Research On Fingerprint Recognition Based On Wavelets

With the rapid development of science & technology and economy, the demands of hu-man identification in a variety of situations are increasingly strong. In order to increase thespeed of identification and to enhance the robustness of identifying algorithms, how to designthe model of identification with high accuracy and speed in the application of embedded fin-gerprint identification systems, in which a lot of noise are attached, has become one of thehottest research topics.Nowadays, as the traditional methods used to detect the detailed features of ridges infingerprint are found very di?cult to meet the requirements, using wavelet transform has beenone of the preferred directions in feature extraction. In this thesis, a model based on theapplication of embedded fingerprint identification has been proposed, using wavelet transformto extract the features of the original fingerprint images, and then using pattern recognition withsuch features by probabilistic neural network. The input fingerprint images are preprocessedby image segmenting under the condition of no distortion as much as possible. By doingso, the amount of data in wavelet transform is reduced and hence the recognition is speededup. Then the fingerprint feature extraction is carried out with two approaches to the featureextraction. The first is to use one dimensional wavelet transform based on Haar wavelet andthe second one is to utilize two dimensional wavelet transform based on Daubechies (dbN)series wavelet.To handle properly the input fingerprint images obtained from the embedded fingerprintrecognition systems that contain many stains and impurities and to meet the increasing de-mands for real-time in the embedded systems, based on the maximum between-class varianceand the traditional variance segmentation a new method to segment the fingerprint images isproposed in this thesis, which enhances the model robustness against noises to some extentwithout slowing down preprocessing. At the same time, research is done on the smoothingfactor of the probabilistic neural network which is used to make feature comparison (i.e., theselection of match sensitivity) and the best dynamic range of the match sensitivity in our modelis proposed. The simulation results indicate that proposed method can increase the accuracy of recognition significantly as long as the value of smoothing factor is selected in this range.