The Research Of Multilevel Fingerprint Classification Based On Large-scale Database

Automatic Fingerprint Classification is the most important part of AutomaticFingerprint Identification System (AFIS), which has been the hotspot and difficulty tothe national and abroad researchers for a long time. With the rapid development andapplication of large-scale database in the society assurance system, have been turnedto the kernel topic of AFIS that how to reduce the searching scope and improve theperformance. Based on the large fingerprint database, this thesis contains thefollowing contributions:1. A manual fingerprint generation and simulation algorithm is developed in thisthesis, since a large number of fingerprint samples and testing data are necessary tothe performance verification of the classification algorithm. The fingerprint imagesgenerated by the two algorithms are valuable to experiment and similar to thatcollecting through acquisition devices on texture features, image quality and otheraspects. Our fingerprint database is established based on the standard fingerprintdatabase, acquired fingerprint database, the manual generated and simulatedfingerprint database, which ensures the subsequent preprocessing and classification.In some promotion significant, the construction of large-scale fingerprint databasesystem provides a new idea to the testing and verification of automatic fingerprintrecognition algorithm additionally.2. Due to the variations in impression conditions, acquisition devices, ridgeconfiguration, and skin conditions, such as wounds, scratch, tearing wounds, dry, wetand oil, different kinds of noises always appear in the collected fingerprint images.The fingerprint image preprocessing is very important to ensure the accuracy ofclassification and improve its robustness. Based on mathematical morphology andcomputer imaging technology, this thesis has studied fingerprint image qualityevaluation, calculation of baseline frequency and multi-scale orientation map,clustering of ridge types and adaptive preprocessing procedure. In this regard, afingerprint image segmentation algorithm combined of statistical gray characteristicsand the orientation map is proposed. Furthermore, the circular Gabor filters isemployed to enhance the fingerprint image. This algorithm has better segmentation effect and strong adaptability than whatever;3. A multi-level fingerprint classification system based on the interaction text,fingerprint type, the average number of ridge lines between core point and delta pointand the mean period, is presented. The interaction text can be added manually toenhance the matching speed and improve the matching accuracy of classification.Also, instead of the manual interaction, the three automatic classification factors canbe utilized to implement the automatic classification of the large-scale fingerprintdatabase.Extensive experiments show that the proposed multi-level fingerprintclassification approach could result in accuracy conclusion, and the classify factor hassignificant representation, as well as the classification algorithm is effectiveness andefficiency.