Research On Pore-based High Resolution Fingerprint Recognition

High resolution fingerprint images refer to the images with resolution higher than1000 dpi.Compared to traditional fingerprint images,high resolution fingerprint images contain more level 3 features such as pores and ridge contours.Among these features,pores are claimed to be stable,and more abundant than minutiae.Introducing pores into fingerprint recognition can improve the accuracy and safety of fingerprint recognition systems.By this reason,pore-based high resolution fingerprint image recognition has attached more and more attention,and has been a hot topic in the field of biometric recognition.Pore-based high resolution fingerprint recognition can be regarded as point matching and indexing problem,which is confronted with many difficulties in practical application,such as noise,outliers and deformations.These factors will reduce the accuracy of pore matching and fingerprint recognition.Moreover,a large number of pores in fingerprints increases the computation of recognition algorithm.These problems limit the engineering application and promotion of high resolution fingerprint identification system.In this dissertation,several novel algorithms are proposed and used for high resolution fingerprint image alignment,pore matching,and fingerprint indexing.These algorithms can improve the accuracy and speed of fingerprint recognition.The main content of this dissertation are as follows.(1)For partial fingerprint recognition,the dissertation proposes a data-driven descending algorithm for image alignment based on ridge contours.In the proposed algorithm,ridge contours are used to calculate the alignment error,and a data-driven descending algorithm is used to estimate the translation and rotation between test and template images.The proposed algorithm can determine the common areas between two fingerprint images accurately.It can avoid alignment failure caused by the insufficient number of feature points in the common areas.In high resolution fingerprint image matching,pores out of common areas are outliers.Removing them in the first step can improve the speed and accuracy of the subsequent pore matching process.(2)For similarity calculation of pores,the dissertation proposes a co-occurrence descriptor of pores based on edges.Similarity calculation of pores is an important step in pore matching.Existing pore matching algorithms use local descriptors to calculate the similarities,and establish the correspondences between pores.Due to the large num-ber of pores in each fingerprint,local descriptors cannot guarantee an accurate similarity calculation.To solve this problem,we propose a co-occurrence similarity calculation algorithm.This algorithm calculates the similarities between pores using the local descriptors and topological structures of pores.The similarities are then used for one-to-one correspondence establishment.The proposed algorithm can avoid outliers with similar local descriptors effectively,and improve the accuracy of correspondence establishment.(3)This dissertation proposes a local topological preserving matching algorithm for pore matching.This algorithm calculates the matching score of each correspondence,and remove the outliers based on the scores.The similarity between two fingerprint images is finally calculated based on the result of pore matching.The local topological preserving algorithm uses a rotational invariant topological structures to evaluate the similarities of pore correspondences.It can avoid the influence of deformation on matching results effectively,thus,improve the accuracy of the matching.The proposed algorithm outperforms the existing pore matching algorithms in terms of accuracy.(4)This dissertation proposes a high resolution fingerprint image retrieval algorithm based on pores.This algorithm uses local descriptors for pore similarity calculation and indexing.A batch graph matching algorithm is then proposed to remove the false matches.This algorithm matches pores in test image to all candidate pore in the registrations simultaneously with a fast speed and high accuracy.The similarities between test image and registrations are finally calculated based on the result of pore indexing.Experimental results demonstrate that the proposed algorithm can improve the accuracy and speed of the high resolution fingerprint indexing,and also verify the advantage of pore-based fingerprint retrieval algorithm over the traditional minutiae-based algorithms.