ML Integrated Facial Expression Recognition on Occluded Faces Using Feature Fusion

Facial expression recognition is becoming a core part of human-computer interaction and emotion detecting systems. However, the real-time scenarios such as the widespread use of face masks presents a big challenge to traditional facial recognition systems. This work proposes a new approach by integrating machine learning concepts, particularly in neural networks and feature fusion. Our main goal is to develop an accurate facial recognition system that can recognize expressions even when the face is partially visible. To achieve this process, we will take a diverse dataset containing images of faces with variable occlusion. Using Data pre-processing techniques, it standardized the training and evaluation procedure. The obtained results conclude the model's ability to accurately recognize facial expressions even in the presence of occluded faces. To achieve this objective, this research work has proposed a Squeeze and Excitation Network (SE-Net) to enhance the relationship in convolutional feature maps. The proposed feature fusion technique combines the local features and global features. Moreover, it creates opportunities for enabling more consideration into scientific research and the synthesis of data from various inputs by eventually validating more complex situations.