A personnel identity verification method using dab fingerprints

Automatic identity verification is becoming more and more important to secure transactions through the growing system of public networks. Among the biometrics currently used for identity verification, the fingerprint has a very good reputation because of its uniqueness and immutability. Its application to commercial transactions still offers some challenges because the system has to operate on-line and unsupervised, in contrast with semi-automatic operation in law enforcement applications.; This dissertation presents a new model for the analysis and synthesis of the ridge orientation of fingerprint patterns. The model uses a linear basis whose order may be adapted to fit the requirements on precision of the approximation. Since the model is linear, its parameters are efficiently calculated using standard linear transform techniques. Additionally, the model allows the evaluation of the specific contribution of each singularity to explain the ridge orientation everywhere. This property has been used to select the singularities in the process of synthesis of the ridge orientation.; The synthesis model has been successfully tested in the context of synthesis of the ridge orientation of 1380 fingerprint images. The results on ridge orientation synthesis show a mean absolute orientation error of 3.6{dollar}spcirc,{dollar} with a standard deviation of 1{dollar}spcirc.{dollar} The results show that the main property of the ridge orientation as a feature for identity verification is that it has a low rejection rate. The equal error point (equal false acceptance and false rejection rate) was found to be 4%. Additional results on the use of more than one test print, more than one reference print, and more than one finger per person are also presented.; The results of an experiment combining the global features (ridge orientation) and the local features (minutiae) at the level of the decision variable are also presented. They show that it is possible to improve the performance of the local features in the region of low false rejection rate using the global features. In this case, the false acceptance rate steadily increases for false rejection rate lower than 5%. The point of equal error for this experiment is 2.5%.