Performance Evaluation of Face Recognition Using Visual and Thermal Imagery with Advanced Correlation Filters

Abstract

This paper presents the face recognition performance evaluation using visual and thermal infrared (IR) face images with correlation filter methods. New correlation filter designs have shown to be distortion invariant and the advantages of using thermal IR images are due to their invariance to visible illumination variations. A combined use of thermal IR image data and correlation filters makes a viable means for improving the performance of face recognition techniques, especially beyond visual spectrum. Subset of Equinox databases are used for the performance evaluation. Among various advanced correlation filters, minimum average correlation energy (MACE) filters and optimum trade-off synthetic discriminant function (OTSDF) filters are used in our experiments. We show that correlation filters perform well when the size of face is of significantly low resolution (e.g. 20x20 pixels). Performing robust face recognition using low resolution images has many applications including performing human identification at a distance (HID). The eyeglass detection and removal in thermal images are processed to increase the performance in thermal face recognition. We show that we can outperform commercial face recognition algorithms such as FaceIt® based on Local Feature Analysis (LFA).