The Study Of Automatic Fingerprint Recognition System

Fingerprint recognition system is an important biological identification method. Everyone has his own unique fingerprints which can identify himself/herself exclusively. Therefore the characters of personal fingerprints have the proper advantages in the biological identification field. More over the collection of fingerprints is very convenient and safe, so fingerprint recognition system has more reliability and price advantages compared with other biological identification technologies.Fingerprint recognition system is mainly comprised of the following processes, collecting of fingerprint data, preprocess of the data, extracting of minutiae, matching fingerprints, and so on. Because the outputs of the preceding process are just the inputs of the following one, the effects of every process will directly affect the following process'precision. This paper discussed the algorithms of these processes in detail and improved some key algorithms in different processes.There are seven chapters in the paper. Chapter one is an introduction, mainly introduced currently fingerprint identification technologies and some other biology identification methods. Chapter seven is a conclusion of the whole paper. The main contents of Chapters two to six are as follows:1. Study the extracting of fingerprint direction. The computation of the fingerprint image direction field is a key technique of the automatic fingerprint recognition system. From the experiments, we compute the direction of fingerprints by the template algorithm and the rapid algorithm from the grey picture directly of fingerprint. Then analyze and compare the effects of the extracting of the two methods. In this paper, we also introduced another efficiently algorithm. We could know from the fingerprint direction comparison graph that the precision of the template method acquisition is much lower, particularly in the line of regression direction variety the violent information of direction gain by region (click the nearby region such as the central point, triangle) is rougher; while the precision of the improved method is higher than that of the template method, which could be verified by the processes of segmenting images and dealing with the singular points.2. Locate precisely the core points and delta points of fingerprint. The singular points are examined according to the Poincare index. First locate the blocks with those singular points roughly in the image of fingerprint direction. Then search the singular points according to the Poincare index in the blocks. Therefore we could get a set of candidate singular points. At last we get the final singular points by k-means method.3. Realize four kinds of two-value algorithms and two kinds of detailed algorithms, namely that the global threshold algorithm, the max squared deviation among classes based algorithm, the adaptive two-value algorithm, the local threshold algorithm based on direction field, Zhang's detail algorithm, and OPTA detail algorithm, respectively. Discuss the different combinations of the methods and the different effects of the experiment. At last, bring forward the improved aspects. Different combinations of two-value algorithms and detail algorithm have different experimental effects. Despite of any two-value and detail algorithm, more false traits in which comprise lots of false bifurcation points or false end points will be produced after thinning for the bad quality of original fingerprint enhanced picture and have the relative centralized distributing. For the areas of good quality will have the desired effect after fining.4. Study the ridge line tracking technique based on the fingerprint detailed images. Then make the algorithm have more adaptability. The post processing algorithm of the detail feature extraction can identify and remove the false feature dot of short line, spike, bridge, annulet, and so on. For triangle, ladder etc. the recognition and cancellations of the complicated structure need our further research5. Carry out the fingerprint classification algorithms based on singular points fingerprint classification and based on the neural network, respectively. Experiment results show that the former algorithm highly depends on the precision of the singular points location. But it's very difficult for the exactly locating of singular points of most of low quality images. Moreover, many live-collected fingerprint images lack of delta field. Therefore, the former algorithm is limited in precision and application scope. The second algorithm doesn't depend on singular points location. Neural network at through a great deal of sample after train, carry on the sort according to the overall situation feature of the fingerprint image, The algorithm compare is also a haleness according to a little bit strange fingerprint classification. How the architecture of the optimization neural net, neural network and raise the generalization capability of the neural net, neural network should be the direction that we study further.6. Perform the point patter matching algorithm in polar coordinates system. Another point matching algorithm based on GA is also studied. In the first algorithm, the images are matched according to similarity of the ridge lines, and the variable circumscription box is movable, so the difficulty of match of the fingerprint image nonlinear deformation could be solved in some measure. In experiments, we found that most fingerprint images can carry out the fingerprint match with that algorithm, but some bad qualified and blurry images couldn't be matched precisely by this method. The main reason is the detail point features couldn't be abstracted in those bad images. The research work of our next move will in two aspect spreads. One is to improve the fingerprint image enhancement algorithm, withdraw the detail dot more accurately, and the further improvement detail matching algorithm, Strengthen the algorithm to the adaptability of the nonlinear deformation. The other is to study the fingerprint matching algorithms not based on detail point matching for the bad quality images. It could be acted as an assistance method of the detail point match method.This paper has studied the theories and technologies in fingerprint identification field and made a good foundation of theories and technologies for the deeper research of this field.