About Past Issues Editorial Board


Research Webzine of the KAIST College of Engineering since 2014

Fall 2023 Vol. 21

Bayesian Approach-based Ghost X-ray Fluorescence for Rapid Element Mapping

August 23, 2023   hit 333

 Ghost imaging-based X-ray Fluorescence enables fast elemental mapping on the surface of a sample. However, achieving accurate element distributions still necessitates a significant amount of measured data. This research demonstrates Bayesian approaches for reconstructing high-quality elemental distributions with less data.

 X-ray Fluorescence (XRF) analysis, a non-destructive inspection method, is a widely used technique for determining the elemental composition of materials. To provide the spatial element information on the surface of a sample, a raster scanning-based XRF system is generally exploited where an object or an x-ray tube is moved one by one. However, when it comes to a large-area sample requiring high-resolution mapping, the raster scanning approach may be time-consuming due to its inherent property. A ghost imaging-based XRF system (GXRF), which is a non-local imaging method, enables the measurement of characteristic X-ray energies emitted from the entire surface of a sample in a single measurement. Unlike the existing imaging technique, ghost imaging (GI) based on the correlation of two beams makes it possible to indirectly acquire an image by separating the imaging and detection processes. Owing to the non-local property of GI, GXRF can provide a notable advantage over the raster scanning-based system, resulting in faster measurement of element distributions. However, GI sometimes requires a large amount of data to acquire high-quality images due to the noise from statistical correlation.

Schematic of a ghost imaging-based X-ray fluorescence system for mapping element distributions and examples of element distributions (Zn and Cu).


Professor Gyuseong Cho's research team from the Department of Nuclear and Quantum Engineering at KAIST has proposed iterative Bayesian approaches to reconstruct high-quality ghost images using a smaller amount of correlation data. These approaches aim to improve the quality of 2D element distributions in the GXRF system. In the first study, using the statistical properties of a photon-counting process and the inherent characteristics of a natural image, a mathematic modeling is conducted and then a regularized maximum likelihood expectation maximization algorithm in the form of an iterative formula has been derived for retrieving a ghost image. Compared with conventional GI reconstruction methods, the proposed method significantly enhances the quality of ghost images. In the second study, a Bayesian denoising method based on Markov random field has been proposed to enhance the quality of reconstructed ghost images. This method incorporates the latent image from previous measurements as prior information in the subsequent reconstruction steps, enabling faster visibility of the unknown object.


This work represents the first attempt to reconstruct and post-process ghost images by using the statistical properties of measured data in the GI system. The findings could make the GXRF system more feasible as a practical application for mapping element distributions.