Design Face Spoof Detection using Deep Learning

Abstract

The recognition of facial patterns has grown in popularity among the various biometric systems in recent years. As a result, there has been a significant advancement in this field. The security of such systems, however, may be of paramount concern given that numerous studies have shown that face recognition systems are susceptible to a number of threats, including spoofing attacks. When a biometric system is used to identify unauthorized users, spoofing refers to the capacity to trick the system into thinking the unauthorized user is the real deal by using a variety of methods to show a fake version of the original biometric attribute to sensing objects. A number of anti-spoofing techniques that do exist protects face spoofing. This study presents a simple and efficient Resnet18 Convolutional Neural Network (CNN) model architecture, which is more beneficial and produces greater accuracy than other handcrafted, machine learning and pre trained Models. This proposed system has three steps such as analyzing possible enhancements to features, loading photographs and their predictions, and face verification of real or spoofs in an image. We have proposed the model using publicly available challenging CASIA dataset. Wrap photo, cut photo, and electronic screen are three numerous varieties of images that capture real people’s faces. The input video split into picture frames in this step. Experimental results demonstrated that proposed approached outperforms similar approaches by showed a 99.12 percentage of real detection accuracy and a 98.20 percentage of spoof detection accuracy. The results of the experiments demonstrate that our proposed detection system indicates a higher results of detection rate rather than earlier used techniques.