ALERT | Awareness and Localization of Explosive-Related Threats

Computationally Efficient Simultaneous Segmentation and Image Reconstruction for CT X-ray Based EDS Systems Screening
F3-D

Download the 2012 Project Report

Despite recent technological advances, reliable detection of explosives in luggage and IEDs remains a challenging problem. For explosives detection in checked luggage, X-ray computed tomog­raphy (CT) is a widely-adapted sensing modality. CT X-ray based explosive detection algorithms are comprised of image reconstruction, segmentation and classification steps where first, three-dimen­sional X-ray attenuation images are formed. Next, the content of the images are decomposed into ho­mogenous connected regions. Finally, the connected components are identified. Typically the image formation and segmentation are computationally the most intensive steps of the Explosive Detection System (EDS) image processing chain. This research aims to develop computationally efficient si­multaneous image reconstruction and segmentation algorithms. In particular, we develop

1. analytic forward models for conebeam X-ray CT that can incorporate realistic system parameters, arbitrary imaging geometries and X-ray propagation models;
2. computationally efficient inversion algorithms for such models based on microlocal analysis;
3. computationally efficient simultaneous image re­construction and segmentation algorithms;
4. iterative filtered-backprojection type algorithms with sparsity constraints. The outcomes of this research have implications, not only in X-ray CT based EDS, but also other applications involving synthetic aperture radar and sonar, as well as X-ray CT medical imaging.