ABSTRACT
Biometric authentication is very important nowadays. This authentication system has various attacking points that are vulnerable to different attacks. Although the biometric traits (face, iris, voice, etc.) seem personal for individuals, though unknowingly, we disclose those traits publicly in our daily activities. Facebook, Twitter, WhatsApp and Instagram are the most frequently used social networking sites. These are used to share similar personal information and make it easily available to the public. Therefore, a person can easily get biometric traits of others and may misuse this or can gain illegitimate access and advantages. To secure this, a spoofing detection algorithm is needed. The present work can detect whether the face presented before a camera is real or fake. Our proposed system consists of three phases: the phase is preprocessing, where the face region is marked and cropped using facial landmark points. The second phase is feature extraction where we compute the dense scale invariant feature transform (DSIFT) descriptors from the detected facial region and represent them using three different approaches. The last phase is for classification, where SVM is used for the classification task and it determines whether the image is an original face or a fake. We have tested the performance of the proposed work using the NUAA database, which is publicly available. The accuracy of the proposed system is superior to some existing methods.
