This study presents the development and evaluation of a secure and efficient real-time face recognition system for school attendance, integrating cancelable biometrics with cryptographic hashing. A total of 115 face samples were collected from students and teachers under diverse lighting, pose, and expression conditions. Images were pre-processed using Contrast Limited Adaptive Histogram Equalization (CLAHE) and Gamma Correction, followed by feature extraction with ResNet-128D, key-based random projection, binarization into 128-bit templates, and SHA-256 hashing. Evaluation results demonstrated an accuracy of 86.09%, precision of 100%, recall of 86.09%, and F1-score of 92.52%, with an average latency of 281.71 ms, remaining well below the operational threshold of 500 ms. Offline pre-processing improved the F1-Score by 7.50% on large datasets and 7.28% on smaller datasets without sacrificing processing speed. From a security perspective, the system achieved zero false acceptances (FAR = 0%) and allowed template regeneration when compromised, reinforcing privacy preservation. These findings validate the feasibility of combining cancelable biometrics with cryptographic hashing to balance accuracy, speed, and security in practical attendance systems. The research underscores its broader applicability to access control and public security, while future work should emphasize adaptive pre-processing, diverse hardware validation, and hardware acceleration for robust real-time deployment.