Transformation from Frontal Face Detector to Multi-view Face Detector by Mirror Reversal

カテゴリ: 部門大会

論文No: SS3-8

グループ名: 【C】平成29年電気学会電子・情報・システム部門大会講演論文集

発行日: 2017/09/06

タイトル(英語): Transformation from Frontal Face Detector to Multi-view Face Detector by Mirror Reversal

著者名: 游 梦博(岩手大学),孫 海天(岩手大学),明石 卓也(岩手大学)

著者名(英語): Mengbo You|Haitian Sun|Takuya Akashi

キーワード: 多視点顔検出器|正面顔検出器|鏡映反転鏡映反転|Multi-view Face Detection|Frontal Face Detector|Mirror Reversal

要約(日本語): Frontal face detection has been developed mature enough to be applied to everyday life while it is still difficult to handle various profile face views practically. A common solution is to build different detectors organized by a decision tree and ensure each detector coping with a view. However, collecting training images for each view costs much time and labor. Most profile face detectors cannot perform as good as the frontal face detector for insufficient training data. We proposed to transform a frontal face detector to a multi-view face detector without further data collection or training. The result verified that our method is able to detect human faces in almost all the head poses in the dataset.

PDFファイルサイズ: 428 Kバイト