Icme Grand Challenge Results on Heterogeneous Face Recognition: Polarimetric Thermal-to-Visible Matching

Abstract

This paper describes the IEEE ICME Grand Challenge on Heterogeneous Face Recognition (Polarimetric Thermal to Visible Matching), presents the submitted face recognition algorithms, and details the evaluation results. The challenge problem, sponsored by ICME and Polaris Sensor Technologies, is motivated by nighttime face recognition and compares state-of-the-art domain adaptive algorithms for cross-spectrum face recognition. Using unique databases containing corresponding polarimetric thermal and visible facial imagery, the algorithms were developed and independently evaluated. A brief summary of each algorithm is described, and the face verification performances in term of equal error rate (EER) and area under the curve (AUC) are reported. The best performing algorithm was a GAN-based approach submitted by the Rutgers University Team.