A comprehensive study on flexible palmprint verification system design

In this work, as a first, two online palmprint verification systems have been designed and the performance evaluations have been carried out on these systems in real-world conditions. The systems, completely developed by the project team, served at the entrance of the PC rooms for three years. In this period, more than 20,000 verification attempts were achieved by over 1,000 users. High verification rates obtained by the online systems have motivated us to focus on free background and touchless systems? developments, which minimize the restrictions on use. In the second part of the works, the touchless system has been designed and tested with the offline experiments, include various real scenarios. In the thesis, the usage of high-performance Gabor wavelets based Kernel Fisher Discriminants method for the palmprint feature extraction; Active Appearance Models for the palm segmentation in the cluttered backgrounds have been proposed. Furthermore, a novel model based region of interest extraction method that uses the nonlinear regression of the palm model (to increase the selection of the palmprint accuracy); modified point-to-curve distance and margin width metrics (to evaluate the performance of the segmentation methods); and a flexible palm model (to utilize in the model based palm segmentation methods) have been developed. The images acquired by the online systems which have large variations arisen from the real-world conditions, and the images acquired by the touchless system, which simulate large number of posture variations and cluttered backgrounds, constituted the palm databases to put them into use of the researchers. The implementation of different verification scenarios and user-specific normalizations in the palmprint verification systems can also be considered as other contributions.