Wavelet-based fingerprint image retrieval (Q1014552)

The authors present a method for fingerprint based identification based on feature vector comparison in which a basic component of the feature vector consists of Gabor coefficients, or other types of wavelet coefficients. After providing an expository introduction to metric access methods for image retrieval and a comprehensive overview of the specific system architecture proposed here, in addition to a discussion of normalization techniques when the images in question are fingerprints, the authors outline several specific possible feature descriptors, including quadtree wavelets, ``Gabor wavelets'', and steerable wavelets. The feature vector components and corresponding metric accessibility properties are described precisely in each case. Numerical comparisons are made based on image ensembles coming from the Bologna FVC 2002 and 2004 fingerprint databases. Based on several numerical experiments the authors conclude that the Gabor wavelets outperform statistically the approaches using other types of wavelets in terms of retrieval accuracy, based on the specific metric accessibility methods employed. No direct comparisons are made with other methods for fingerprint identification and comparison.