SVM Based Detection of Real and Spoofed Faces Based on Handcrafted Features

Infringement in biometric authentication is one of the crucial concerns in today's scenario. Many of the robust authentication systems fall prey to crude spoofing attacks and their performance had been drastically dropped due to several unseen attacks owing to technological advancements available at no cost. The objective of the proposed work is to design a robust spoof detection system that can sustain even the worst undetectable unseen interference that can fool a state-of-the-art authentication technique. The work comprises four stages and consists of removing artifacts, extracting the optimum features conventionally, reducing the dimensionality of the feature vector and the classification system. The 75:25 percent train: test analysis showed that the selected dimension of the handcrafted features using the Gaussian kernel of the Support Vector Machine was able to produce remarkable results overthe IDIAP dataset. The classification accuracy was 100% for 3488 test images and approximately 99.54% for 6000 test images.