Geometric based feature extraction and finger vein recognition based on feature fusion

Biometric recognition (BT) systems are a set of systems that automatically verify or identify in-dividuals' identities. During this process, characteristic features unique to each individual, such as finger vein (PD) pattern, palm vein pattern, fingerprint, retina, face, signature, voice, etc. are used. PD pattern, which is one of these features, is gradually expanding its areas of usage and popularity compared to other BT systems due to its additional advantages. Especially in places where security and applicability are prioritized, the use of PD recognition systems is increasingly expanding. In order to enhance the suc-cess rates of PD recognition systems and increase the efficiency of the system, important stages such as preprocessing and feature extraction are being continuously explored through new research. Studies on the reduction of noise in PD images, the extraction of new features, and the development of new met-hods using these features are brought to the literature. Studies are made to remove the noise, caused by infrared rays, or light scattering occurring from the structural properties of the tissue or optical and mo-tion distortions occurring in the process of acquiring PD images. In this thesis, the noise presented in the images was attempted to be removed by employing the Homomorphic Filter (HF) and Perona-Malik Anisotropic Diffusion (PMAD) methods during the preprocessing stage. Additionally, in order to enhance the success rates, two new features, namely Horizontal Total Proportion (HTP) and Vertical Total Propor-tion (VTP), have been proposed in addition to the features obtained from existing feature extraction methods used in the literature. These features have been introduced to the literature using the Horizontal and Vertical Total Proportion (HVTP) feature extraction method. The extracted new features were com-bined with both spatial and frequency domain features with the help of the feature fusion technique, and improvements in the success rates were observed by using different classification methods. Through these studies, the effective utilization of features associated with PD, as well as the enhancement of PD recognition performance, has been achieved, contributing to the development of more reliable and high-accuracy applications in identity recognition processes.