Classification And Identification Of Fingerprint Images

Due to the uniqueness and invariability of fingerprints, the automated identification based on fingerprints is becoming an attractive alternative to the traditional methods of identification. However, there is much special noise in fingerprint images, and many methods wouldn't be publicized, so an ideal automated fingerprint identification system (APIS) is still a difficult research subject.At present, APIS may be grouped into two major categories - APIS based on minutia features and APIS based on statistical features. In this paper, the part of APIS based on minutia features includes mainly computing orientation, filtering, binarization, thinning, feature extraction, postprocessing and minutia matching; the part of APIS based on statistical features is composed of reference point location, extracting feature, classification and matching. Some new approaches are brought forward:1) In the preprocessing, a new binarization method of fingerprint images based on the orientation and the dynamic threshold is proposed, which has the excellent capability of noise resistance. This method makes fully use of the orientation and the characteristics of grayscale change, gets the binarized images from the primitive fingerprint images directly, instead of the series of processing such as smoothing, enhancement and binarization.2) In the aspect of classification, a new feature vector for classification is developed. This kind of vector can be gotten directly from the orientation images. Compared with other classification features, it's advantages are computational simplicity and low dimension, so it is more suitable for classification.3) In the classification networks, this paper not only introduces the traditional center clustering method and the back-propagation (BP) network, but also applies a hidden-layer-structure-adaptive radial basis function (RBF) network to fingerprint classification.Author practices all algorithms mentioned in this paper with Visual C++ on the PHI computer. The experiment results demonstrate that new methods can enhance processing effects of fingerprint images and improve the classification performance.