Thermal–visible registration of human silhouettes: A similarity measure performance evaluation

Infrared thermal facial image sequence registration analysis and verification

“…In this study, the NMI similarity index [20][21][22][23][24] is taken as the optimizing objective function of genetic algorithm [25][26][27][28], as shown in formula 2-3.…”

Automatic visible and infrared face registration based on silhouette matching and robust transformation estimation

“…This is appropriate since the silhouettes are quite stable and they are sufficient to recover a rigid/affine transformation. The idea of using silhouette has been previously studied mostly in the case of videos [7], while in our problem it is for still images.…”

“…For example, visible information is better for establishing a discriminative face model, while thermal IR images are not affected by illumination variation or face disguise. However, successful image fusion requires a critical and challenging step that the image pairs to be fused have to be correctly co-registered on a pixel-by-pixel basis [7][8][9]. In this paper, we focus on registration of visible and thermal IR face images for fusion-based face recognition.…”

WELD: Weighted Low-rank Decomposition for Robust Grayscale-Thermal Foreground Detection

“…Given the potentials of grayscale-thermal data, however, the related research is limited by the lack of a comprehensive video benchmark. On the one hand, the existing grayscale-thermal dataset, like OSU Color-Thermal [16] and LITIV [17,18], contain small number of videos with less challenging and induce significant bias. In addition, these datasets are used for different computer vision tasks, such as registration, fusion, and tracking, and thus are not suitable for addressing different challenges in moving object detection.…”

“…For example, OSU Color-Thermal [16] IM, IS, DS Torabi [17] IM, LI, IS Torabi [18] IM, IS dataset [16] contains six thermal/color video sequence pairs recorded from two different locations with only people moving. Other two grayscale-thermal datasets are collected by Torabi et al [17,18]. The detailed cha Most of them however suffer from their limited size, low diversity and high bias.…”