Periocular Recognition Based on Features from Thermal and Visible-Light Images

Periocular recognition can be used for authentication even when the mouth is hidden by a mask and is thus potentially useful in a wide range of situations. A method is proposed here that uses not only visible-light images but also thermal images for periocular recognition with the expectation that the method will be robust against changes in illumination and facial expression. The NVIE database was used for experimental evaluation. It consists of visible-light and thermal facial images with various expressions captured in different lighting environments. After a score for each subject image was calculated from the features, score-level fusion was performed using the XGBoost algorithm. Experimental results show that our proposed method is more accurate than a method using only visible-light images.