Robust Fingerprint Anti-Spoofing: Integrating Wavelet Scattering and Convolutional Neural Network

Fingerprint recognition systems are vulnerable to spoofing attacks, undermining their security. Traditional PAD methods, using handcrafted features, struggle to generalize across spoof materials and sensors, while deep learning models require large datasets and high computation. We propose a hybrid approach integrating a wavelet-scattering network (WSN) with a CNN, leveraging WSN for stable texture features and CNN for high-level representations. By processing local patches around minutiae, our method reduces computational costs. Evaluated on LivDet2015, it outperforms MobileNet in intra-sensor tests with known spoofs, reducing ACE by 15.25%, and shows robust generalization. in cross-material and crosssensor scenarios, offering a balanced, interpretable, and robust solution for fingerprint PAD.