A multichannel approach to ngerprint classi

Fingerprint classiication can provide an important indexing mechanism in a nger-print database. An accurate and consistent classiication can greatly reduce ngerprint matching time for large databases. In this paper, we present a ngerprint classiication algorithm which is able to achieve an accuracy better than previously reported in the literature. We classify ngerprints into ve categories: whorl, right loop, left loop, arch, and tented arch. The algorithm separates the number of ridges present in four directions (0 , 45 , 90 , and 135) by ltering the central part of a ngerprint with a bank of Gabor lters. This information is quantized to generate a FingerCode which is used for classiication. Our classiication is based on a two-stage classiier which uses a K-nearest neighbor classiier in the rst stage and a set of neural networks in the second stage. The classiier is tested on 4; 000 images in the NIST-4 database. For the ve-class problem, classiication accuracy of 90% is achieved. For the four-class problem (arch and tented arch combined into one class), we are able to achieve a classiication accuracy of 94:8%. By incorporating a reject option, the classiication accuracy can be increased to 96% for the ve-class classiication and to 97:8% for the four-class classiication when 30:8% of the images are rejected.