Application of Deep Learning Models for Face Recognition

Tóm tắt

In recent years, face recognition technology has emerged as a reliable and efficient biometric authentication method for automating attendance management. This study presents the design and implementation of an integrated attendance and payroll management system based on facial recognition and liveness detection. The proposed system utilizes the Histogram of Oriented Gradients (HOG) algorithm for face detection and a deep convolutional neural network (CNN) to generate 128-dimensional facial embeddings through the face_recognition library. These feature vectors enable fast and accurate one-to-many comparisons using Euclidean distance. To ensure data security and prevent spoofing attacks, the system incorporates a three-layer verification mechanism, including token validation, real-time image analysis, and one-to-one facial revalidation. Experimental evaluation demonstrates that the model achieves an average accuracy of 88–90% under standard office lighting conditions with a processing time of 0.5–1 second per recognition, meeting real-time performance requirements. The integration of recognition, attendance tracking, and payroll generation within a single full-stack architecture significantly reduces manual intervention, enhances transparency, and improves operational efficiency in enterprise environments.