Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.250445
Title: Automated materials discrimination using 3D dual-energy X-ray images
Author: Wang, Ta Wee
ISNI:       0000 0001 3561 5011
Awarding Body: Nottingham Trent University
Current Institution: Nottingham Trent University
Date of Award: 2002
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
The ability of a human observer to identify an explosive device concealed in complex arrangements of objects routinely encountered in the 2D x-ray sereening of passenger baggage at airports is often problematic. Standard dual-energy x-ray techniques enable colour encoding of the resultant images in terms of organic, inorganic and metal substances. This transmission imaging technique produces colour information computed from a high-energy x-ray signal and a low energy x-ray signal (80keV < E < 140keV). The broad nature of this materials discrimination places plastic explosive in the same organic window as other innocuous organic items. Also, images of a threat substance which has been masked by other materials will result in colour encoding which is proportional to the effective atomic number of the threat material and the masking material. The work presented in this thesis enables the effective atomic number of the target material within a specified window (6.6 < Ze/f < 13) to be automatically discriminated from many layers of overlapping substances. This is achieved by applying a basis materials subtraction technique to the data provided by a wavelet image segmentation algoritlim. This imaging technique is reliant upon the image data for the masking substances to be discriminated independently of the target material. Further work investigated the extraction of depth data from stereoseopic images to estimate the mass density of the target material. A binocular stereoscopic dual-energy x-ray machine previously developed by the Vision Systems Group at The Nottingham Trent University in collaboration with The Home Office Science and Technology Group provided the image data for the empirical investigation. This machine utilises a novel linear castellated dual-energy x-ray detector recently developed by the Vision Systems Group. This detector array employs half the number of scintillator-photodiode sensors in comparison to a conventional linear dual-energy sensor. The castellated sensor required the development of an image enliancement algorithm to remove the spatial interlace effect in the resultant images prior to the calibration of the system for materials discrimination. To automate the basis materials subtraction teclmique a wavelet image segmentation and classification algorithm was developed. This enabled overlapping image structures in the x-rayed baggage to be partitioned. A series of experiments was conducted to investigate the discrimination of masked target materials. It was found that the system noise produced significant errors in the polynomial equations used for estimating the thickness of the aluminium and plastic basis materials. However, a successful demonstration of an automated technique for discriminating a plastic plate in some realistic scenarios has been demonstrated. Although, the technique will only work correctly if the materials masking the target fall within the window of effective atomic number defined by the chosen basis materials. Thus, for instance a steel mask would produce a false negative result. In order to discriminate it accurately a material would require the determination of its mass density. This could he provided within the reported basis material subtraction theory if the thickness of the target were known. However, the depth resolution (±6.7mm) produced by the experimental stereoscopic system was found to he too coarse for inclusion in the automated material discrimination program.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.250445  DOI: Not available
Keywords: Pattern recognition & image processing
Share: