Comparison of Feature Descriptors for Visible and Thermal Face Recognition

摘要

Face recognition, as one of the biometric methods, in controlled environments reaches recognition accuracy of more than 99%. Regardless of that, variations in lighting conditions, pose, facial expressions and occlusions result in performance degradation. Including thermal infrared facial images in face recognition systems can provide a solution to these problems, especially for limitations caused by illumination. Both thermal and visible face recognition systems can use the same, common image processing feature descriptors. This paper compares the two most common feature descriptors, HOG and LBP, and examines their performance both in visible and thermal face recognition applications in order to find more effective feature descriptor for each image sensor.