Improving Biometric Security Through Multimodal Fusion and Deep Hashing

Abstract

Biometric security technologies are increasingly important for protecting sensitive information and securing access control. There are inherent problems related to spoofing, privacy and data security in traditional monomial biometric systems. In this paper, we proposed a novel deep learning framework to enhance the biometrics security by using multispectral face, iris and fingerprint information. Combining deep hashing into the proposed fusion framework, a strong binary multimodal latent representation is generated which is robust in presence of fake attempts. The proposed approach also integrates a hybrid security framework (combining cancellable biometrics and secure sketch method) for improving security of biometric templates. Furthermore, deep auto encoder algorithm is applied for feature extraction to get improved encoded features in order to boast security. The efficacy of the approach is demonstrated on a multimodal face, iris and fingerprint biometric database, resulting in improved performance along with enhanced privacy through cancelability and unlink ability of biometrics templates. Deep hashing function is also tested on an image retrieval dataset task as well standard one where the network structure could be applied’re used and it shows similar adaptability.