A novel face detection algorithm using thermal imaging

Abstract

Face detection is an important step for successful face recognition system. Variation in lighting conditions result in degradation in performance of face detection. Thermal images are not affected by variation in illumination. In this paper, we have proposed a novel face detection algorithm by using thermal imaging. In our proposed algorithm, Otsu's thresholding method is used for converting the thermal images into binary form. Horizontal projection of the image is calculated to identify the global minimum that helps in identifying height and width of the head region. We have developed a UTAR-YK Thermal database to verify the performance of the proposed algorithm. Experimental results show that our proposed algorithm has an average accuracy of 92.16%, that is measured by finding the overlapping ratio between the automatically-detected face region and manually drawn face region. An average 1.13 bounding box ratio, which is the comparison of the size of the box between the automatically-detected face region and manually drawn face region.