Research On Key Algorithms Of Automatic Fingerprint Identification

Biometric systems are rapidly gaining acceptance as one of the most effective technologies to identify people. A biometric system is essentially a pattern recognition system that acquires raw data from an individual, extracts a notable feature set from the raw data, compares this feature set against the feature sets stored in the database, and executes an action according to the result of the comparison. In biometric community, fingerprint recognition is the most popular method. Nowadays, there are a lot of fingerprint systems existing in a wide range of applications:from physical access control to criminal investigation and from inmates managing to corpse identification. However, there are still many problems expected to be solved which are shown in real-world applications and evaluations. In this thesis, we focus on the feature selection for sensor interoperability, fingerprint segmentation, estimation of average inter-ridge distance, fingerprint scaling, system performance provement algorithm, and complexity reduction for non-additive measure.The main contributions of the thesis can be summarized as follows.1. The need for sensor interoperability has increased tremendously because of the widespread deployment of biometrics systems. in various applications and the request of users. The feature selection is an important issue among the roles of biometrics system. So the feature selection has become to a fundemental problem of sensor interoperability. To our best knowledge, no systematic study has been conducted to ascertain its effect on biometric systems.In this thesis, the relationships among person, sensor and feature are discussed. Two factors, i.e., the inherent performances gap and the drop of performance caused by coordinating two sensors, can be regarded as reasons of sensor interoperability problem. We discuss the feature selection for sensor interoperability using a case study in fingerprint segmentation. We have illustrated the impact of the feature selection on the sensor interoperability of a fingerprint system. Experiments show that the various features exhibit different sensor interoperability using different sensors.2. Fingerprint segmentation is an important step in an automatic fingerprint identification system. In this thesis, we proposed a hybrid fingerprint segmentation framework that combines block-wise method with pixel-wise method. Because these methods are both linear algorithm, the hybrid method is also linear. This hybrid method takes advantage of both the block- and pixel-wise methods. On one hand, it has lower computational complexity. on the other hand, experimental results indicate that the hybrid method performs much better than original ones.3. The average inter-ridge distance of a fingerprint image is an important characteristic of fingerprint texture attribute and widely used in many applications of fingerprint recognition. As an important parameter, the precision of estimation probably affects segmentation, enhancement and classification procedure of fingerprint recognition seriously.This thesis proposes a novel algorithm which employs discrete Fourier transform, discrete information entropy theory, and weighted Euclidean distance to compute the average inter-ridge distance of fingerprint image accurately. In order to evaluate the performance of the new method, an experimental scheme based on man-made experiment-data sets and typical fingerprint images is proposed. The experimental results show that the proposed method is able to estimate the average ridge distance accurately.4. In this thesis, we experimented with the idea that sensor interoperability of fingerprint system caused by differences in resolution and scanning area can be improved by introducing a novel scaling module. For the purpose of facilitating system configuration, we have developed a series of scaling methods, including scaling factors and the graph-or template-based scaling parameter estimation. In graph-based scaling methods, we have explored the application of various technologies in estimation of the average inter-ridge distance. In template-based scaling methods, we have developed an estimation algorithm using Delaunay triangulation algorithm. We have developed some practical measures for the scaling factors, e.g. manual estimation, fingerprint image restruction and the average inter-ridge distance based method. Experimental results indicated that the accuracy and robustness of fingerprint system can be improved effectively by embedding such a scaling module into the traditional framework of fingerprint systems under multi-sensor situation.5. In this thesis, we proposed a novel idea that an existing fingerprint system can be improved by introducing a system performance improvment module. The novel method we proposed is consists of two steps:starting from the multiple enrolled impressions, attempt to convert them to points in multidimensional space via analyzing one-on-one matching results; then a matching step is achieved by calculating the distance between query image and the centroid of multiple enrolled impressions in multidimensional space. We formulate this task as a distance computation in multidimensional space task. Experiments indicate that the system using our method performs much better than the original system.6. In this thesis, a novel polynomial method is proposed to solve the parameter estimation problem for Choquet integral. The basic idea of our method is to regard the problem as a sequential one at first; and then we use Bayesian inference method to solve the problem. Using our method, the computational complexity for the non-additive measure is reduced from O((n+K)*22n) to K*O(n2logn). Specifically, the semantic information of the 2n variables is not lost. This method can be utilized to real-time applications. We provide statistical performance guarantees for the proposed method.The experiments show that the performance achieved by using this method is better than that of traditional methods. We propsed a Mobius representation and Choquet integral based performance improvment method of fingerprint system.