Research And Implementation Of Automatic Fingerprint Classification By Technology

As one of the most broadly used technology in biometrics-based verification, fingerprint recognition is widely used in the field of international trade,criminal act and a course of justice. So the research of an automatic fingerprint identification system(AFIS) has important technical value and practical value.There are many algorithms in field of fingerprint verification, but they still have some shortages to be researched. This thesis emphasizes on the research of automatic fingerprint classification and makes a discussion with the related content and proposes some correlative improved methods.1. Preprocessing. In this stage present a segmentation algorithm owing to Canny operator. This algorithm has a simple principle but good segmentation effect. It can comparatively keep the fingerprint pattern information as integrity. And this thesis improves the algorithm whose locate reference point basing on directional image, using method of replacing the iterative process with thickness template and reducing the algorithm's complexity. Through experimentation it shows that this algorithm locate the reference point more exactly than conventional poincare index algorithm.2. Feature extraction. First analyse the advantages and shortcomings of singularity extracting method which bases on poincare index and introduce a method which bases on statistic information;Then make improvements to the algorithm which extract features based on half region field of directions.we show the improved algorithm can extract the classified information more accurate and has revolves invariability.3. Classification. Select the Support Vector Machine (SVM) as the classifier and on the foundation of doing research on the multi-class classification,the thesis design a kind of SVM-based binary tree to carry on the multi-class classification. Numerical experiments on large problems demonstrate the effectives of the method over conventional methods such as muti-class SVM approaches with "one-to-one"and "one-to-the others",and it has the certain practical value.