This dissertation addresses different image processing problems faced during the development of two different identification systems (i) an automated system for postmortem identification using dental records (dental radiographs), (ii) an automated ear identification system. Automating the postmortem identification of deceased individuals based on dental characteristics is receiving increased attention especially with the large number of victims encountered in mass disasters, as 9/11 attack, and Tsunami. The Automated Dental Identification System (ADIS) can be used by law enforcement agencies to locate missing persons using databases of dental x-rays of human remains and dental scans of missing or unidentified persons. ADIS provides functionality for users to upload the reference records, and submit identification queries using submitted records. ADIS then produces a short matching list of possible matches for the dental experts to verify. The ear identification system can be used at access point of restricted areas; this system helps identify a person from surveillance videotapes.

This dissertation introduces new high performance approaches for three image-processing problems of the ADIS record preprocessing stage. For the first, we introduce an automatic hierarchical approach to the problem of cropping dental image records into films. Our approach is heavily based on concepts of mathematical morphology and shape analysis. Testing reflects an overall error of ∼ 3%. For the second, we address the problem of teeth contour extraction using active contour without edges. This technique is based on the intensity properties of the overall region of the tooth image. It extracts a very smooth and accurate tooth contour. For the third, we enhance the existing techniques for automatic classification of teeth into four classes (molars, premolars, canines, and incisors); as well as the construction of a dental chart, which is a data structure that guides tooth-to-tooth matching. We tackle this composite problem using appearance-based features (low computational-cost) for assigning an initial class, followed by applying a string matching with don't care technique based on teeth neighborhood rules. Adding the don't care character allows the technique to work in the presence of missing tooth, which represents 21% of the database. Our approach achieves 82% teeth labeling accuracy based on a large test dataset of films.

For ADIS, also we introduce new techniques for the problem of fast dental image retrieval. We use Eigen images to reduce the dimensionality of each tooth, as well as other teeth contour descriptors. The main features of this search engine are that it completes the search in order of seconds and it reaches a reasonable accuracy with a relatively short candidate list.

For ear identification, we develop different components of a viable automated method for ear identification system. We automate the Iannarelli ear identification system, which had been used manually for years. We extract the ear external and internal curves, and use these curves to calculate the different Iannarelli distance measurements. We evaluated the system performance based on statistical analysis of a large dataset of thousands ear images, where the identification rate is 90% for rank 1 image.