An Enhancement of Eigenface Algorithm Applied for Identifying Spoofing Attacks in Facial Recognition

Facial recognition is a biometric authentication technology that identifies individuals using their facial features in images and videos. The rise of spoofing attacks poses significant risks, especially for facial recognition, as malicious actors are impersonating individuals by misusing their identities. The accuracy of facial recognition can be impacted by factors like face occlusions, image low-resolution, and the distance of the user's face from the camera, leading to potential misidentification. This research enhanced the Eigenface Algorithm applied to identify spoofing attacks in facial recognition. The algorithm extracts facial features and transforms them into eigenvectors for improved accuracy. Previous studies showed limitations in face detection at varying distances, prompting the incorporation of a distance-based scale in the enhanced algorithm, targeting a range of 30 cm to 60 cm from the camera. In this study, the researchers implemented OpenCV2 and stored the trained dataset in a YAML file. The dataset was generated by capturing multiple images with varying environments and distances, which were then preprocessed to resize and convert them to grayscale. Results showed a confidence level of 93.83% for the enhanced algorithm, a significant improvement from 57.38% with the existing one, and a faster average recognition time of 0.0141 seconds compared to 0.0216 seconds, demonstrating a 36.45% increase in confidence and a speed improvement of 0.0075 seconds. The distance-based scaling improves the efficiency of the Eigenface Algorithm by preventing recognition attempts at unsuitable distances, thus enhancing usability in practical applications.