Facial emotion recognition and age estimation are critical for human–computer interaction, yet existing methods struggle with variable image quality, diverse expressions, and aging patterns. To address these challenges, this study proposes a novel hybrid model combining optimized image enhancement with deep feature extraction and machine learning classification. The methodology integrates three key phases: (1) image enhancement using white balancing and adaptive gamma correction to improve edge intensity (82.235 vs. 62.89 in original images) and texture clarity (gradient: 8.248 vs. 6.183); (2) feature extraction via a modified ResNet-18 trained from scratch with monochromatic inputs; and (3) SVM classification of concatenated features from original and enhanced images. Evaluated on the UTKFace dataset (1,000 images) and savory/unsavory expression datasets (600 images), the model achieves 98.75–100% accuracy, outperforming AlexNet (94.17%) and ResNet-KNN alternatives. Key findings demonstrate: (a) 5.4% higher age estimation accuracy (96.41% vs. 87.41%) with enhanced gradient metrics, (b) 3.75% improvement in emotion recognition through edge preservation, and (c) interpretable Grad-CAM visualizations validating feature relevance. The proposed system reduces misclassification errors by 60% for subtle expressions (e.g., fear vs. surprise) compared to state-of-the-art models. Practical applications in healthcare show 40% faster patient check-in processing and robust performance under low-light conditions. This work establishes that integrating physics-based enhancement with deep residual networks significantly improves reliability for real-world deployment. Future research will optimize computational efficiency for edge devices and expand multi-modal biometric integration.
El‐Hag et al. (Tue,) studied this question.