Quantitative comparison of two nonlinear approaches for synthetic biometric verification

Abstract

Biometric verification is one of the leading security issues. Being the unique signature (e.g., fingerprint) of human beings, biometric pattern recognition serves as an interesting branch of study in disciplines of digital image processing. Due to the fact that the accuracy of pattern recognition determines the correctness of decision making, two nonlinear approaches have been introduced, where independent component analysis (ICA) and discrete wavelet transform (DWT) are among the most effective methods. After sensing information is digitized into image matrix files, data matrices could be analyzed using ICA algorithms. 2D DWT is capable of compressing fingerprints and then reconstructing patterns by reorganizing approximation components and revised components of horizontal detail, vertical detail and diagonal detail via thresholding. Satisfactory results can be obtained from both approaches, however, it is hard to show which approach could be better performed from visual appealing. In this case, one set of quantitative measures is presented to evaluate outcomes of biometric verification on a basis of ICA and 2D DWT, where the objective measures of the discrete entropy, discrete energy, relative entropy, contrast, homogeneity, dissimilarity and mutual information are used to illustrate drawbacks and merits of these two dominating biometric recognition approaches.