Fusion of operators for heterogeneous periocular recognition at varying ranges

Abstract

Cross-spectral matching of active and passive infrared (IR) periocular images to a visible light periocular image gallery is a challenging research problem. This scenario is motivated by a number of surveillance applications such as recognition of subjects at night or in harsh environmental conditions. This problem becomes even more challenging with a varying standoff distance. To address this problem a new compound operator named GWLH that fuses three local descriptors – Histogram of Gradients (HOG), Local Binary Patterns (LBP) and Weber Local Descriptors (WLD) – applied to the outputs of Gabor filters is proposed. The local operators encode both magnitude and phase information. When applied to periocular regions, GWLH outperforms other compound operators that recently appeared in the literature. During performance evaluation LBP, Gabor filters, HOG, and a fusion of HOG and LBP establish a baseline for the performance comparison, while other compound operators such as Gabor followed by HOG and LBP as well as Gabor followed by WLD, LBP and GLBP present the state-of-the-art. The active IR band is presented by short-wave infrared (SWIR) and near-infrared (NIR) and passive IR is presented by mid-wave infrared (MWIR) and long-wave infrared (LWIR). In addition to varying spectrum, we also vary the standoff distance of SWIR and NIR probes. In all but one case of the combination of spectrum and range, GWLH outperforms all the other operators. A sharpness metric is introduced to measure the quality of heterogeneous periocular images and to emphasize the need in development of image enhancement approaches for heterogeneous periocular biometrics. Based on the statistics of the sharpness metric, the performance difference between compound and single operators is increasing proportionally with increasing sharpness metric values.