Fingerprint Image Enhancement And Classification

According to the fast improvement of computer technology, automated fingerprintidentification system is the main research topic of biometric recognition. However, theperformance of fingerprint identification systems is limited by the quality of fingerprintimages which are often influenced by different factors. So it is necessary to improve thefingerprint image quality and to make them fit for the requirement of real applications.The preprocessing of the fingerprint images was discussed in this paper and much moreattention was paid to fingerprint image enhancement with orientation filtering and Gaborfiltering, some improvements was made to the Gabor filter.At first, we discussed the traditional methods of fingerprint preprocessing such asimage normalization, fingerprint image segmentation and the computation of orientationfield. And then we introduce a continuous orientation field algorithm which is moreprecise and smooth than former algorithms, and provides the reliable basis for the work.And then orientation filter and Gabor filter algorithms for fingerprint image enhancementwere introduced and studied in this paper. Specifically for orientation filtering we adoptthe continuous orientation field. On the research of Gabor filter, two methods ofcomputing texture frequency and the impact of a few parameters over Gabor filter werepresented. Since tow-dimensional Gabor filter can be separated into one-dimensionalGaussian band pass filter and one-dimensional Gaussian low pass filter which isperpendicularly, two-dimensional Gabor filters is implemented by two one-dimensionalfilters. Thus the complexity is reduced. Experimental results indicate that the newalgorithm has higher ability to resist noises and to enhance quality of fingerprint image.The fingerprint classification is an important link of automation fingerprintidentification system. This paper roughly introduces some fingerprint classificationalgorithms nowadays, especially the classification combined by binary tree and SVM(Support Vector Machine). This algorithm uses binary tree theory to decompose theproblem into three 2-class classification problems, then uses Support Vector Machine tooptimize the three hyperplanes, The combination of the two exerted the superiority for2-class classification of SVM over other algorithms completely. Experimental resultsshow that this algorithm improves the efficiency of fingerprint classification.