Enhanced facial emotion detection based on deep feature whale optimization algorithm with Hyper Capsule Generative Adversarial Network

Abstract

to detect and interpret human emotions from facial features. This method enhances emotion recognition in real-time applications like human-computer interaction and surveillance. Traditional facial expression systems rely heavily on handcrafted features, which limits flexibility. They are often sensitive to environmental noise and fail in dynamic or spontaneous expressions.

Objectives: The proposed system presents a hybrid approach for emotion-based facial expression recognition, integrating advanced techniques at multiple stages of the processing pipeline.

Methods: Initially, Discrete Wavelet Transform (DWT) is employed for pre-processing, enhancing feature extraction by decomposing facial images into multiple frequency components. This is followed by Facial Landmarks-Based Segmentation, which isolates critical facial regions that are most indicative of emotional expression. For feature selection, the Whale Optimization Algorithm (WOA) is utilized to identify the most relevant features, thereby reducing dimensionality and enhancing the model’s efficiency. Classification is performed using a novel Hyper Capsule Generative Adversarial Network (HCGAN-G), which combines the representational strength of capsule networks with the generative capabilities of GANS to improve recognition performance, especially in complex and subtle emotional states.

Results: The effectiveness of the system is rigorously evaluated using a comprehensive set of performance metrics, including accuracy 95.4%, sensitivity 94.2%, specificity 96.3%, precision 93.5%, root mean square error 0.25, area under the curve 0.97, and F1-score 0.89, demonstrating its robustness and reliability in emotion recognition tasks.

Conclusions: The proposed hybrid emotion-based facial expression recognition system significantly improves on traditional methods by combining rule-based and data-driven approaches. This architecture integrates the hierarchical spatial feature learning capabilities of capsule networks with the generative abilities of GANs. As a result, the system can accurately classify even subtle and complex emotional expressions.