Semester

Spring

Date of Graduation

2013

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Hany Ammar

Committee Co-Chair

Donald A Adjeroh

Committee Member

Guodong Guo

Committee Member

Richard Jurevic

Committee Member

Xin Li

Committee Member

Matthew C. Valenti.

Abstract

Automated dental identification is one of the best candidates for postmortem identification. With the large number of victims encountered in mass disasters, automating the process of postmortem identification is receiving an increased attention. This dissertation introduces new approaches for different stages of Automated Dental Identification system: These stages include segmentations, classification, labeling, and matching:;We modified the seam carving technique to adapt the problem of segmenting dental image records into individual teeth. We propose a two-stage teeth segmentation approach for segmenting the dental images. In the first stage, the teeth images are preprocessed by a two-step thresholding technique, which starts with an iterative thresholding followed by an adaptive thresholding to binarize the teeth images. In the second stage, we adapt the seam carving technique on the binary images, using both horizontal and vertical seams, to separate each individual tooth. We have obtained an optimality rate of 54.02% for the bitewing type images, which is superior to all existing fully automated dental segmentation algorithms in the literature, and a failure rate of 1.05%. For the periapical type images, we have obtained a high optimality rate of 58.13% and a low failure rate of 0.74 which also surpasses the performance of existing techniques. An important problem in automated dental identification is automatic classification of teeth into four classes (molars, premolars, canines, and incisors). A dental chart is a key to avoiding illogical comparisons that inefficiently consume the limited computational resources, and may mislead decision-making. We tackle this composite problem using a two-stage approach. The first stage, utilizes low computational-cost, appearance-based features, using Orthogonal Locality Preserving Projections (OLPP) for assigning an initial class. The second stage applies a string matching technique, based on teeth neighborhood rules, to validate initial teeth-classes and hence to assign each tooth a number corresponding to its location in the dental chart, even in the presence of a missed tooth. The experimental results of teeth classification show that on a large dataset of bitewing and periapical films, the proposed approach achieves overall classification accuracy of 77% and teeth class validation enhances the overall teeth classification accuracy to 87% which is slightly better than the performance obtained from previous methods based on EigenTeeth the performance of which is 75% and 86%, respectively.;We present a new technique that searches the dental database to find a candidate list. We use dental records of the FBI's Criminal Justice Service (CJIC) ADIS database, that contains 104 records (about 500 bitewing and periapical films) involving more than 2000 teeth, 47 Antemortem (AM) records and 57 Postmortem (PM) records with 20 matched records.;The proposed approach consists of two main stages, the first stage is to preprocess the dental records (segmentation and teeth labeling classification) in order to get a reliable, appearance-based, low computational-cost feature. In the second stage, we developed a technique based on LaplacianTeeth using OLPP algorithm to produce a candidate list. The proposed technique can correctly retrieve the dental records 65% in the 5 top ranks while the method based on EigenTeeth remains at 60%. The proposed approach takes about 0.17 seconds to make record to record comparison while the other method based on EigenTeeth takes about 0.09 seconds.;Finally, we address the teeth matching problem by presenting a new technique for dental record retrieval. The technique is based on the matching of the Scale Invariant feature Transform (SIFT) descriptors guided by the teeth contour between the subject and reference dental records. Our fundamental objective is to accomplish a relatively short match list, with a high probability of having the correct match reference. The proposed technique correctly retrieves the dental records with performance rates of 35% and 75% in the 1 and 5 top ranks respectively, and takes only an average time of 4.18 minutes to retrieve a match list. This compares favorably with the existing technique shape-based (edge direction histogram) method which has the performance rates of 29% and 46% in the 1 and 5 top ranks respectively.;In summary, the proposed ADIS system accurately retrieves the dental record with an overall rate of 80% in top 5 ranks when a candidate list of 20 is used (from potential match search) whereas a candidate size of 10 yields an overall rate of 84% in top 5 ranks and takes only a few minutes to search the database, which compares favorably against most of the existing methods in the literature, when both accuracy and computational complexity are considered.

Share

COinS