Efficient Fingerprint Spoof Detection Using Texture-Guided Ensemble Feature Learning and Random Forest Classification

Fingerprint-based biometric systems often face risks due to spoof attacks. These attacks involve using fake fingerprint samples to bypass authentication. This research aims to improve fingerprint spoof detection using an efficient and accurate method. The main problem addressed is the failure of existing systems to detect spoofs with high reliability across different sensors and materials. The proposed method uses a Random Forest classifier supported by a new algorithm called Texture-Guided Ensemble Feature Learning. This algorithm selects texture-focused regions from fingerprint images. It assigns weight to regions with clear texture differences. These regions show variations between genuine and fake prints. The feature set includes pore spacing, residual noise, gray-level energy, entropy, and statistical values. Preprocessing steps enhance fingerprint image quality by cropping the region of interest. Edge detection and contrast adjustment help in identifying useful texture zones. The Texture-Guided Ensemble Feature Learning algorithm then selects key features from these zones. This selective learning improves the quality of inputs to the classifier. The Random Forest model then learns the pattern of both real and fake prints using these features. The method is tested using the LivDet 2009, 2011, and 2013 datasets. These datasets contain images from different sensors and spoof materials. The proposed system reaches an accuracy of 97.4 percent. It also shows fast prediction speed in real-time testing. The method works well across different spoof types. This improves fingerprint system security by detecting spoof attacks more effectively than older methods.