Contact Lens Detection using Transfer Learning with Deep Representations

Abstract

Iris scans are most promising, secure and stable than any other biometrics such as fingerprints etc. However, the Iris Recognition Systems (IRS) are exposed to the possibility of being attacked by spoofing using contact lenses. Consequently, detecting the presence of contact lenses becomes obligatory for an IRS to work meticulously. The main contribution of this paper is to present a transfer learning technique relying on a pre-trained deep Convolutional Neural Network (CNN) to extract features followed by principal component analysis (PCA) based feature selection. At last, a cubic Support Vector Machine (cSVM) is used for training Error-Correcting Output Code (ECOC) multiclass model to detect the presence and type of a contact lens. The pre-learnt CNN architecture used here is trained on a huge size of dataset which can be transferred to contact lens detection with a smaller sized dataset. The proposed method is evaluated on two publicly available databases namely, IIIT-D and ND in order to ascertain the adaptability of our method. The comparative analysis affirms the performance superiority (i.e., correct classification rate (CCR)) of the proposed method over the state-of-the-art contact lens detection algorithms.