| 講演名 | 2019-07-06 プロファイル顔検出のために拡張方式を適用する可能性を探索する 鄒 敏(岩手大), 游 夢博(西北農林科技大), 明石 卓也(岩手大), |
|---|---|
| PDFダウンロードページ |  PDFダウンロードページへ |
| 抄録(和) | %% v3.1 [2018/04/18] %documentclass[Proof,technicalreport]{ieicej}documentclass[technicalreport]{ieicej}usepackage{graphicx}usepackage[T1]{fontenc}usepackage{lmodern}usepackage{textcomp}usepackage{latexsym}usepackage{amsmath,amssymb}%usepackage[fleqn]{amsmath}%usepackage{amssymb}defIEICEJcls{texttt{ieicej.cls}}defIEICEJver{3.1}newcommand{AmSLaTeX}{% $mathcal A$lower.4exhbox{$!mathcal M!$}$mathcal S$-LaTeX}%newcommand{PS}{{scshape Post-Script}}defBibTeX{{rmfamily Bkern-.05em{scshape ikern-.025em b}kern-.08em Tkern-.1667emlower.7exhbox{E}kern-.125em X}}jtitle{プロファイル顔検出のために拡張方式を適用する可能性を探索する}%jsubtitle{}etitle{Exploring the Possibility of Applying Expansion Scheme for Profile FaceDetection}%esubtitle{}authorlist{% authorentry[zou@scv.cis.iwate-u.ac.jp]{鄒 敏}{Min ZOU}{Iwate} authorentry[ymb@nwafu.edu.cn]{游 梦博}{Mengbo YOU}{china} authorentry[akashi@iwate-u.ac.jp]{明石 卓也}{Takuya AKASHI}{IwateAkashi}}affiliate[Iwate]{岩手大学理工学研究科hskip1zw〒020--8551 岩手県盛岡市上田4--3--5}{Graduate School of Science and Engineering, Iwate Universityhskip1em%, Division of Design and Media Technology4--3--5, Ueda, Morioka, Iwate, 020--8551 Japan}affiliate[china]{西北農林科技大学hskip1zw〒712100 中国陝西省咸陽市楊陵区}{Department of Computer Science, College of Information Engineering, Northwest A&F University\No.22, Rd.Xinong, Yangling, Shaanxi, 712100 China}affiliate[IwateAkashi]{岩手大学理工学部hskip1zw〒020--8551 岩手県盛岡市上田4--3--5}{%Academic Group of System Sciences & Technology, Faculty of Science and Engineering, Iwate Universityhskip1em%, %Department of Systems Innovation %Engineering, Computer, Intelligence and Media Technology, %Graduate School of Science and Engineering, Division of Design and Media Technologyhskip1em4--3--5, Ueda, Morioka, Iwate, 020--8551 Japan}%MailAddress{$dagger$hanako@denshi.ac.jp, % $daggerdagger${taro,jiro}@jouhou.co.jp}begin{document}begin{jabstract}教育の質を向上させるには，顔検出および検知技術が必要である．先生は学生の表情や頭の傾きなどを把握できれば，授業のスピードや内容を変えることができる．深層学習と他の方法を用いて，センシング技術における人間の正面顔検出は大きな成果が出ており，精度も非常に高い．しかしながら，プロファイル顔検出に関する研究はそれほど多くはない．本稿では，最初のステップとしてプロファイル顔検出のための方法を探索する．既存の正面顔検出分類器を利用する拡張方法を提案した．実験結果が我拡張方法が横顔の検出に有効であることを示した．end{jabstract}begin{jkeyword}多視点顔検出，正面顔検出，拡張方法，コンピュータビジョンend{jkeyword}begin{eabstract}Face detection and sensing technology are necessary to improve thequality of education. Because the teacher can change the speed and content bygrasping the facial expressions and head tilt. Using Deep Learning and othermethods, frontal face detection of human face in sensing technology achievedgreat progress, and the accuracy is very high. However, the researches towardsprofile face detection are less. In this paper, we will work on exploring themethod for the profile face detection as the first step. We proposed the methodof expansion scheme, which taking advantage of the existing frontal facedetection classifier. The experimental results show that expansion scheme iseffective for implementing profile face detection. end{eabstract}begin{ekeyword}Multi-view Face Detection, Frontal Face Detection, Expansion Scheme, Computer Visionend{ekeyword}maketitlesection{Introduction} In recent years, face recognition has attracted much attention. The researches of face detection has rapidly expanded by not only engineers but also neuroscientists, since it has many potential applications in computer vision communication and automatic access control system. As a first step of face recognition, face detection is very important. However, face detection is not straightforward because it has lots of variations of image appearance, such as pose variation (frontal, non-frontal), occlusion, image orientation, illuminating condition and facial expression. So many researches about frontal face detection had achieved good results and the high accuracy. The method they proposed can successfully detect frontal faces in a wide variety of images~cite{osuna1997training}~cite{yang2000snow}~cite{rowley1999neural}~cite{schneiderman2000statistical}~cite{sung1994example}~cite{viola2001rapid}. While profile face detection methods are always not reliable which inspires some methods~cite{li2002statistical}~cite{schneiderman2000statistical} to specifically address the profile problem. Profile face detection is difficult as profile face consist of different roll, pitch an yaw angles of the 3D space. So it is difficult to develop a detection system to cover all the combinations of roll, pitch and yaw angles. The method named boosted cascade classification is proposed in~cite{viola2001rapid}, which is also called Viola-Jones(VJ)framework to achieve fast processing of frontal face detector. There is another classification called SURF cascade framework, which was proposed in 