Work Package 5: Automated detection, comparison and correlation of X-ray images
Leader: Irène Dorion (Smiths Detection)
The main aim of this work package is to improve efficiency and effectiveness of cargo security screening by developing sophisticated algorithms for automated detection and screener assistance. Besides these automated systems, the interpretation and evaluation of X-ray images is simplified by optimised image rendering and the display of additional meta-information. Key ingredient for both automated detection and simplified screener work will be a data base storing reference images and meta-information. The range of candidate objects to be detected will be defined based on a risk analysis to be conducted with end users, scientific and security experts (see also WP2). Due to plethora of objects and the complexity of the surrounding cargo environment, supposedly, no generic algorithm will be able to detect all objects. Therefore, a set of different algorithms shall be designed and tested that individually detect specific threats only. The categorisation of threats shall be performed based on the urgency of detection, the set of available data, the potential to generate new reference images and the development effort for the new algorithms. The combination of these algorithms will ease and fasten operator work in an industrial-conform, cost effective way with next generation non-intrusive technologies for enhanced cargo screening and inspection
The first step is the definition of a new inspection concept (T5.1). The following steps are the studies of methods and algorithms to offer computer assisted detection and localisation of irregularities in cargo X-ray images (T5.2). Modern image and signal processing algorithms will be applied and further developed for automated detection of threats in X-ray images using different signatures based on material, structure and shape information. Applied visual cognition research will be used to identify how expert screeners recognize threats in order to further enhance automated detection and screener assist technology (T5.3) as well as classification methods (T5.4) fitted to these images.
The involved partners have extended experience in classification algorithms based on Support Vector Machines (SVM) and derivatives. Other methods applicable to X-ray images are well known by the involved partners and will be discussed at an early stage of the project to define the most promising ones versus the targeted applications (e.g., approaches based on some optional segmentation step followed by texture and pattern analysis are promising candidates to be combined with shape and orientation based approaches that may be derived from human vision mechanisms). From these methods and algorithms, a software implementation will be done and tested against the prepared reference data base. The detection results will be assessed as well as the improvement in term of the inspection capabilities through the use of assisted decision tools for checking the imaged cargo content versus freight transportation documents (T5.5). Thereby the actual parameters like photon energy, spatial resolution and efficiency of the detector system and the size of the X-ray focal spot are taken into account. As a result the adapted images provide a comparable contrast scale and equivalent sampling.
To Work Package 4 To Work Package 6