2.5D Palmprint Recognition using Signal level Fusion and Graph based Matching

Effective fusion schemes for combining complementary information from multiple sensors can improve the performance of applications like multi-biometric recognition. Multi-biometric recognition have become very popular as a counter measure for direct spoofing attack. Palmprint has evolved as a popular trait over the years for multi-biometric recognition due to its non-intrusiveness, low cost for capture device, and stable feature set. However, 2D and 3D palmprint in isolation can become vulnerable to spoofing attacks. In this paper, we propose a signal level fusion scheme for 2D and 3D palmprints to obtain an improved multimodal biometric system. In particular, we first generate a 2.5D palmprint data using standard deviation based locally adaptive signal level fusion of 2D and 3D palmprints. A graph based template matching framework is designed next for the purpose of recognition. Comprehensive comparisons with several existing works indicate the benefit of our proposed solution.