Research On Fingerprint Recognition Based On Direction Field And Deep Learning

In recent years,with the rapid development of information technology,image processing algorithms have been widely applied in biometrics.Compared with other biometric identification methods,fingerprint identification has a higher universality,but also more accurate and convenient,has been successfully applied to security applications in various industries.Planar fingerprint comparison is usually used in the field of security protection,which means that the fingerprint image contains more complete information and controllable quality.At present,there are many mature fingerprint recognition algorithms,but almost all methods use the texture information of the fingerprint image to extract the feature points of the fingerprint,and then perform the feature point comparison.In applications,some scenarios require one-to-one comparison of fingerprints,such as the financial domain;while some scenarios require one-to-many comparisons,such as fingerprint access control,attendance,and so on.If only the feature points are used for comparison,when the fingerprint library capacity is relatively large,the one-to-many comparison will make the comparison process take too long,which is not conducive to the further promotion of fingerprint recognition.In the process of processing and extracting feature points of classical fingerprint images,it is generally necessary to calculate the direction field of the fingerprint image,and then directionally filter the fingerprint image based on the direction field,so as to obtain the detailed features of the fingerprint more accurately;or use the directional field to calculate the macro features of the fingerprint,such as the center,the triangle point,and the pattern determination.This direction field data is discarded after use.In fact,the direction field of the fingerprint characterizes the distribution of fingerprint lines,which is a macroscopic feature between the macroscopic features such as fingerprint center and the detailed features such as feature points.Therefore,in a one-to-many fingerprint comparison,the direction field can be used for fingerorint screening.As long as the algorithm is designed properly and the alignment time of the direction field is smaller than that of the fingerprint detailed feature alignment,most fingerprints can be quickly excluded,thus improving the speed of one-to-many comparison of fingerprint.On the other hand,the directional field is a macroscopic feature that is independent of the fingerprint detail features.In any case,the comparison of the directional fields during the comparison process can greatly improve the alignment accuracy.Therefore,designing a fast fingerprint direction field alignment algorithm is very meaningful for the speed and accuracy of fingerprint recognition.As deep learning gradually enters the public's sight,it has also made great strides in many fields such as computer vision,natural language processing and signal processing,and its powerful learning ability has been recognized by more and more people.In view of this,we can consider the deep neural network to simulate the cognitive process of the human brain gradually abstracting,select the directional field information including the overall shape of the fingerprint and the local flow as the input of the network,and obtain the feature through autonomous learning,thereby achieving an alignment of the fingerprint direction field.This paper adopts the deep network structure,uses the direction field information of the fingerprint as the input of the network,and finally realizes the fingerprint comparison by giving the two fingerprint similarity scores by supervising the directional field characteristics of the self-learning fingerprint.First,the direction field information of the fingerprint is calculated as the original data set of this paper.Secondly,the fingerprint data is preprocessed,and the fingerprint data without the center point and the center point at the edge of the image is removed,filtered and cropped.The fingerprint data suitable as the training set is used to reduce the interference to the final recognition due to the unreasonable label.Then,the network framework is constructed based on the full connection layer,and the direction field information of the fingerprint is input into the network in the form of an array,wherein the same type of fingerprint given label is 1,and the different types of fingerprint given labels are 0.In the network,the input information and output information of each layer are spliced to form the input of the next layer,so that the layer and the layer form a dense connection.Finally,the network is trained so that the final model has the ability to identify the same type of fingerprint and different types of fingerprints.Experiments show that the proposed method has a higher recognition rate.