Face Spoofing Detection Using Image Distortion Features

Accurate biometric system for authentication is the need of the hour in today’s scenario. In face spoofing attack a person tries to pretend to be a valid user by using photo or video of an authorized person and gets illegitimate access. Hence it is essential to develop a robust and authentic face spoof detection system in order to protect the privacy about the person. Centre of attraction of this paper is face spoofing detection system using Image Distortion Analysis (IDA). Analysis of distortion of an image to identify spoof attack is the principle consideration of the proposed system. This paper extracts four different IDA features-Specular Reflection, Blurriness, Chromatic Moment and Colour Diversity which are concatenated together to form IDA feature vector. IDA feature vector is further used for face spoof detection using Support Vector Machine (SVM) and Artificial Neural Network (ANN). This paper highlights the performance of accuracy of SVM and ANN using in term of True Acceptance Rate (TAR) and True Recognition Rate (TRR). The performance of classifier was tested on the data-set of public-domain face spoof databases MSU MFSD face images. It was observed that the TAR vs TRR for SVM is 94.4% while that for ANN is 88.9%. The performance of accuracy indicated that the spoof detection based on IDA using SVM is more secured than ANN.