الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 5 |  |  |
| | | from | 5 | | |
Abstract
A New Technique For On-Line Fingerprint Verification Systems Fingerprint verification is one of the most important biometric technologies, which has drawn a substantial amount of attention recently, because they can be easily attained, quickly classified, and very likely to be found at crime scenes. Therefore, fingerprints are collected and stored everyday in a wide range of applications. Generally, the insufficient performance of previous feature extraction and matching algorithms applied in on-line fingerprints verification systems are due to causes either concerning the accuracy or time requirements. The degradation of the accuracy of previous algorithms is a result of inefficient techniques applied in registration or feature extraction and matching. The inadequacy of time requirement is either due to a very high required computational load or a very high storage requirement. This thesis introduces a new technique for on-line automatic fingerprint verification system. This new technique consists of a new registration technique and a new feature extraction and matching technique. The new registration technique is called the three points reference point location technique (TRL). This technique depends on locating three different points by three different techniques (Wegstein’s technique, Poincaré-Index, and multiple resolution analysis of the orientation field). Then, the fingerprint image is divided into four quarters and the basic minutiae (i.e. bifurcations and endpoints) are extracted in each quarter independently. In the proposed technique, the features extracted in each quarter are treated before supplying them as an input to the Kohonen’s map. Finally, a Comparison between the new technique and the circular sampling technique (non-minutiae based technique) and the Hough-transform based technique (minutiae based technique) is made in order to test the performance by computing the matching errors (FAR and FRR) and the required CPU time for execution and by drawing the ROC curve. The comparison shows that the new proposed reference point location technique makes a reduction in the percentage of variation in X-coordinate and Y-coordinate in comparison with other reference point location techniques. Moreover, the performance of the new proposed technique is nearly as the performance of the Hough-transform based technique (minutiae based technique), but concerning the required CPU time for executing the proposed registration and matching technique is less than the Hough-transform based technique . In the proposed reference point location technique, each quarter of the fingerprint image, formed by the intersection of the two drawn lines by the new registration technique, is treated independently of other quarters. Thus, a corrupted or distorted region (quarter) has a less effect on the overall performance. On the other hand, the average insertion error and the average allowed variation between the 35 35 matrices supplying the Kohonen’s map of the input and template fingerprints is obviously less than that of the basic minutiae. Thus, it decreases the possibility of false minutiae and their effect on the performance on feature extraction and matching. In addition, the spatial self-organization of Kohonen’s map transforms feature similarity to storage location adjacency