2011~cite{li2011face}. It is derived from VJ framework and adopt not Haar-like features but multi-dimensional SURF features~cite{bay2008speeded} to describe local patches. ~cite{jones2003fast}built 12 different detectors to handle in-plane rotation classes of frontal face and a right profile detector to handle out-of-plane rotation from about 3/4 view to full profile. Instead of running all detectors on all regions, a decision tree is trained to choose the appropriate detector for that viewpoint. However, the size of the training sets is as small as 2424 and the number of training images for profile face is about 1/3 of that for frontal face which results in the limited performance. VJ framework provides a fast solution with boosting cascade classifier using the Haar features for the frontal face detection. This paper proposed to use the slipping window optimized expansion scheme to transform the frontal face detector to be profile face detector by making use of the over-representation characteristic of the frontal face detector trained with Haar-like features. This paper is structured as follows: Section 2 introduces the related research in face detection and the contributions of the researches. Section 3 introduces our method for exploring the possibility of applying expansion scheme for profile face. Section 4 introduces the experimental results. The conclusion and future work is given in Section 5. begin{figure*}[tb] vspace{-20mm} begin{center} includegraphics[width=1.28textwidth,natwidth=800,natheight=182]{./figs/1.png} end{center} caption{Two cases of expanding profile faces to make new target images; (a),(d)ROI contains asymmetrical face, which have different size about the left and right sides; (b)is the result that expanded from (a); (d)is the result that expanded from (c).} label{fig1} end{figure*} section{Related Works} The research in 2000s by Viola and Jones~cite{viola2001rapid}~cite{viola2001robust} had great impact in the field of face detection. There are three main contribution of their researches. First, a new image representation which is called ''Integral Image'', allows the features used by the detector for quickly and efficiently computing the sum of values in a rectangle subset of a grid. They proposed the method called Haar feature whiched is very efficient to compute due to the integral image technique, and provide good performance for building frontal face detector. Second, they proposed the learning algorithm, based on AdaBoost. The AdaBoost (Adaptive Boosting) algorithm is generally considered as the first step towards more practical boosting algorithm~cite{freund1997decision}. There research selects a small number of critical visual features and yields extremely efficient classifiers. Third, they proposed the attentional cascade structure. In the research of~cite{viola2001rapid}, a degenerate decision tree, which is called a ''cascade'' was proposed. It allows background regions of the image to be quickly discarded while taking attention to more computation on promising object-like regions. Another method called SURF cascade(Speeded Up Robust Features~cite{bay2008speeded}) is also used in face detection~cite{li2011face}. Since Haar features have limited representation capacity, that it has difficulty to deal with variations due to pose and illumination. In order to improve the detection accuracy, some researches extended Haar features to Haar-like features, joint Haar features, sparse features~cite{huang2006learning}, polygon features~cite{pham2010fast}, etc. Actually, Haar-like features improved the accuracy, on the other hand, they showed slower detection than Haar feature. SURF cascade was presented in ~cite{li2013learning}for fast and accurate object detectors. This research compared SURF cascade detector with the other algorithms on detection accuracy and speed. The experiments of this research showed the result that although SURF feature is smaller than that of VJ framework in model-size due to short cascade but also achieve the results on par with state-of-art detectors. This paper explores the potential of existing frontal face detector of Haar-like features cascade training with VJ framework. A novel idea is proposed to verify that the existing frontal face detector can be used to detect profile faces. section{Method} subsection{Expansion Scheme} There are a number of techniques for quickly and accurately detecting frontal faces, which is symmetrical horizontally. The actual situation is that faces are always rotated while being captured into an image. This paper focuses on exploring of the detection of asymmetrical face, which have different width between the left side and the right side. Firstly, we expanded the narrower half of the face manually and make the profile face look like a frontal