Face re-identification across pose in thermal infrared spectrum based on local texture descriptors

Face re-identification is a challenging task which is aimed to check similarity of two faces shown in the images. Face recognition system have been investigated since many years mostly in visible domain. We investigate face recognition methods based on facial images acquired in far-infrared range (thermal spectrum). The main reason for using thermal infrared for face recognition is to observe people in night conditions. However, this task is not free of challenges. In this paper we investigate the impact of various head positions on efficiency of face re-identification. The paper presents our measurement approach, results of many series of tests as well as performance metrics of re-identification based on three state-of-the-art facial descriptors.