dc.description.abstract | Due to the explosive growth and demand of information security, biometric features based personal identification systems gradually dominate identification techniques in many applications. There are many physiological features, such as faces, fingerprints, and iris images, which have been extensively studied for personal verification purpose in the past few decades. In the past, many literatures discussed biometric verification by only using either palmprint features or palm-dorsa vein-patterns features. Both of them are constrained by some certain limitations (e.g., utilization of fixed pegs to constrain the palm position while acquiring palm images, and requirements of adequate lighting conditions) which hinder the practicality of applications.
In this thesis, a novel method is proposed to remove the limitations imposed by docking devices so that visible thermal images can be captured from the pegs-free platform. Furthermore, the palm features are successfully extracted from interest of interest (ROI) by using the novel proposed histogram of iterative thresholding (HIT) technique. Finally, we combine the two palm features, which are principal palmprints and palm-dorsa vein-patterns to enhance the accuracy rate.
In our work, template matching and support vector machine (SVM) are designed to verify the query images. First of all, linear correlation function is adopted in template matching, and then query image is matched with the reference templates to measure the similarity between the two different palm images. Secondly, we train the data with the selected RBF kernel in SVM. The k-fold cross validation is used to obtain a 98.2% accuracy rate. Experimental results demonstrate that our proposed algorithm is reliable, feasible, and adaptable in practical applications.
| en_US |