Research On The Technologies Of Feature Extraction And Recognition For Low Quality Fingerprint Image

Fingerprint recognition is one of the technologies of biometric authentication. It plays a more and more important role in many regions. Although, fingerprint recognition has been extensively studied and many advances have been made on it, there are still many problems expected to be solved which are shown in actual applications and evaluations (e.g. FVC). As a consequence, in recent years, many academies and industries have been making an in-depth research on fingerprint recognition technologies. This thesis has studied the main technologies of fingerprint recognition, mainly including feature extraction of low quality image and fingerprint matching, and makes the following contributions:(1) Ridge orientation estimation and segmentation of low quality fingerprint image are explored, and a neural network based method to train the correctness of ridge orientation is proposed. The trained network can also be used to segment fingerprint image. Due to the limit of ridge orientation estimation method based on gray gradient and low-pass filtering, we train the correctness of the estimated ridge orientation and those image blocks with incorrect estimated orientation can be segmented. At the same time, the incorrect estimated orientation can be revised according to the around correct orientations. Also, the trained network responds to a ridge block with different values for different orientations: the responding value for the real ridge orientation is larger than other orientations, and thus the trained network can be used to estimate ridge orientation. Experimental results show that these methods can improve minutiae extraction.(2) The types of fingerprint image regions are analyzed, and a secondary segmentation method to segment remaining ridges, which is the afterimage of the previously scanned finger, is proposed. Many existing segmentation methods can effectively separate non-ridge regions and unrecoverable ridge regions from the regions of clear or recoverable ridge structure, called primary segmentation, but are not able to segment the afterimage ridges. Secondary segmentation follows the steps of primary segmentation to cut the afterimage regions from the foreground of primary segmentation, and thus is able to avoid extraction of false feature.(3) A fingerprint enhancement algorithm based on Gabor filter is improved. Texture filtering is a popular method of fingerprint enhancement. In order to speed up the run of program, filters have to be implemented by using lookup tables, which leads to the discretization of parameters. The discretization and rapid change of ridge orientations around the singularpoints will sometimes produce blocky effect. Besides, fingerprint image will take on different enhanced results with respect to different shapes and sizes of the filters. We investigate into these problems and revise the Gabor based fingerprint enhancement. Experimental results show that the revised method leads to an improved performance.(4) A fingerprint matching method based on multiple reference minutiae is proposed. Most existing fingerprint identification systems match two fingerprints using minutiae-based method. Typically, they choose a reference minutia from the template fingerprint and the query fingerprint, respectively. When matching the two sets of minutiae, the template and the query, firstly the reference minutiae pair is aligned coordinately and directionally, and secondly the matching score of the rest minutiae is evaluated. This method guarantees satisfactory alignments of regions adjacent to the reference minutiae. However, the alignments of regions far away from the reference minutiae are usually not so satisfactory. We propose a minutia matching method based on global alignment of multiple pairs of reference minutiae. These reference minutiae are commonly distributed in various fingerprint regions. When matching, these pairs of reference minutiae are to be globally aligned, and those region pairs far away from the original reference minutiae will be aligned more satisfactorily. Experiment shows that this method leads to improvement in system identification performance.(5) An approach to matching fingerprints using minutiae and orientation field is proposed. It first aligns two fingerprints based on their minutiae, and then computes the consistency of their orientation fields. Experiments show that this approach leads to substantial improvement in performance compared with current minutiae based approach.(6) A minutiae relationship representation and matching method based on curve coordinate system is proposed. For each minutia, a curve coordinate system is established, and the coordinates of other minutiae in this coordinate system is computed. Thus, the coordinate relationship between each pair of minutiae can be evaluated. These relationships are used for pairing minutiae between the template fingerprint and the query fingerprint by means of transferring reference minutiae. The algorithm is tested on FVC2004DBs which include many highly distorted fingerprints. The results show that the proposed algorithm achieves improved matching accuracy and is able to cope with highly distorted fingerprints.(7) A quality estimation method of fingerprint image is proposed. This method is based on the correctness estimation of ridge orientation using the trained neural network. The quality estimation method aims at directing choosing template fingerprint at the enrollment stage of a recognition system. Besides, a merging method of template fingerprint features is proposedto serve for the enrollment. Experimental results show that these methods improve the accuracy at the recognition stage of the system.(8) A mean period based approach for fingerprint classifying is presented. It computes the mean period of every fingerprint and sorts fingerprints according to their mean periods. This approach speeds up the searching of fingerprint database efficiently. And perfect performance is possible if combining this method with traditional fingerprint classification. A demo system of fingerprint recognition is modularizedly and layeredly designed andimplemented. The demo packs the main algorithms proposed in this thesis.