view:9062 Last Update: 2021-12-12
Fingerprints Feature Extraction and Novel Fuzzy Neural Network Classification
Fingerprint recognition is an interesting subject in biometric. As theirs pattern complication, classifying and identifying them with uncertainty are critical problems in pattern recognition. Fingerprint feature extraction is most important process for identifying them as well. In this paper besides of using effective methods for extraction of features a five layer feed forward Fuzzy Neural Network composed by fuzzy neurons and its associated learning algorithm has been introduced for classification of shifted and distorted patterns. For generating input patterns to feed the matching network, the algorithm extracts singular points and minutiae of each pattern by using the optimum segmentation and recovering methods. The pattern of these features which are extracted of different shapes of fingerprints, let call it "feature map", will be encoded and applied to the intelligent classifier, The network first fuzzifies the entry then computes its similarities to all of the learned patterns, finally the network select the learned pattern of highest similarity and return its specific class illustrator of a person, as a non fuzzy output. We applied the database of almost 100 fingerprint acquired from ten people contain two types of minutiae and three types of singular points. The learning speed and matching capability are improved by using the feature coding method. Necessary time for feature extraction and classifying of 100 different fingerprints with 220*176 dimensions and 315 dpi resolutions is less than 20 second. Also necessary space for storing image bank becomes small, up to 2.54Kbyte per each feature map. This FNN is a classifying system with supervised learning algorithm.