Towards a model for estimating image difficulty in X-ray screening

In this study, we developed a first computational model for estimating image difficulty of X-ray images of passenger bags. Based on Schwaninger (2003) three image-based factors are proposed as predictors of image difficulty; view difficulty of the threat item, superposition by other objects, and bag complexity (i.e. clutter and transparency of the bag). First, these factors were validated using detection experiments. We then developed computer-based algorithms to estimate the image-based factors automatically. Finally, we could show that our computational model can better explain human performance than human ratings of the image-based factors