Fingerprint Identification Based On Directional Filter Banks

Fingerprint is one of the most reliable research objectives of the biology identification techniques for its uniqueness and inflexibility. Because of its good feasibility and practicality, fingerprint identifcation technique has become one of the most revalent, convenient and reliable identity authentication techniques. Fingerprint image preprocessing is the first step of automatic fingerprint identification system. whose quality will directly affect the result of automatic identification system.Using non-separable 2-D filter, the DFB(Directional Filter Bank) is able to extract the edge and texture efficiently, and acquire 2n directional information. It has widely been applied to the fields of image de-noising, segmentation, enhancement, identification, etc. for the high directional precision of its decomposition.This paper accomplishes the following achievements.(1) Analyzing the characteristics of DFB. Because the DFB can extract the edge and texture of image efficiently,and the fingerprint image has obvious local directionality,so it's feasibility when processing the fingerprint with DFB.(2) Determining the reference point of fingerprint and ROI centered with the reference point. The accuracy of the reference point location will effect the subsequent process of the fingerprint identification directly. At present,the method of fingerprint reference point location has good effect in good image only. It has bad effect in bad image even that it can't locate the reference point. Futher reseach is need for better reference point location algorithm.(3) Decomposing the fingerprint image into sub-images with different directional with DFB, which relative to the directional imformation of angle. And decompose the noises into sub-image with different orientation to achieve the de-noising.(4) Processing the directional sub-image based on orientation energy, extract the dominant directional energy of sub-image, calculate the distance between the input feature vector and template feature vector to complete the matching of fingerprint finally.(5) Preprocessing a fingerprint image with DFB conbined with Gabor filters. Experiment results show that the algorithm can reduce the noise, connect the broken fingerprint and enhance the ridges.