Research On Crease Detection And Its Applications For Fingerprint Recognition

With long history of research on fingerprint verification and spread of applications, fingerprint recognition based identity verification becomes more and more popular and obtains more attentions. All of current products about fingerprint verification are based on the minutiae-matching algorithm. However, the tradition minutiae-matching algorithm will extract many spurious minutiae when processing fingerprint images with low quality, especially from old people. It is because some creases exist in the fingerprint. As a result, the recognition ratio will be decreased. How to remove the spurious minutiae generated by creases becomes a bottle neck for fingerprint verification technology. In this paper we propose to remove the spurious minutiae by detecting creases at first. Crease is regarded as the noise in the traditional methods, but in this paper we treat it as a feature of the fingerprint, an independent signal. We give the definition for the crease, study how to model it and detect it. Based on the basic study of the crease, we improve the minutiae-matching algorithm by firstly removing the spurious minutiae.Our paper focuses on the following aspects.We are the first in literatures to define the crease and lead a thorough study on it, including modeling and detection. We propose a mathematical model to represent the crease. We use a parameterized rectangle to simplify the crease representation for efficiently storage.We classify all creases into two groups: invariable creases and temporary creases. We propose two applications related to creases. One is to developing the minutiae-matching algorithm by firstly removing the spurious minutiae. The other is to developing the crease-matching algorithm.We design an optimal filter to detect creases, and prove the optimization. We select the appropriate parameters for the filter based on lots of statistics. At the same time, we estimate which creases will be detected based on the current filter. Finally, a multi-channel detection framework is developed for detection. Experimental results demonstrates the efficiency and robust of the detection method. Based on the detected creases, we implement the minutiae-matching algorithm by firstly removing the spurious minutiae, experimental results demonstrates that it is very helpful to improve the recognition performance by removing the spurious minutiae in advance. We also implement the crease-match algorithm for fingerprint verification. Although we have experimented on a small image set, results are very promising. This application will be continued to study in the future.