Abnormal Crowd Behavior Detection using Optical Flow
Are there real-time methods for detecting abnormal crowd behavior?
Slides from my dissertation defense are available below. More information forthcoming.
Exploring the Matching Score Space of Face Images
Does a face matching algorithm group people together in the same manner as humans?
This exploratory research project in face recognition focused on analyzing similarity scores from large-scale face matching experiments using hierarchical clustering. Similarity scores in face recognition represent the proximity between pairs of images as computed by a matching algorithm, and given a large set of images and the proximities between all pairs, a similarity score space can be defined. Hierarchical clustering analysis was applied to the similarity score space to develop various taxonomies. Given the number of subjects in the dataset, we used hierarchical methods to aggregate images of the same subject. We also explored the hierarchies occurring in larger groups, in which we saw clusters form based upon gender and ethnicity. Research in clustering biometric scores is still widely unexplored. Clustering can be used to prune datasets for faster matching experiments and to explore biases in face recognition algorithms. Thus, I aim to revisit these topics in future research projects.
Face Recognition of Identical Twins
Can a computer tell the difference between identical twins?
This research project studied the state of facial recognition of identical twins. For two consecutive years members of the Computer Vision Research Team at the University of Notre Dame collected high resolution images and videos from pairs of identical twins at the Twins Day Festival in Twinsburg, OH. The performance of face recognition on identical twins was compared using commercial and open source software. Additionally, we highlighted the stark difference in the non-match distribution between matching experiments containing twins and those without. This work resulted in two publications.