Image Quality and Texture-Based Features for Reliable Textured Contact Lens Detection

Textured contact-lens detection in iris biometrics has been a significant problem. In this paper, we propose a novel approach based on image quality and texture-based features for presentation attack detection for patterned/textured contact lens detection. The proposed approach employs the image quality features computed using Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) and texture features computed from Binarized Statistical Image Features (BSIF) to detect presentation attacks based on contact lenses. An efficient comparator using Spectral Regression Kernel Discriminant Analysis (SRKDA) is used for computing sample scores. The Fischer Discriminant Ratio (FDR) weighted fusion is used to perform score-level fusion from both models. The proposed method is tested on LivDet-Iris 2017 Clarkson, Notre Dame, and IIITD dataset. The experiments show noticeable results in detecting textured contact lens in iris samples for both same-set evaluation and cross-set evaluation.