Study On Fingerprint Image Quality Assessment Based On SVM Classification

With the increasingly high security requirements for individual identity, the biometric technology depending on physical characteristics of human is widely used in various areas of identity authentication. Fingerprint identification is the most effective biometric technology because of its individual uniqueness, stability, and simplicity. At present, the fingerprint identification technology has been widely used in network security, criminal identification, access control systems, ATM occasions and so on.Automated Fingerprint Identification System is an integrated automatic identification system including the modules of fingerprint image acquisition, processing, feature extraction, identification and matching. The performance of this system is heavily dependent on the quality of the fingerprint images. High-quality fingerprint images can ensure a high recognition rate. So if the quality of fingerprint images is first assessed, it is determined whether the texture is clear, whether the image can be used for feature extraction, fingerprint matching or identification. Therefore, low-quality fingerprint images can be refused so that the performance of Automated Fingerprint Identification System will be greatly improved.In this thesis, we do amount of research of the quality assessment for fingerprint image with reference on the quality assessment for digital image and the definition in biometric field. The main work is summarized as follows:(1) The fingerprint image quality is defined after the research on the methods of the quality assessment for digital image, in combination with the definition in the biometric field. This definition gives the content and direction about the quality assessment for fingerprint image in this thesis.(2) Three assessment factors are designed which are the image ambiguity factor, the texture feature factor and the utility factor. They are the comprehensive reflection of the fingerprint image quality from three aspects:the fuzzy extent of the overall image, the clarity of ridge and valley and the impact on the procedures of the fingerprint identification. (3) A quality assessment algorithm for fingerprint image based on SVM classification is proposed to overcome the the shortcomings of the scoring function assessment method with artificial setting the thresholds. And a multi-SVM classification algorithm is proposed to classify the fingerprint images according to different standards.(4) In this thesis, the proposed algorithms have been tested. With the removal of a certain proportion low-quality fingerprint images, the matching accuracy and ROC curve are tested. The results show that the assessment accuracy is impoved compared with the linear scoring function.