Fingerprint Classification Based On Frequency Spectrum Energy And J-Divergence Entropy

Automation fingerprint recognition system is hotspot of biometrics recognition,and fingerprint classification is one of key technology in fingerprint recognition.In large fingerprint database,it is time consuming that the input fingerprint compared with the reference fingerprints in database.In order to reduce searching time and complexity of calculation,reference fingerprints in database are classified into several classes,and are put into the Corresponding database.The input fingerprint is only compared with the reference fingerprints in the corresponding database.In a sense,actually fingerprint classification is rough process of fingerprint recognition and it provides an index mechanism of fingerprint recognition system.Through native and foreign scholars have made a large number of researches,current fingerprint classification algorithm can't satisfy the application of low quality fingerprint images.So each step of fingerprint classification is researched in this paper.Fingerprint classification in this paper is based on fingerprint directional graph. Firstly,fingerprint directional graph is computed.Then fingerprint feature is extracted and dimension of fingerprint feature is reduced.At last,the new feature is sent into classifier.The research contents are as follows:一,Extraction of fingerprint directional graph is one important step of fingerprint classification,but for low quality fingerprint images,it is difficulty for current methods to extract accurate directional graph.So a new method based on frequency spectrum energy is proposed.Firstly,transform the fingerprint images from spatial domain to frequency domain,then the directional images are constructed according to frequency spectrum energy.Because of the distribution characteristics of frequency spectrum energy,this method not only extract accurate directional graph, but also has some tolerance to noise.Experimental results show the validity.二,High dimensional feature not only affects performance of fingerprint classification,but also affects speed of fingerprint classification.Because dimension of the fingerprint feature is too high,the method based on j-divergence entropy is proposed.J-divergence entropy is smaller and probability distribution with two classes is bigger.It is beneficial for fingerprint to classify.The new method is compared with usual methods of descending dimension,and experiment results have proved this method is feasible.三,SVM classifier is used in this paper.SVM is a kind of method with rigorous mathematical explanation and it is by optimal hyper plane to classify fingerprint.In this paper,fingerprint is classified into five classes and for each two classes,one SVM is constructed.Firstly,SVM classifiers are trained by training set,and then the input fingerprint is send into SVM classifiers.The experiments prove the efficiency of the approach.