face. The existing Haar-like feature cascade detector is applied to test whether the modified face can be detected. Various profile views of human face are tested to draw the reliable conclusion. However, when we input a similar frontal face which had been expanded, the cascade detector can still recognize it as a frontal face, see Figure 1. The experimental result showed that the profile face can be detected by the detection technical of frontal face. The contribution of this paper is that a novel expansion scheme is proposed to detect the profile face using a frontal face detector. We call this method ''Expansion scheme". subsection{Sliding ROI} The sliding window method plays an integral role in object detection, as it allow us to localize exactly ''where''in an image an object resides. The procedures of sliding window method can be illustrated as follows: begin{itemize} item Firstly, select the ROI (region of interest), and let it move in the target image from the left to right, from top to bottom. item Secondly, a cascade detector is used to make sure if there are frontal faces from each ROI. If the detector reports a "positive", the coordinates will be recorded as the result. And it means there are faces found. item Thirdly, if there are no face found after checking sliding ROI, the searching will stop and report that there are no face. It will slide to another ROI until all the possible regions are checked. end{itemize} subsection{Image Feature} As mentioned in 3.1, we used the expansion scheme to expand the asymmetrical face, then using the sliding window method to search in the target image. Then Haar-like feature cascade is used to make sure if there are frontal faces from each ROI. Haar-like features are digital image features used in object recognition. They owe their name to their intuitive similarity with Haar wavelets and were used in the first real-time face detector~cite{viola2001rapid}. Haar-like features can efficiently reduce/increase the in-class/out-of-class variability, thus making the classification easier~cite{lienhart2002extended}. The Haar-like features can describe the ratio between the dark and bright areas within a kernel. One typical example is that the eye region on the human face is darker than the cheek region, and the method of Haar-like feature can efficiently catch that characteristic effectively~cite{chen2008hand}. begin{figure*}[tb] vspace{-20mm} begin{center} includegraphics[width=1.28textwidth,natwidth=800,natheight=551]{figs/2.png} end{center} caption{(a)and(c)are the result images that not being expanded; (b)and(d)are the result images that used expansion scheme} label{fig3} end{figure*} section{Experimental results} To verify whether the expansion scheme would work, the experiment is designed as follows: Firstly, various profile views of human face images are selected from the Pointing'04 face database. Secondly, these images are manually modified according the expansion scheme. Thirdly, both the original images and the modified images are tested using the frontal face detector. In most cases, the frontal face detector cannot find any faces in the original images since most images contains a profile face. However, the frontal face detector was able to find faces in the modified images, which means the proposed expansion scheme worked. In other words, the proposed expansion scheme has made a profile face detected by a frontal face detector. Figure 2 displays the asymmetrical profile face detected by Haar-like features on database Pointing'04 ~cite{gourier2004estimating}. Without the expansion scheme, the profile face couldn't detected by the Haar-like feature method. After expansion, our method correctly detects the best match of the profile face in 5 of the images. section{Conclusion} Experimental results proved that it is possible to detect the profile face by the method of frontal face detection using expansion scheme. And expansion scheme is effective in the detection of profile face. This paper developed expansion scheme to transform the frontal face detector to be the profile face detector without any more processes of data collection or training. Haar-like features, which we used for the profile face detector provides a strong performance both on the accuracy and on the process of the speed. In the future, will expand the profile face automatically and try other methods to improve the speed and accuracy of profile face detector, and we will take measures to detect multiple faces using the proposed method. bibliographystyle{ieicetr}bibliography{template}end{document} |
