Hybrid Deep Learning System for Robust and Scalable Face Authentication

Abstract

A biometric face identification system plays a crucial role in authentication system, requiring accuracy and computational efficiency, particularly in resource-constrained environments. The objective of this study is to develop a lightweight deep learning framework for facial authentication that achieves high performance while maintaining low computational cost. The proposed method integrates a hybrid architecture combining 1-Dimensional Convolutional Neural Networks (1D-CNNs) and Long Short-Term Memory (LSTM) networks. The preprocessing pipeline includes face detection, grayscale conversion, image enhancement, cropping, resizing to a standardized input size and normalization. Subsequently, Linear Discriminant Analysis (LDA) is applied for dimensionality reduction while preserving discriminative facial features. The CNN component extracts local spatial features from the facial image, while the LSTM component models the sequential dependencies among the extracted feature representations, enabling the network to capture contextual relationships and improve biometric face recognition accuracy. The system was evaluated on two benchmark datasets, MUCT and CASIA-WebFace, achieving classification accuracies of 100% and 99.8%, respectively. These results demonstrate the effectiveness of the proposed hybrid architecture in learning discriminative facial representations and its strong generalization capability across datasets. The novelty of this work lies in the integration of a lightweight CNN-LSTM framework with LDA to achieve a balance between high recognition accuracy and computational efficiency, making it suitable for real-time facial authentication in edge and IoT environments.