Histogram Matched Visible and Infrared Image Registration for Face Detection

Face detection is a common image processing task in applications such as pedestrian monitoring, crowd flow prediction, security video surveillance, and fever screening. Thermal imaging is used in conjunction with a visible light camera for improved monitoring of people. Often information from multi-modal sources such as IR/visible images is fused for efficient face detection. The accuracy of the face detection system relies on the quality of the image registration procedure. In this paper, we propose an enhanced intensity-based image registration technique for thermal and visible facial images. The method uses a histogram matching scheme to equalize the intensity levels of the IR/visible images for proper image registration. The proposed scheme when validated with facial images shows an improved registration quality when quantified using structural similarity (SSIM) and mutual information (MI) metric for facial IR and visible images.