Robust Facial Feature Extraction Using Deep Learning Under Varying Image Conditions

Abstract

Facial recognition has become one of the most prominent applications of computer vision, driven by advances in deep learning techniques. The performance of facial recognition systems is strongly influenced by the quality of feature representations and the model's ability to generalize across different image conditions. In real-world applications, facial images are frequently affected by variations in illumination, facial pose, expression, image resolution, and occlusion, which may influence feature consistency and recognition accuracy. This study aims to investigate facial feature extraction and evaluate the generalization capability of deep learning models under varying image conditions. The proposed approach employs a Convolutional Neural Network (CNN) as a feature extractor, with pretrained architectures and embedding-based representation learning. The evaluation is designed around illumination variations, pose changes, resolution degradation, expression variations, and partial facial occlusion. Model performance is assessed using accuracy, precision, recall, F1-score, cosine similarity, and Receiver Operating Characteristic–Area Under the Curve (ROC-AUC), subject to an appropriate face-verification protocol. Embedding distribution analysis is also used to investigate the stability of facial representations across image conditions. The study is expected to clarify the robustness of facial feature representations, the sensitivity of deep learning models to image variations, and the factors affecting model generalization. The findings may contribute to the development of more reliable facial recognition systems for unconstrained environments..