Face Detection in Thermal Image Based on Two-Stage CNNs

• Item type: SIG Technical Reports(1)
• 公開日: 2018-05-03
• タイトル: Face Detection in Thermal Image Based on Two-Stage CNNs
• タイトル / 言語: en
• タイトル: Face Detection in Thermal Image Based on Two-Stage CNNs
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 一般セッション
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Kyushu University
• 著者所属: Kyushu University
• 著者所属: Kyushu University
• 著者所属: Kyushu University
• 著者所属: Osaka University
• 著者所属: Kyushu University
• 著者所属(英): en
• 著者所属(英): Kyushu University
• 著者所属(英): en
• 著者所属(英): Kyushu University
• 著者所属(英): en
• 著者所属(英): Kyushu University
• 著者所属(英): en
• 著者所属(英): Kyushu University
• 著者所属(英): en
• 著者所属(英): Osaka University
• 著者所属(英): en
• 著者所属(英): Kyushu University
• 著者名: Yibo, Guo; Atsushi, Shimada; Hideaki, Uchiyama; Chao, Ma; Hajime, Nagahara; Rin-ichiro, Taniguchi
• 著者名(英): Yibo, Guo; Atsushi, Shimada; Hideaki, Uchiyama; Chao, Ma; Hajime, Nagahara; Rin-ichiro, Taniguchi
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: The IR images generated from an infrared camera show the temperature distribution of object. Because facial temperature is stable and independent of ambient lighting, an IR camera can be used in detecting facial regions in indoor facilities. The face detection algorithm uses Adaboost with local features, such as Haar-like, MB-LBP, and HOG features in the thermal image. However, these features can only be extracted by manual and few of them can be used. In this paper, we proposed a method that detects faces with thermal image by using CNN, which do not need to design the features and maintain a well performance. The face detector we provided has two stage: a CNN for facial region proposal and another CNN for calibration. The calibration network is used to adjust the detetcion window location in order to obtain a better recall. Finally, we compare the performance on thermal image dataset of our method with method that uses different designed features such HOG, and find that the face detector can maintain a similar or even better result, while being less complexity.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: The IR images generated from an infrared camera show the temperature distribution of object. Because facial temperature is stable and independent of ambient lighting, an IR camera can be used in detecting facial regions in indoor facilities. The face detection algorithm uses Adaboost with local features, such as Haar-like, MB-LBP, and HOG features in the thermal image. However, these features can only be extracted by manual and few of them can be used. In this paper, we proposed a method that detects faces with thermal image by using CNN, which do not need to design the features and maintain a well performance. The face detector we provided has two stage: a CNN for facial region proposal and another CNN for calibration. The calibration network is used to adjust the detetcion window location in order to obtain a better recall. Finally, we compare the performance on thermal image dataset of our method with method that uses different designed features such HOG, and find that the face detector can maintain a similar or even better result, while being less complexity.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA11131797
• 書誌情報: 研究報告コンピュータビジョンとイメージメディア（CVIM）; 巻 2018-CVIM-212, 号 2, p. 1-4, 発行日 2018-05-03
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8701
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会