Distorted Fingerprint Image Identification Combined With Orientation Field Feature

As our society has become electronically connected and more mobile, surrogate representations of identity such as passwords and ID cards can't be trusted to establish a person's identity any more. Biometric recognition is considered more reliable because biometrics identifiers intrinsically represent the individual's bodily identity. Because of the well-known distinctiveness and persistence properties as well as cost and maturity of products, fingerprint has been the most widely deployed biometric characteristics. More and more automatic fingerprint identification systems (AFIS) have been developed and deployed in government and civilian domains. However, the progress achieved in fingerprint verification is still quite far from people's expectation. Most fingerprint identification systems are dependent on quality of the captured fingerprint images, and the performance may drop dramatically on recognizing a poor-quality fingerprint image. Both the accuracy and efficiency of existing AFIS need to be improved.Minutia is the most widely adopted fingerprint feature in AFIS, as it is easier to detect and represent in computer, and more discriminative as well. However, in poor-quality fingerprint images, few reliable minutiae can be extracted. Also imperfect minutiae detection algorithms may deliver spurious minutiae or miss genuine minutiae. As a result, it is necessary to combine some other fingerprint features with minutiae in AFIS. Fingerprint orientation field describes the basic shape, structure and direction of a fingerprint. It can provide sufficient fingerprint details beyond minutiae, and it is extracted in most existing AFIS for its importance in fingerprint enhancement, fingerprint classification, etc.This thesis focuses on the research of combining orientation field with minutiae in automatic fingerprint image identification, especially in phases such as fingerprint image pre-alignment, fingerprint feature extraction, fingerprint local matching and fingerprint global matching. Toleration of nonlinear distortion in fingerprint images is also under consideration. Specifically, the thesis makes the following contribution:1. Most traditional fingerprint image alignment methods are based on features such as minutiae and singular points, and a best alignment is always pursued even at much higher computation costs. However, as for the nonlinear distortion, such a best alignment may do not exist at all, and too much accuracy is often unnecessary. We propose to align two fingerprint images coarsely using orientation field in the early stage, and then segment the overlapped regions in both impressions. To avoid spending too much time correlating two orientation images, we define fingerprint orientation difference field, and propose to change orientation field into orientation difference field first before orientation-field-based fingerprint image alignment. It also helps to reduce the effect of global distortion on orientation images. 2. Fingerprint experts always locate a minutia first by the macro-structure. However, the global topological structures of fingerprints are often ignored for the less discrimination. In order to introduce them into AFIS, we propose to divide a fingerprint image into several homogeneous regions by the fingerprint orientation field feature, and further to use such sub regions to help finding candidate minutiae pairs in local minutiae matching. Different strategies for the query fingerprint image and the template fingerprint image are adopted to improve the robustness of the method.3. Local minutiae matching allows to quickly and robustly determine pairs of minutiae that match locally, but minutiae descriptors encoding relationships between a minutia and its neighbor minutiae may suffer from problems such as spurious minutiae producing and genuine minutiae missing. Based on the analysis of a classic orientation-based minutia descriptor, we propose to improve the descriptor in order to cover more nearby regions around the centered minutia and to produce a local minutia matching similarity value more accurately.4. Nonlinear distortion is common in fingerprint images acquired by sensor. It degrades the performance of AFIS severely. We propose a method to adjust the minutiae position and direction on the basis of multiple reference minutiae alignment method. The query fingerprint image and the template fingerprint image are aligned first using multiple reference minutiae alignment method, and some landmark minutiae pairs from two images are computed, then minutiae around each paired landmark minutia are adjusted by specific rules. We also present a principle for distorted fingerprint image matching. That is,"aligning globally while matching locally, aligning coarsely while matching accurately". Based on the principle, we design a local minutia descriptor, and further use the descriptor to align and match two fingerprint images, while the similarity computing method is especially studied.