Fingerprint Ridge Features Extracting And Matching

Fingerprint is one of deeply researched and widely used biometrics to identify a person, because of its universality, distinctiveness, permanence and easy collectability. Many advances have been made in research and applied widely to area of forensic, financial, military applications and so on. However, recognition accuracy of fingerprint images of low quality, small area and badly deformed need to be improved in applications. Many algorithms have been proposed to solve these problems. These algorithms are mainly based on the minutiae extraction and matching. As to our observation, analysis and experiments, to describe fingerprint by ridge features accord more to human experts to match fingerprints than by minutiae. Ridges features have more information which could be used for matching, and can distinguish some fingerprints with little minutiae or a lot of false minutiae, which could not be recognized correctly by minutiae features. Therefore, fingerprint recognition accuracy could be improved for matching with ridge features. In this thesis the fingerprint ridge features are defined, and the algorithms to represent and extract ridge features and to match fingerprints by ridge features are researched, then a fingerprint recognition application based on ridge features are designed. The main contributions are summarized as following:1. An algorithm based on Log-Gabor filter to extract fingerprint ridge area is explored. Log-Gabor filter is robust to images with the different skin conditions in dry or wet, especially for the local area of low contrast. The Log-Gabor filter could turn the ridge area bright and make the non-ridge area dark. Fingerprint segmentation method by dynamic gray threshold could remove the non-ridge area of low gray value. According to the orientation reliability, the block in clear ridge area is more reliable with small difference between the blocks around. The disordered ridge with noises area could be segmented by low orientation reliability. Segmentation experiments on FVC2004 DB2 and FVC2006 DB1 show that fusion segmentation by Log-Gabor filter and orientation reliability is robust to segment fingerprint images, which could deal with the fingerprint image in different skin conditions and segment foreground of clear ridge area from disordered non-ridge area. The proposed algorithm segments effectively the ridge area for featres extacting and matching and reduces the bad effect of false features.2. Representation of fingerprint ridge features for matching is proposed. The fingerprint ridge features include point features, line features and topology features. The point features are the coordinates, orientation and point types of the two end points of a ridge. The minutiae features(terminations and bifurcations) from a fingerprint are the subset of its point features. The line features consist of the two end points, the max curvature, length and ridge types. The topology features only describe the relationship o f the neighboring ridges. Theoretically the fingerprint which could be verified by minutiae can also be verified by ridge features, for the ridge features contain more information than minutiae. And the fingerprint ridge structure and not be constructed from the minutiae features, but could be constructed properly from the ridge features. So some fingerprints, with little minutiae or a lot of false minutiae, cannot be recognized by minutiae, but could be recognized correctly by the ridge features. What is more, the ridge features are stored by the topology features, so it is easy to find the corresponding ridges for matching, and the matching time reduces, while all minutiae should be explored to find the corresponding minutiae. So the ridge features are more effective representation of fingerprint features than minutiae.3. The extraction algorithms of ridge features from thinned image are studied. After orientation estimation, fingerprint segmentation, enhancement and thinning, a fingerprint image turns to a thinned image with the ridges only 1 pixel width. We have reformed the minutiae extracting algorithm from thinned image to extract ridge features besides minutiae. The algorithm is based on the minutiae extraction, with a little more computation and storage to extract these ridge features of the two end points, the max curvature, length, ridge types and topology than minutiae. Experiments show that ridge features are more effective than minutiae in fingerprint matching and fast matching.4. The new matching algorithm of ridge features is proposed, including two stages of alignment of special ridges and matching with ridge features. Eight types of special ridges to align two fingerprints are defined and extracted. The special ridges are aligned with high priority first, to get the alignment parameters. After alignment, the line features and the topology features are used for searching the minutiae pairs. The minutiae similarity is computed with threshold for the minutiae pairs. And the matching score of the template and query fingerprint. Experiments on FVC2006 DB3 and DB4 show that ridge feature extracting and matching algorithm improves the accuracy of fingerprint recognition.5. A fast matching algorithm is proposed based on the ridge features. After the special ridges are extracted, The Ridge Reliability is defined according the topology features, line features and point features. According to the statistics and analysis of high reliability ridges in FVC fingerprint databases, only the ridges with high re liability are used for matching. Experiments on FVC2006 DB3 and DB4 show that the fast matching algorithm could greatly decrease the matching time but only decreases the accuracy a little, which is tolerable in application. And for FVC2006 DB3, when the fingerprint images are badly deformed, our algorithm can improve the fast matching accuracy better than global alignment.6. A new method to extract fingerprint features from subareas divided by image quality is proposed, to optimize the procedure of fingerprint recognition and improve the practicality of application. It is found that most images are not so bad and only parts of these images are of low quality in many fingerprint images in applications. We propose a new method to extract fingerprint features from subareas divided by image quality. The image quality of all subareas is estimated first. Ridges are traced and minutiae are extracted directly from gray level image for high quality subareas. Orientation estimation, enhancement, binarization and thinning are executed for poor quality subareas, and then the orientation of the poor quality subareas is adjusted according to the ridge orientation of the high quality subareas. Finally ridges are traced and minutiae are extracted for low quality subareas. Statistic results of feature extracting time of experiments on 4 FVC databases show that the proposed algorithm to distinguish between low quality subareas and high quality subareas costs less computation to extract features in high quality image subareas and reduces the computation time of fingerprint recognition.Above algorithms are implemented, and the experimental results and analysis are shown in the thesis,based on the FVC fingerprint databases.