The use of image segmentation techniques as a preprocessing step in luggage detection process is due to that the decomposition of a relatively complex scene into several simple sub-scenes can ease the analysis of the scene. Specifically, effective image segmentation approaches are able to have (1) each sub-scene contain a set of items possessing some particular features; (2) the objects (threats) be more visible in a simple sub-scene; and (3) interference of harmless items reduced.
The objective of x-ray image segmentation is to seek image segmentation algorithms effectively extracting features or objects of interest from a given luggage scene. Namely, image segmentation approaches are expected to show items invisible or unclear in the original scene such that the luggage scene can be easily and accurately interpreted by screeners.
Technical Approach:Image slicing and entropy-based multilevel thresholding approaches were mainly investigated. We selected several image slicing and entropy-based multilevel thresholding approaches and applied them to our luggage scans. Besides, a entropy-based multilevel thresholding approach, referred to as valley method, was suggested. Several key issues in this study include (1) the automatic selection of threshold values and number; and (2) time consumed to separate a common luggage scene by using a particular image segmentation algorithm.
Resulting slices generated by using iterated conditional modes (ICM) method in which the number of classes is set to 5. |
No publications currently available for this project.
This research is being conducted at the IRIS Lab by Yue Zheng under the supervision of Dr. Besma Abidi and Dr. Mongi A. Abidi.