Advanced Facial Feature Extraction and Detection using Kernel based Deep Transfer Learning and Ensemble Model
M Rithani, R S SyamDev, Sri Vathsan M M, V K Sowrish · 2025
Facial feature extraction and detection systems, particularly for gender and age recognition, are crucial for security surveillance and targeted marketing. This study employs transfer learning models (Xception, DenseNet121, Inception V3, ResNet-50, and VGG16) and an ensemble approach on the UTKFace dataset. While individual models achieve accuracies of up to 94%, the ensemble model enhances this to 98%. Key contributions include using kernel PCA for dimensionality reduction and squeeze-and-excitation blocks for feature recalibration, advancing the precision and efficiency of facial analysis tasks.