Author

Xuan Xu

Date of Graduation

2016

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Thirimachos Bourlai

Committee Co-Chair

Jeremy Dawson

Committee Member

Yuxin Liu

Abstract

Face, iris and fingerprint based biometric systems are well explored areas of research. However, there are law enforcement and military applications where neither of the aforementioned modalities may be available to be exploited for human identification. In such applications, soft biometrics may be the only clue available that can be used for identification or verification purposes. Tattoo is an example of such a soft biometric trait. Unlike face-based biometric systems that used in both same-spectral and cross-spectral matching scenarios, tattoo-based human identification is still a not fully explored area of research. At this point in time there are no pre-processing, feature extraction and matching algorithms using tattoo images captured at multiple bands. This thesis is focused on exploring solutions on two main challenging problems. The first one is cross-spectral tattoo matching. The proposed algorithmic approach is using as an input raw Short-Wave Infrared (SWIR) band tattoo images and matches them successfully against their visible band counterparts. The SWIR tattoo images are captured at 1100 nm, 1200 nm, 1300 nm, 1400 nm and 1500 nm. After an empirical study where multiple photometric normalization techniques were used to pre-process the original multi-band tattoo images, only one was determined to significantly improve cross spectral tattoo matching performance. The second challenging problem was to develop a fully automatic visible-based tattoo image registration system based on SIFT descriptors and the RANSAC algorithm with a homography model. The proposed automated registration approach significantly improves the operational cost of a tattoo image identification system (using large scale tattoo image datasets), where the alignment of a pair of tattoo images by system operators needs to be performed manually. At the same time, tattoo matching accuracy is also improved (before vs. after automated alignment) by 45.87% for the NIST-Tatt-C database and 12.65% for the WVU-Tatt database.

Share

COinS