Enhancing Long-range Images for X-ray Inspection via Reconstruction

Industrial Problems Seminar 

Scott McCloskey
Kitware

Abstract

Several challenges stand in the way of long-range X-ray imaging and tomography, including sparse photon arrival (resulting in low signal-to-noise ratios), difficulty in source/detector placement (resulting in few views), and movement of the sensor and/or object (resulting in motion blur). The commonly-used convolutional blur model does not apply to long-range X-ray imaging, where only a few photons from a pulsed linear accelerator reach the detector. We develop a physically-informed forward model, including a novel blur model for the low photon count setting, and apply it within an analysis by synthesis framework. This allows us to generate high-quality images of the object that are consistent with the physical constraints imposed by the dim and blurry inputs. Our enhanced images have improved contrast and reduced noise, improving interpretability.