학습 시간 단축과 보안 강화를 위한 새로운 얼굴 인식 방식

초록

Facial recognition technology has become an essential tool in contactless identity verification, particularly following the COVID-19 pandemic. However, traditional image-based systems face two significant challenges: high computational costs due to extensive training datasets and privacy risks associated with storing facial images. This study introduces a novel coordinate-based learning approach that leverages landmark data and perspective transformation to address these issues. By using 68 landmark coordinates extracted from facial images, the proposed method eliminates the need to store raw images, thus safeguarding personal data and reducing data size for faster training. Perspective transformation further enhances recognition accuracy by generating frontalized data, improving performance for diverse angles and expressions. Experimental results demonstrate that the proposed system achieves a training time reduction of approximately 68% compared to conventional methods while maintaining over 98.5% accuracy. This study presents an effective solution for secure, efficient, and accurate facial recognition in moder`n biometric systems.

키워드