Explosives represent a continuing threat to aviation security. The US Department of Homeland Security (DHS) currently has requirements for future explosive detection systems (EDS) that include increased probability of detection and decreased probability of false alarm for a larger set of objects and with reduced minimum masses . An EDS uses computed tomography (CT) technology to measure the physical characteristics of objects in baggage. CT provides an image of objects based on their X-ray attenuation. The attenuation depends on the material being scanned and is also a function of the energy of the incident X-ray photons. In conventional CT systems, also called single-energy CT, the scan is performed with a single source spectrum and energy-integrating detectors, and the ability to determine the chemical composition of the scanned materials is limited. In Multi-Energy CT (MECT), multiple energy-selective measurements of the attenuation are taken. Since additional energy-dependent information can lead to enhanced material discrimination, MECT can potentially provide improved detection capability over conventional single-energy CT.
In MECT systems for security, potential threats are found by identifying and segmenting the scanned objects. Typically, one set of images (e.g., the conventional attenuation image) is used for object segmentation and another set of images (e.g., the effective atomic number and density images) is used for object labeling and discrimination of the resulting segmented areas. These two processes of segmentation and labeling are typically decoupled. The segmentation process, which is usually based on single-energy data, is challenging due to image streaks and metal artifacts. Multi-energy material information is not exploited in this process, even though it might improve the segmentation. In addition, the reconstruction of reliable material parameter images for labeling the materials is challenging due to artifacts and noise created by clutter objects in the luggage. In this project, we develop methods that perform object identification and segmentation jointly using MECT data. This joint approach allows for material information to be exploited in the object segmentation process and incorporates the segmentation into the object material identification. The result is reduced artifacts and improved object localization and material labeling.