ALERT | Awareness and Localization of Explosive-Related Threats

False alarm reduction in automatedprocessing of X-ray Backscatter images
F3-A4

Download the 2012 Project Report

X-ray imaging systems relying on backscattered or transmitted energy are widely used in security screening of both individuals and vehicles. In collaboration with an industrial partner, we have focused on developing novel algorithms to address image processing challenges that arise with these systems. In a first thrust, we have focused on airport X-ray backscatter (XBS) systems, building on our prior (Y3) work. In these systems, radiation exposure to the travelling public is minimized by using extremely low X-ray dosage, resulting in images with relatively low signal-to-noise ratio (SNR). This creates a need for effective image denoising approaches that, unlike standard smoothing ker­nels, preserve edge information to allow for the accurate identification of threat objects. We have found non-local means (NLM) denoising, a recently developed patch based approach, to be effective for XBS denoising. However, NLM is very computationally demanding and can blur weak edges. We have therefore developed fast, edge-preserving NLM variants to address these issues and have dem­onstrated performance gains on data. More recently, we have launched a second thrust to investigate super-resolution zooming techniques for vehicle scanning systems. In this work we are investigating the use of learned dictionaries and sparsity-based reconstruction methods to correct for artifacts generated by vehicle motion.