Segmentation and Feature Extraction Methods for Ocular Modalities

The demand for better and robust security is an undisputed part of our interconnected world. Biometrics based solutions offer an alternative to passwords and other form of authentications. Especially, Ocular biometrics has gained higher demand based on their proven recognition rates and reliable texture patterns over time. Physical and psycho-physiological aspects of human eye reveal unique signatures from physical appearance and cognitive state. Physical aspects of human eye include unique texture pattern from Iris, Retina and Sclera. Based on uniqueness, universality and availability, ocular modalities can be potentially be used as biometric token. Iris biometrics is considered as the optimal solution with given image acquisition in near infrared environment. Iris biometrics in visible spectrum (due to darker irises) and off-angle iris recognition is challenging. As an attempt to address the limitation from Iris recognition, we design a hyper focal imaging platform for visible spectrum iris recognition. Lateral lighting is used for illumination and hyper focal imaging technique is used to acquire super sharp iris image. In this study, visible spectrum iris recognition is successfully performed but the proposed setup limits its applications to cooperative users and requires additional light source. Usage of conjunctival vasculature (CV) recognition is one of the alternative solutions to overcome limitations from iris recognition. Red blood vessel pattern seen on white of eye is termed as conjunctival vasculature. CV can be potentially used as biometric modality due to its uniqueness, availability in visible spectrum and its maximum exposure in gazing. In this dissertation, we design and evaluate the scleral segmentation algorithms, vascular pattern enhancement techniques, feature extraction methodologies and feature enhancement methods for CV recognition. We introduce a new sclera segmentation algorithm using morphological properties of human eye in visible spectrum. This study demonstrates the effective use of the multi-scale multi-directional transform (MSMDT) for feature extraction from CV. Non-linear feature enhancement and feature mapping in various MSMDT domain are successfully demonstrated for the differentiation of texture. Psycho-Physiological aspects of human eye reveal cognitive state of human brain. In this dissertation, startle eye blink signals are used for assessment of credibility from 250 FPS video records. Using upper eyelid tracking at 250 FPS and ensemble of SVMs, we were able to detect deceptive individuals with sensitivity, specificity, correct rate of up to 0.7857, 1, and 0.8636 (5-fold cross validation).

Name thesis:degree_name

Ph.D. (Doctor of Philosophy)

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Electrical and Computer Engineering (UMKC)

Grantor dc:publisher

University of Missouri--Kansas City

Year dc:date.issued

2015

Author dc:creator

• Tankasala, Sriram Pavan

Advisor dc:contributor.advisor

• Derakhshani, Reza

Language dc:language.iso

en_US

Handle dc:identifier.uri

https://hdl.handle.net/10355/58686

OAI identifier oai:identifier

oai:mospace.umsystem.edu:10355/58686