Face emotion recognition is an evolving field within computer vision and artificial intelligence, finding applications in areas such as human-computer interaction, psychological analysis, and security systems. This project aims to develop an effective face emotion recognition system using MATLAB, with a focus on feature extraction using AlexNet, a pre-trained convolutional neural network (CNN).
The primary goal of this project is to accurately identify and classify human emotions from facial expressions, specifically targeting emotions such as happiness, sadness, neutrality, anger, surprise, fear, and disgust. The methodology involves several stages: image acquisition, preprocessing, feature extraction using AlexNet, and emotion classification.
In the image acquisition phase, facial images are sourced from existing datasets. The preprocessing stage involves enhancing the quality of these images through techniques such as grayscale conversion, resizing the images to the required size of AlexNet, noise reduction, and histogram equalization, ensuring consistency and improving the performance of subsequent stages.
Feature extraction is the core of this project, utilizing AlexNet to extract high-level features from the facial images. AlexNet, a deep learning model known for its robust performance in image classification tasks, is employed to leverage its convolutional layers for automatic feature extraction. This involves resizing the facial images to match the input size of AlexNet (227x227 pixels) and using the network to extract deep features from these images.
MATLAB's Deep Learning Toolbox and Image Processing Toolbox provide the necessary tools and functions to implement and integrate these methodologies efficiently. When the testing code is executed, the webcam is activated to capture live facial images, and then the detected emotion is classified in real-time.
The results indicate that utilizing AlexNet for feature extraction significantly enhances the accuracy of emotion recognition compared to traditional feature extraction methods. This approach benefits from the rich, hierarchical features learned by AlexNet, which are highly discriminative for facial expressions.
In conclusion, this project successfully demonstrates the effectiveness of using AlexNet for feature extraction in face emotion recognition with MATLAB. It highlights the potential for combining deep learning with conventional machine learning classifiers to achieve high accuracy. Future work could focus on real-time processing capabilities and integrating more sophisticated deep learning models to further improve accuracy and generalization across diverse datasets.