Deep Neural Network Evaluation for Emotion Recognition through Facial Expressions Analysis

Emotion recognition through facial expressions is paramount in marketing, healthcare, affective computation, and security applications. This research evaluates Deep Neural Network architectures for emotion recognition based on facial expression images and discrete emotion classification. The datasets used are JAFFE and a custom dataset. The selected Deep Neural Networks are VGG16, InceptionV3, ResNet50, Transformer model, VGG16 with LSTM and VGG16 with a Transformer. The Networks use transfer learning with pre- trained ImageNet weights. Results show that VGG16 outperforms other models in accuracy for both datasets, while hybrid models like VGG16 with Transformer show lower accuracy. These results indicate that further fine-tuning and exploration of these architectures are necessary. Investigating different methods to combine the strengths of each architecture could lead to better outcomes. This research highlights the strengths and limitations of each architecture, guiding future advancements in emotion recognition technology.