Automatic Classification of Fingerprint Images

Classification of fingerprints into disjoint categories can result in more accurate and faster matching. Fingerprint experts classify fingerprints into the following broad categories: whorls, leff loops, right loops, arches and composites. Traditionally, syntactic approaches have been employed for fingerprint classification. But the resuslts of classification using this approach have not been encouraging due to factors such as large variations within patterns of the same class and the sensitivityof syntactic methods to noise. In this paper, we pose i~ngerpnnt class~hcat~on as a stat~stcal pattern class~f~cahon problem We employ features from the d~rectional transform of the fingerprint image mstead of the frngerpnnt Image rtself. We construct the h~stoqramof e~qht d~rectrons andcompute the texture features from the co-occurence matrix of the direction image. We seiect the best feature subsets using the Whitney method and exhaustive search. The resulting recognition accuracy is promising, but additional experiments on a larger dataset are needed to establish the robustness of theproposed classification scheme.