Fingerprint Classification And Its Application To Orientation Field Reconstruction

Automatic fingerprint identification technology has more advantages than other techniques at universality,uniqueness,collectability,acceptability and other aspects.Fingerprint classification and the extraction of orientation field is the key of the fingerprint identification technology,which is vital to improve the performance of subsequent fingerprint processing.With the development and progress of the society,the research of fingerprint identification technology is more mature.However,it is still some shortcomings to be improved.At present,the finger scars,peeling,the defects of equipment acquisition and improper pressing methods will lead to the relatively low quality of the collected fingerprint images.Most existing methods of computing the orientation field can correctly estimate the orientation field of high quality fingerprint images.Only the method based on dictionary construction is more effective for low quality fingerprint,while this method is poor in real time.We consider the important problem which the different types of fingerprints in the same locations will have different characteristics.Therefore,some improvements on the construct dictionary based orientation field rebuild algorithm in this paper to solve above problems..Experimental results certify that the algorithm can accurately reconstruct the orientation field of low quality fingerprint and reduce cost time.The major content and innovation are summed as follows:(1)A classification method based on fingerprint LBPV(local binary pattern variance)features is proposed.The fingerprint is divided into four parts by taking the fingerprint cores as the reference point and extracting the LBPV feature of each part.Then the improved KNN algorithm is used to get the probability of each type.Finally,the weighted fusion strategy is used to get the type of the fingerprint.Experimental results show that this feature has a good classification ability.(2)A fingerprint orientation field rebuild based on classification is proposed.Local dictionaries are constructed for each type of fingerprint.When a low quality fingerprint is input,this fingerprint will be classified after estimating the pose and then rebuild orientation field by querying the corresponding type local dictionary.By classifying the fingerprints,each location can contain rich texture information and need less number of dictionaries,which makes the localized dictionaries look-up more quickly and improves the real-time performance of the algorithm.The improved mediodsk-clustering algorithm is used to replace the original mediodsk-clustering algorithm in obtaining the prototype orientation patches,which greatly improves the clustering speed.