Multimodal image registration for effective thermographic fever screening

TL;DR

This paper presents a multimodal image registration method combining landmark and edge detection techniques to accurately localize canthi regions in IR and white-light images for effective fever screening.

Contribution

It introduces a coarse-fine registration strategy that achieves high accuracy in localizing canthi regions for thermographic fever screening.

Findings

Registration accuracy within 2.7 mm

Effective localization of canthi regions

Improved fever screening precision

Abstract

Fever screening based on infrared thermographs (IRTs) is a viable mass screening approach during infectious disease pandemics, such as Ebola and SARS, for temperature monitoring in public places like hospitals and airports. IRTs have found to be powerful, quick and non-invasive methods to detect elevated temperatures. Moreover, regions medially adjacent to the inner canthi (called the canthi regions in this paper) are preferred sites for fever screening. Accurate localization of the canthi regions can be achieved through multi-modal registration of infrared (IR) and white-light images. We proposed a registration method through a coarse-fine registration strategy using different registration models based on landmarks and edge detection on eye contours. We evaluated the registration accuracy to be within 2.7 mm, which enables accurate localization of the canthi regions.