Face Spoofing Detection using Hybrid Kernel Approach with CNN, SVM Classifiers

Abstract

Face antispoofing methods has been built up quite a while since the face affirmation frameworks were satisfactorily related. The standard methods in this subject simply utilize the whole area of person face. Regardless, uncommon facial parts continually have different structures and the full-face model possibly debilitate the error of the particular parts. Thusly, setting up the particular model for every facial part can improve the execution of against spoofing. Here, we propose another procedure of face against spoofing utilizing half and half CNN for facial parts. We separate the face into a couple of areas and dependent on different parts, applying the CNN representation for it, which will set up the crossover DCT-CNN. Moreover, we connect on the cross breed model to prepare a SVM classifier. Utilizing SVM, The face picture is portioned into number of various squares and LBP Features are taken, at that point SVM is utilized for deciding if the information picture relates to live or counterfeit face. We tried the proficiency of our procedure on open available databases, picture, video, mask attacks and the investigations exhibit our proposed methodology can secure acceptable results with respect to the top class methodologies.

 Keywords: Face-spoofing attack, Local Binary Pattern, CASIA dataset, SVM, CNN