Research And Implementation Of Two Quality Evaluation Methods For Fingerprint Images

In recent years, the technique of automatic fingerprint identification has been a research focus in academia. Many researchers have done much work on fingerprint identification. But until now, some technical difficulties have always been baffling the development of fingerprint identification, such as the pre-processing and matching of low-quality fingerprints. Consequently, the analysis and measure of fingerprint quality becomes an important aspect of current fingerprint research. The effective resolution of such questions has significant meaning for development of fingerprint technology.In this paper, we have deeply research on fingerprint quality evaluation from two viewpoints: one is the analysis and study on quality features' evaluation performances, the other is the research of different fusion methods for quality features. The main research contents are as follows:1) Presently, the fingerprint quality index mainly bases on local or global analysis, namely block image evaluation and whole image evaluation. For the perfection of quality evaluation standards, this paper proposes a new hybrid way for analysis. Firstly, we use the Fourier spectrum of fingerprint to present the global quality, then the standard deviation method is used to compute the quality of the block image, and take the distance of the block to the singular point area as weight to average blocks' score. In the end, we combine the results of two methods to denote the whole fingerprint quality, make up the shortcoming of evaluating the singular point area simply using Fourier spectrum. The experiment results show that this is a reasonable and effective method.2) Aiming at the problem of different fusion methods for multi quality index, this paper proposes a quality evaluation way based on SVM. In the selection of quality index, we fully take into account local and global characters and spatial and frequency domain features, use three quality features, which are standard deviation, Fourier spectrum and gradient. These features are carefully analyzed, and the spatial distribution chart of quality is built based on the three features. Through careful observation, we find that it is non-linear separable. So SVM is presented to classify the fingerprints. We test this method in public database, and compare with single index evaluation methods. The result shows that the proposed method can evaluate fingerprint quality more exactly.