Finger Knuckle Pattern Recognition Through The Fusion of Major and Minor Knuckles

Fingerprints, palm veins, face recognition, DNA, palm print, hand geometry, iris recognition, retina, voice, gait, signature, and other physical or behavioral features have long been employed in biometric systems. The finger knuckle print is a new biometric feature that has piqued the interest of academics in recent years. Recently it was discovered that the skin's knuckle image pattern comprises of wrinkles or lines, and that the texture pattern created by the finger knuckle is very unique in each user, making the surface unique for biometric identification. The minor finger knuckle patterns can be utilized as standalone biometric patterns or in conjunction with the major finger knuckle patterns to increase performance. A vast number of research suggest that multibiometric fusion and multi-modality employed can greatly increase the biometric identification system's recognition rate, anti-attack, and resilience which might be incredibly useful in forensics applications and other related domains. In this study, a multimodal biometric system which combines minor and major finger knuckles is developed and experimented on PolyU-FKP finger knuckles datasets. Feature extraction techniques used include hand-crafted feature extraction descriptors, PCA and BSIF, CNN models, AlexNet and modified AlexNet. The results obtained showed that major finger knuckle system fared better in both PCA and BSIF, accounting for the clearer patterns on the major finger knuckle, based on early testing results comparing it to the minor finger knuckle system. Additionally, the outcomes demonstrate that when the two traits are mixed at different phases, the system is noticeably improved, particularly in the case of PCA, where up to 15.1% improvement was achieved. The best accuracy overall obtained is a 100% in AlexNet model.