Impact of Activation, Optimization, and Regularization Methods on the Facial Expression Model Using CNN.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Facial expressions are crucial for human-computer interaction and robot communication.
• Deep learning, particularly convolutional neural networks (CNNs), shows promise in image categorization tasks.
• Accurate facial expression recognition (FER) is essential for advancing human-computer collaboration.

Purpose of the Study:

• To develop and implement a novel FER technique using CNNs.
• To enhance human-computer interaction and robot communication through improved emotion recognition.
• To address the limitations of traditional FER methods.

Main Methods:

• Utilized a novel FER technique based on activations, optimizations, and regularization parameters.
• Employed convolutional neural networks (CNNs) for image-based emotion recognition.
• Trained and evaluated the model on facial expression databases like CK+ and JAFFE.

Main Results:

• The model successfully recognized a range of emotions including happiness, sadness, surprise, fear, anger, disgust, and neutrality.
• Performance was evaluated using activation, optimization, regularization, and hyperparameter tuning.
• Achieved superior network performance compared to state-of-the-art FER techniques.

Conclusions:

• The proposed CNN-based FER technique demonstrates high accuracy and effectiveness.
• This approach surpasses traditional FER methods relying on handcrafted features.
• The findings contribute to more sophisticated human-computer interaction and robotic communication systems.