@techreport{oai:ipsj.ixsq.nii.ac.jp:00175410,
 author = {Chao, Ma and Ngo, Thanh Trung and Hideaki, Uchiyama and Hajime, Nagahara and Atsushi, Shimada and Rin-ichiro, Taniguchi and Chao, Ma and Ngo, Thanh Trung and Hideaki, Uchiyama and Hajime, Nagahara and Atsushi, Shimada and Rin-ichiro, Taniguchi},
 issue = {1},
 month = {Nov},
 note = {An infrared camera is able to capture temperature distribution as an infrared (IR) image. It is a powerful tool in human related applications, such as human face recognition in complex illumination and fever screening in public places relying on facial temperature. Since facial temperature is almost constant, it is easy to find the facial region on an IR image. However, a simple temperature thresholding is not always working for detecting face stably. It is a standard for face detection to use Adaboost with local features such as Haar-like, MB-LPB, and HoG in visible image. However, there are very few research works using these local features in IR domain. In this paper, we propose an AdaBoost based training method to mix these local features for face detection in IR domain. In experiment, we captured a dataset of 20 people including 14 males and 6 females with variations of 10 different distances, 21 poses, and with/without glasses. We showed the proposed mixed features has an advantage over all of the regular local features using leave-one-out cross-validation., An infrared camera is able to capture temperature distribution as an infrared (IR) image. It is a powerful tool in human related applications, such as human face recognition in complex illumination and fever screening in public places relying on facial temperature. Since facial temperature is almost constant, it is easy to find the facial region on an IR image. However, a simple temperature thresholding is not always working for detecting face stably. It is a standard for face detection to use Adaboost with local features such as Haar-like, MB-LPB, and HoG in visible image. However, there are very few research works using these local features in IR domain. In this paper, we propose an AdaBoost based training method to mix these local features for face detection in IR domain. In experiment, we captured a dataset of 20 people including 14 males and 6 females with variations of 10 different distances, 21 poses, and with/without glasses. We showed the proposed mixed features has an advantage over all of the regular local features using leave-one-out cross-validation.},
 title = {Mixed Features for Face Detection in Thermal Image},
 year = {2016}
}