Localization of Eye Region in Infrared Thermal Images using Deep Neural Network

Abstract

Infrared (IR) thermal images are produced by recording the thermal radiation emitted from the object. It has been increasingly used in industrial installations, surveillance, evaluating human health, etc. Medical IR thermal images are used as a tool for early diagnosis of certain diseases in the human beings. It can detect the heat patterns of the body as well as irregular blood flow. Compared to visual images, thermal images have low resolution and low contrast. Hence in IR thermal images eye detection is more challenging, for ocular surface disease diagnosis. This article proposes an automatic eye localization method from IR thermal images using YOLO v2 object detector. The performance of eye localization in IR thermal images using YOLO v2 shows mean average precision 97% and mean intersection over union (IoU) 90% for test images.