Hand-Based Feature Level Fusion for Single Sample Biometrics Recognition

Single sample biometrics recognition may lead to bad recognition result in real-world applications. To solve this problem, we present a novel feature level biometrics fusion approach by combining two kinds of biometrics: palmprint and middle finger image, both of which can be acquired from one hand image. We first utilize a manifold learning method to find the local embedding subspaces of palmprint and middle finger images, and then use principal component analysis (PCA) to extract the concatenated feature. To do so, a well performance could be obtained for the reason that the local structures of single model biometrics are preserved, while the redundancies between them are reduced. Comparing with single modal biometrics and score level fusion, the experimental results illustrated the average recognition rate of the proposed approach was significantly promoted to 98.71%. The performance comparisons in terms of cumulative match characteristic (CMC) curves for different recognition approaches were also presented to demonstrate the strength of the proposed fusion scheme.