Free-Form Deformation Approach for Registration of Visible and Infrared Facial Images in Fever Screening

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Infrared thermography (IRT) is used for fever screening during pandemics.
• Accurate core body temperature estimation relies on identifying specific facial regions like the inner canthi.
• Identifying these regions in IR images is challenging due to limited features compared to visible images.

Purpose of the Study:

• To develop and evaluate multi-modality image registration (MMIR) techniques for accurate canthi localization in IR facial images.
• To improve automated fever screening by enabling precise temperature measurements.

Main Methods:

• Developed MMIR techniques combining visible and IR facial images.
• Employed free-form deformation (FFD) models based on edge maps for enhanced registration accuracy post-affine transformation.
• Compared Demons and cubic B-spline FFD algorithms for medical image registration.

Main Results:

• The Demons algorithm demonstrated superior performance over the cubic B-spline algorithm in FFD.
• Registration accuracy achieved was 2.8 ± 1.2 mm using control point correspondence.
• This accuracy enables reliable automated localization of canthi regions in IR images.

Conclusions:

• MMIR techniques, particularly using the Demons algorithm, effectively improve canthi localization in IR facial images.
• The developed methods facilitate accurate and automated fever screening for asymptomatic individuals.
• Enhanced registration accuracy supports the use of IRT for public health surveillance during disease outbreaks.