Denoising approaches for X-ray personnel screening systems

Abstract

Personnel screening using X-ray backscatter (XBS) systems or other advanced imaging technologies is a key component of aviation security. XBS images specifically reveal objects placed on the body, such as contraband and security threats, as well as anatomical features at or close to the surface. In fielded XBS systems, the 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), with noise being Poisson in nature. 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 that non-local means (NLM) denoising, a recently developed patch based approach, is an extremely effective edge-preserving technique for XBS denoising. Here each image patch is compared to other image patches and a similarity-weighted average is performed. Unfortunately, the most widely used NLM techniques are computationally demanding. In this paper, we explore fast NLM denoising techniques and adapt recent algorithms to the XBS problem. In addition, we explore fast methods for improving NLM denoising along weak edges in the image. We demonstrate the effectiveness of the technique in terms of anomaly detection on a dataset of acquired XBS images.