Detail Restoration and Tone Mapping Networks for X-Ray Security Inspection

Hyo-Young Kim,Seung Park,Sung-Jea Ko,Yong-Goo Shin,Seung-Won Jung

doi:10.1109/access.2020.3035086

Hyo-Young Kim, Seung Park + Show 3 more

Open Access

https://doi.org/10.1109/access.2020.3035086

Copy DOI

Abstract

X-ray imaging is one of the most widely used security measures for maintaining airport and transportation security. Conventional X-ray imaging systems typically apply tone-mapping (TM) algorithms to visualize high-dynamic-range (HDR) X-ray images on a standard 8-bit display device. However, X-ray images obtained through traditional TM algorithms often suffer from halo artifacts or detail loss in inter-object overlapping regions, which makes it difficult for an inspector to detect unsafe or hazardous objects. To alleviate these problems, this article proposes a deep learning-based TM method for X-ray inspection. The proposed method consists of two networks called detail-recovery network (DR-Net) and TM network (TM-Net). The goal of DR-Net is to restore the details in the input HDR image, whereas TM-Net aims to compress the dynamic range while preserving the restored details and preventing halo artifacts. Since there are no standard ground-truth images available for the TM of X-ray images, we propose a novel loss function for unsupervised learning of TM-Net. We also introduce a dataset synthesis technique using the Beer-Lambert law for supervised learning of DR-Net. Extensive experiments comparing the performance of our proposed method with state-of-the-art TM methods demonstrate that the proposed method not only achieves visually compelling results but also improves the quantitative performance measures such as FSITM and HDR-VDP-2.2.

Highlights

To detect unsafe or hazardous objects quickly and in a noninvasive manner, X-ray inspection systems have been extensively used in many security applications [1]
To train convolutional neural network (CNN) without requiring laborious expert retouching or traditional tone mapping (TM) methods, this paper presents a dataset synthesis technique based on the BeerLambert law [32] as well as a novel loss function that can be used to train our networks without paired HDR-LDR images
EXPERIMENTAL RESULTS we present qualitative and quantitative performance comparison results to demonstrate the superiority of the proposed method compared to the state-of-the-art TM methods [10], [21], [24], [29], [37]

Summary

Introduction

To detect unsafe or hazardous objects quickly and in a noninvasive manner, X-ray inspection systems have been extensively used in many security applications [1]. To visualize HDR X-ray images that have 12- to 16-bit precision on a standard 8-bit display device, X-ray inspection systems often apply tone mapping (TM), which shrinks the intensities of the HDR image to the target display range [1], [3]. Global TM methods apply the same mapping function to all the pixels in the HDR image [4]–[11]. Ward [5] used a simple linear function to compress image contrast instead of the absolute luminance. Larson et al [4] applied a histogram adjustment technique to preserve the histogram distribution of the original HDR image. Khan et al [11] applied a histogram-based TM after perceptual quantization to enhance the dark regions and compress the bright ones. Global TM is simple, fast, and can preserve the

Methods

Results

Conclusion