The Research On Method Of Automatic Fingerprint Classification

Fingerprint Classification is the most important part of Automatic Fingerprint Identification System(AFIS), which has been the hot spot and difficult point for a long time. Fingerprint classification classifies the input fingerprint into a particular category n a reliable and steady manner, making the query just in a subset of the fingerprint database, thus reducing the response time. If a fingerprint classification system that has high classification speed and accuracy is designed, it can not only enormously improve the use efficiency of the AFRS, but also the store of character data and search of database will be more convenient.On the research of fingerprint classification, although many algorithms have been proposed, it is still a challenging problem. For fingerprint classification, the current research works mainly focus on the two fields:The first field is how to improve the accuracy of Henry classification mode. The second field is to extract a new fingerprint classification feature and classify fingerprints into some new classes. This paper researches the above two fields respectively, and puts forward a new classification method for each research direction respectively.The main works and innovations are as follows:1. The reference point detection method based on orientation pattern labeling in fingerprint images is improved. When the reference point can not be detected by only one labeling, the method based on orientation labeling solve this problem by continuously changing the quantization ranges and relabeling the orientation image. This process can not guarantee that all fingerprints can detect a reference point and this process is time consuming. This paper solves the problem by using the image morphology method. When the reference point can not be detected by only one labeling, the proposed method solves this problem by continuously eroding and relabeling the orientation image. Experiment results show that the improved method has a quicker speed and a higher detection rate than the original orientation pattern labeling algorithm.2. A local ridge distance based approach for fingerprint classification is presented. It firstly extracts a local region centered at the reference point. Secondly, it computes the local ridge distance of every fingerprint by using the Fourier transform. Lastly, classify fingerprints of fingerprint database according to their local ridge distance, thus fingerprints that have the same local ridge distance are in the same class. Experiment results show that this approach can speed up the retrieving of the fingerprint database efficiently.3. A fingerprint classification method based on the local orientation feature and ELM is presented. It firstly extracts a local region centered at the reference point. Secondly, it computes the orientation for each local block of the region. The sine and cosine of the twice orientation of each block are used to form a classification feature vector. Finally, ELM is used to classify fingerprints into five categories:arch, tent arch, whorl, left loop and right loop. Experimental results illustrate that the classification feature is more robust and the classification method can perform better on both speed and accuracy.4. A fingerprint classification system based on the local orientation feature and ELM classifier is completed. This paper by putting the code of the reference point detection, the direction field computation, local orientation feature extraction and classification by ELM classifier together to achieve the fingerprint classification system.