Ph.D. Tezi Görüntüleme | |||||||||||||||||||||
|
|
||||||||||||||||||||
| Summary: In this study, it is aimed to develop a palmprint verification system that is portative, completely free background and does not have any hand movement restricting apparatus. In such a system, the hand must be successfully segmented and alignment problems must be resolved. In the study, hand segmentation was performed with Active Appearance Model based segmentation. Alignment problems encountered when the hand is not parallel to the camera plane have been tried to be overcome with two new original approaches (Palm2AAM and AAM2Palm). The proposed approaches within the scope of the study require depth knowledge. For this purpose, a stereo image acquisition device was prepared using two cameras and 3 new palmprint dataset were collected with this device. Experimental studies on these datasets have examined the effects of the proposed methods on verification performance. In the experiments, both coding-based and subspace-based methods were applied for palmprint feature extraction. In particular, it has been shown that coding techniques, which have been proven successful in the literature, are very unsuccessful in images under perspective distortion. Subspace based methods approach to desired performance values when used with Gabor filter. In addition to these, three combined models have been proposed with the help of the advantages that the stereo camera system offers and the effect of the hand position on the verification performance was shown by determining the hand posture angle.
Key Words: Unrestricted Palmprint Verification, Stereo Camera, Palm2AAM, AAM2Palm, Perspective Correction. | |||||||||||||||||||||