Learning Hierarchical Visual Codebook for Iris Liveness Detection

Iris liveness detection is an important module in an iris recognition system to reduce the risks of being spoofed by fake iris patterns at the sensor input. A general framework is proposed to detect multiple types of fake iris images based on texture analysis. A novel iris pattern representation method namely hierarchical visual codebook (HVC) is proposed to encode the distinctive and robust texture primitives of genuine and fake iris images. HVC takes advantages of both locality-constrained linear coding and vocabulary tree. Therefore, it can achieve less visual code quantization error, capture salient texture pattern sparsely, and reduce the dependence on coding at the upper level of vocabulary tree. To establish a benchmark for research of iris liveness detection, we develop a large fake iris image database including various fake iris images. Extensive experimental results demonstrate that the proposed method achieves 99% accuracy in fake iris detection.