Anomaly Detection in Advanced Imaging Technology Systems Based on Graph Theory [INACTIVE]
NOTE: As a result of the ALERT Biennial Review conducted in March of 2018, this project has been concluded and will not be funded in Year 6.
Explosive are often difficult to detect in an imaging modality due to obscuration by background clutter. The difficulty arises from the fact that both the shape, size, intensity of various parts of the background as well as the explosive exhibit significant variability. Consequently, methods that are capable of discriminating anomalous regions in the context of variable background are required. We propose anomalous cluster detection methods for detection and identification of irregular shapes and outliers in non-stationary background. Our proposed method is based on modeling background image as a spatial random field supported on a spatial graph and identifying clusters on the graph that are likely to be anomalous. The problem of searching for anomalous shapes is combinatorial and we propose to leverage new results on mirror-descent for semi-definite programming problems for rapid detection and identification of explosives.
One of the main challenges in dealing with the threat of terror is the appearance of unknown improvised explosives. There is a need for rapid assessment of the yield, sensitivity, and safe disposal of these explosives. [...] We suggest using computational methods as a first rapid response to these threats.- Year 5 Annual Report
Faculty and Staff Currently Involved in Project
Students Currently Involved in Project
- Alp Durmus Acar