Finger knuckle print recognition based on multi-instance fusion of local feature sets

Biometrics has become one of the reliable averages to construct the recognition systems of personal identity. Recent studies have attracted the attention of researchers for a new method finger-knuckle-print (FKP), which focuses on the related skin patterns of the outer surface around the phalangeal joint of ones finger. It was discovered that the finger-knuckle print (FKP) allows discrimination between different people. Adaptation of feature extraction and matching to increase the distinction effectively between individuals plays a key role in such an FKP based personal authentication system. In this paper, we present a novel approach use of multi-instance feature fusion based on micro texture in spatial domain provided by uniform local binary pattern (ULBP) to circumvent the influence problem of the sub-image size on the recognition rate. For classification, we have used the minimum distance classifier and experimented with two different distance measures: Euclidean and City-block. The experiments clearly show the superiority of the multi-instance verification approach than using any single instance verification over individual classifiers on the published PolyU knuckle database.