Performance Evaluation And Analysis of LBP Utilization In Face Spoofing Detection With Deep Learning

With the development of technology recently, almost all universities in Indonesia have implemented online attendance. An online attendance system using facial recognition is a technology that is able to identify a person's face from a digital image. Based on the problems that are often faced today, during attendance the system is not able to distinguish real faces or prints. Based on these problems, we need a method that can help avoid face spoofing. In this study, a deep learning-based method was designed using a dataset of real face images and printed faces by utilizing the LBP (Local Binary Pattern) feature extraction technique to distinguish real faces or printed faces. From the research that has been done, the test results show that the performance of face spoofing classification with the LBP method can bring improvements to the three architectures. The ResNet34 model obtained an accuracy value of 98%, InceptionV3 obtained an accuracy value of 97%, and EfficientNet-B4 obtained an accuracy value of 99.4%. In the VGG16 architecture, the usage of original image yield superior results with an accuracy value of 99.5%. From the evaluation results above, it can be concluded that although in this study the LBP image did not significantly outperform the original image, this study found that face spoofing detection using feature extraction techniques can be considered and studied more deeply.