Multi-Feature Facial Complexion Classification Algorithms Based on CNN.

Area of Science:

• Computer Vision
• Medical Imaging Analysis
• Machine Learning

Background:

• Facial complexion variations can indicate underlying health issues.
• Subtle facial feature distinctions make accurate complexion classification difficult.
• Convolutional Neural Networks (CNNs) show promise for image analysis tasks.

Purpose of the Study:

• To develop and evaluate novel multi-feature facial complexion classification algorithms.
• To improve the accuracy and efficiency of facial complexion analysis using deep learning.
• To identify optimal facial regions of interest (ROIs) and feature extraction strategies.

Main Methods:

• Proposed three distinct CNN-based algorithms: multi-feature fusion, splicing, and independent training.
• Extracted and utilized features from specific facial ROIs (nose, forehead, philtrum, cheeks).
• Trained and validated algorithms on a dataset of 721 preprocessed facial images.

Main Results:

• Multi-feature fusion and splicing algorithms achieved 95.98% and 93.76% accuracy, respectively.
• An optimal approach combining multi-feature CNN with machine learning reached 97.78% accuracy.
• The arrangement of ROI features (nose, forehead, philtrum, cheeks) proved optimal for classification.

Conclusions:

• Multi-feature deep learning algorithms, particularly fusion-based approaches, significantly outperform single-image analysis (e.g., EfficientNet at 89.37%).
• The strategic combination and arrangement of features from multiple facial regions are crucial for high-accuracy complexion classification.
• These findings offer new research avenues in facial complexion classification and deep learning applications for health monitoring.