Characterization, Similarity Score and Uniqueness Associated with Perspiration Pattern

Vulnerabilities in biometric systems including spoofing has emerged as an important issue. The focus of this work is on characterization of ‘perspiration pattern’ in a time-series of fingerprint images for liveness detection. By using information in the high pass bands of the images the similarity score for the two images is calculated to determine the uniqueness of the perspiration pattern. In this wavelet-based approach, the perspiration pattern is characterized by its energy distribution in the decomposed wavelet sub bands. We develop a similarity matching technique that is based on quantifying marginal distribution of the wavelet coefficients. The similarity match technique is based on Kullback-Leibler distance, which is used to decide ‘uniqueness’ associated with the perspiration pattern. Experimental results show good separation resolution in similarity scores of inter (43 subjects) and intra (12 subjects over 5 months) class comparisons. This may be considered as a robust liveness test for biometric devices.