Similarity Measures Implementation on Face Authentication using Indonesian Citizen ID Card

The aim of this study is to demonstrate to computer science students the application of basic algebra concepts, particularly distance, in face authentication. Our simulation presents the results obtained from face authentication using five face images from 39 respondents and an Indonesian citizen ID card (KTP). We utilized MTCNN and RetinaFace for face detection and FaceNet512 for feature extraction. The KTP and images were then compared using Cosine, Euclidean, and Euclidean-L2 distances. The outcomes indicate that RetinaFace achieved a 100% success rate in image detection during this simulation, whereas MTCNN only reached 99.15%. The Cosine Distance method obtained the best Specificity and Accuracy results for both face detection techniques. In terms of Recall score, the RetinaFace-Euclidean Distance method shows a better performance than the other available scenarios.