2D/3D biometric and biomedical image matching using line constraints
Date of Issue2013
School of Computer Engineering
The work reported in this thesis is a research on 2D and 3D biometric and biomedical image registration, which is of current interest. Three body parts (palmprint, hand veins, and brain surfaces) identification/ registration have been investigated in this research. The three parts are selected from three categories (outer surface, beneath the skin, and internal organ). The goal of the research is to develop 2D/3D algorithms for robust and effective matching in palmprint/vein identification systems or surface registration. Biometrics is the study of measurable biological characteristics of identifying a person using their physical or behavioral characteristics. Personal identification and verification based on two-dimensional (2D) biometrics features have attracted more and more attention in today’s society. Among the biometrics features, palmprint/hand vein have high potential to become major ones. The tradeoff between discriminative power and user friendliness is a big challenge to today’s palmprint/hand vein identification systems. During the extraction process of palmprint/hand vein, there will always be a problem of distorted or broken lines which causes difficulty in the matching process. Moreover, most palmprint/vein matching systems principally rely on perfect alignment, which is a challenging job in image preprocessing. Researchers used special image acquisition devices to restrict the movement and positioning of the hand during the acquisition process in order to improve the image quality to solve rotation and translation problem. However, the lack of mobility consequently results in the restrictive application of palmprint/vein authentication system. Motivated by the above difficulties, we are interested in developing a rotation/translation invariant algorithm capable of solving the broken line problem as well. In this work, we propose a novel 2D approach for palmprint/hand vein matching. The lines are retrieved as primitives and presented as line segments. A set of pair-wise angle relationships among lines are extracted to present the local structure of line pattern. Pair-wise angle relationships are simple, invariant to translation and rotation, robust to end-point erosion and segment error, and sufficient for discrimination. In order to keep positional information as well, dynamic region division technique is proposed. The algorithm has been applied to three different databases including a benchmark palmprint database, a vein pattern database, and a mobile phone palm print database and achieved 100%, 87.5%, 83.5% accuracy respectively. Besides using lower quality images, experiments have also been concluded to test the sensitivity of the algorithm when the palm print images have been modified or distorted. The results of the experiment show a high 97% match when palm print images have been skewed to 10 degrees. The matching experiment has shown encouraging results, which implicate that line segments can provide sufficient information for palmprint and hand vein recognition. Moreover, an interesting application, fortune telling has been investigated and been proved to be useful. After getting promising results in matching 2D biometrics images captured from surface of human body, we extend our research to some part inside human body. The brain stem and vestibular system are two of the key organs in the balance control system. Significant anomalies of the balance function, proprioception and oculomotor reflexes have been reported, and morphological difference has been found in the vestibular system. However, there is no reported work on studying the morphological alterations in brain stems in the Adolescent Idiopathic Scoliosis (AIS) subjects in an objective and quantitative way. Motivated by this, we developed mathematical models, which facilitate the morphometry of brainstem surfaces in AIS and normal controls. In 3D surface registration, we present a rigorous algorithm to register brainstem surfaces for the disease analysis of Adolescent Idiopathic Scoliosis. The basic idea is to extract four consistent features, which describe the global geometry of the brainstem, to guide the surface registration. Using the Ricci Flow method, brainstem surfaces are parameterized conformally onto quadrilaterally-faced hexahedron, which naturally induces the feature landmark-matching brainstem registration. Our registration algorithm can guarantee the exact landmark correspondence between brainstem surfaces. A shape energy can then be defined to measure the local shape variations between different brainstem surfaces. We have tested our algorithms on 30 real brainstem surfaces extracted from MRIs of 15 normal subjects and 15 AIS patients. Experimental results show the efficacy of the proposed algorithm to register brainstem surfaces, which matches landmark features consistently. The computed registration can be used for the morphometry of brainstems. It has high potential to apply our proposed 3D surface registration to biomedicine applications.
DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition