Shape Variance-Based Feature Extraction for Biometric Fingerprint Analysis

The analysis of biometric data has applications in a variety of fields including security, behavioral studies, surveillance and monitoring. A typical problem when analyzing such data is the identification of meaningful features. The challenge is to find relationships and correlations within the data that produce feature vectors which are robust to noise, acquisition errors, geometric transformations, or to other distortions. To this end, we propose a feature extraction technique that is based on the empirical distribution of shape variance as a fundamental criterion for analyzing and clustering the originating data. We apply the method to an image dataset consisting in fingerprints from multiple users, and show that a description of the fingerprint profile based on the variance of ridge flexion provides robust features with applicability in identification and pattern matching problems. We evaluate the extracted features using k-means clustering, by matching the generated clusters with the distribution of data labels. The features are obtained in an unsupervised manner and allow accurate matching of the data instances while exhibiting robustness to geometric transformations.