Date of Award
Master of Science in Computer Engineering
Department of Electrical and Computer Engineering
Gilbert L. Peterson, PhD
In criminal investigations, it is not uncommon for investigators to obtain a photograph or image that shows a crime being committed. Additionally, thousands of pictures may exist of a location, taken from the same or varying viewpoints. Some of these images may even include a criminal suspect or witness. One mechanism to identify criminals and witnesses is to group the images found on computers, cell phones, cameras, and other electronic devices into sets representing the same location. One or more images in the group may then prove the suspect was at the crime scene before, during, and/or after a crime. This research extends three image feature generation techniques, the Scale Invariant Feature Transform (SIFT), the Speeded Up Robust Features (SURF), and the Shi-Tomasi algorithm, to group images based on location. The image matching identifies keypoints in images with changes in the contents, viewpoint, and individuals present at each location. After calculating keypoints for each image, the algorithm stores the strongest features for each image are stored to minimize the space and matching requirements. A comparison of the results from the three different feature-generation algorithms shows the SIFT algorithm with 81.21% match accuracy and the SURF algorithm with 80.75% match accuracy for the same set of image matches. The Shi-Tomasi algorithm is ineffective for this problem domain.
DTIC Accession Number
Fogg, Paul N. II, "Forensics Image Background Matching Using Scale Invariant Feature (SIFT)Transform and Speeded Up Robust Features (SURF)" (2007). Theses and Dissertations. 2730.