Purchase this article with an account.
Natsuko Toyofuku, Thomas F. Schatzki; Feasibility of feature-based contraband detection in x-ray images. Journal of Vision 2005;5(8):958. doi: 10.1167/5.8.958.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Detection of illegal and prohibited items (contraband) passing through airports is a complicated and difficult task. The USDA uses x-ray inspection, interviews, and luggage and canine searches to keep our agriculture safe from foreign bugs & disease brought in through air traffic. Contraband includes, but is not restricted to, fresh fruits, plants, meat, soil, grain, and seeds.
X-ray inspection is relied upon because it is fast and allows for a non-invasive look into the contents of passenger baggage. Inspectors use the x-ray images to judge if a bag must be physically opened and searched for contraband. The drawback of x-ray inspection is that the images are monochromatic, compressed jumbles of lines and textures that must somehow be translated into presumed 3D representations that can be recognized as contraband items. This requires months of experience and is difficult to train for (there is no consistent vocabulary to describe “what to look for”).
We are investigating a new image feature based (IFB) approach to improve inspection performance. Instead of looking for specific contraband (e.g. fruit), inspectors will look for particular patterns of curves or textures associated with contraband items. IFB is fast, adaptable, provides a consistent vocabulary, and does not require months of onsite training.
Contraband occurs in ∼5% of bags. The number of bags that can be searched is limited by staff capacity, thus increasing the frequency of contraband-containing bags (CB) in that selection is desirable. Current max search capacity is ∼10% of all x-rayed bags. X-ray inspection results in a CB frequency of 18% (vs 5%). However, 82% of the bags were searched unnecessarily, resulting in passenger delays and wasted manhours. Using IFB and pulling only 3.6% of the bags, CB frequency would be 10%. If IFB pulled 10% of the bags, we predict that CB would increase to ∼21%. Training naïve observers in IFB and refinements of the feature set will be investigated next.
This PDF is available to Subscribers Only