art accuracy of 98.4%.