| 抄録(英) | %% v3.1 [2018/04/18] %documentclass[Proof,technicalreport]{ieicej}documentclass[technicalreport]{ieicej}usepackage{graphicx}usepackage[T1]{fontenc}usepackage{lmodern}usepackage{textcomp}usepackage{latexsym}usepackage{amsmath,amssymb}%usepackage[fleqn]{amsmath}%usepackage{amssymb}defIEICEJcls{texttt{ieicej.cls}}defIEICEJver{3.1}newcommand{AmSLaTeX}{% $mathcal A$lower.4exhbox{$!mathcal M!$}$mathcal S$-LaTeX}%newcommand{PS}{{scshape Post-Script}}defBibTeX{{rmfamily Bkern-.05em{scshape ikern-.025em b}kern-.08em Tkern-.1667emlower.7exhbox{E}kern-.125em X}}jtitle{プロファイル顔検出のために拡張方式を適用する可能性を探索する}%jsubtitle{}etitle{Exploring the Possibility of Applying Expansion Scheme for Profile FaceDetection}%esubtitle{}authorlist{% authorentry[zou@scv.cis.iwate-u.ac.jp]{鄒 敏}{Min ZOU}{Iwate} authorentry[ymb@nwafu.edu.cn]{游 梦博}{Mengbo YOU}{china} authorentry[akashi@iwate-u.ac.jp]{明石 卓也}{Takuya AKASHI}{IwateAkashi}}affiliate[Iwate]{岩手大学理工学研究科hskip1zw〒020--8551 岩手県盛岡市上田4--3--5}{Graduate School of Science and Engineering, Iwate Universityhskip1em%, Division of Design and Media Technology4--3--5, Ueda, Morioka, Iwate, 020--8551 Japan}affiliate[china]{西北農林科技大学hskip1zw〒712100 中国陝西省咸陽市楊陵区}{Department of Computer Science, College of Information Engineering, Northwest A&F University\No.22, Rd.Xinong, Yangling, Shaanxi, 712100 China}affiliate[IwateAkashi]{岩手大学理工学部hskip1zw〒020--8551 岩手県盛岡市上田4--3--5}{%Academic Group of System Sciences & Technology, Faculty of Science and Engineering, Iwate Universityhskip1em%, %Department of Systems Innovation %Engineering, Computer, Intelligence and Media Technology, %Graduate School of Science and Engineering, Division of Design and Media Technologyhskip1em4--3--5, Ueda, Morioka, Iwate, 020--8551 Japan}%MailAddress{$dagger$hanako@denshi.ac.jp, % $daggerdagger${taro,jiro}@jouhou.co.jp}begin{document}begin{jabstract}教育の質を向上させるには，顔検出および検知技術が必要である．先生は学生の表情や頭の傾きなどを把握できれば，授業のスピードや内容を変えることができる．深層学習と他の方法を用いて，センシング技術における人間の正面顔検出は大きな成果が出ており，精度も非常に高い．しかしながら，プロファイル顔検出に関する研究はそれほど多くはない．本稿では，最初のステップとしてプロファイル顔検出のための方法を探索する．既存の正面顔検出分類器を利用する拡張方法を提案した．実験結果が我拡張方法が横顔の検出に有効であることを示した．end{jabstract}begin{jkeyword}多視点顔検出，正面顔検出，拡張方法，コンピュータビジョンend{jkeyword}begin{eabstract}Face detection and sensing technology are necessary to improve thequality of education. Because the teacher can change the speed and content bygrasping the facial expressions and head tilt. Using Deep Learning and othermethods, frontal face detection of human face in sensing technology achievedgreat progress, and the accuracy is very high. However, the researches towardsprofile face detection are less. In this paper, we will work on exploring themethod for the profile face detection as the first step. We proposed the methodof expansion scheme, which taking advantage of the existing frontal facedetection classifier. The experimental results show that expansion scheme iseffective for implementing profile face detection. end{eabstract}begin{ekeyword}Multi-view Face Detection, Frontal Face Detection, Expansion Scheme, Computer Visionend{ekeyword}maketitlesection{Introduction} In recent years, face recognition has attracted much attention. The researches of face detection has rapidly expanded by not only engineers but also neuroscientists, since it has many potential applications in computer vision communication and automatic access control system. As a first step of face recognition, face detection is very important. However, face detection is not straightforward because it has lots of variations of image appearance, such as pose variation (frontal, non-frontal), occlusion, image orientation, illuminating condition and facial expression. So many researches about frontal face detection had achieved good results and the high accuracy. The method they proposed can successfully detect frontal faces in a wide variety of images~cite{osuna1997training}~cite{yang2000snow}~cite{rowley1999neural}~cite{schneiderman2000statistical}~cite{sung1994example}~cite{viola2001rapid}. While profile face detection methods are always not reliable which inspires some methods~cite{li2002statistical}~cite{schneiderman2000statistical} to specifically address the profile problem. Profile face detection is difficult as profile face consist of different roll, pitch an yaw angles of the 3D space. So it is difficult to develop a detection system to cover all the combinations of roll, pitch and yaw angles. The method named boosted cascade classification is proposed in~cite{viola2001rapid}, which is also called Viola-Jones(VJ)framework to achieve fast processing of frontal face detector. There is another classification called SURF cascade framework, which was proposed in 2011~cite{li2011face}. It is derived from VJ framework and adopt not Haar-like features but multi-dimensional SURF features~cite{bay2008speeded} to describe local patches. ~cite{jones2003fast}built 12 different detectors to handle in-plane rotation classes of frontal face and a right profile detector to handle out-of-plane rotation from about 3/4 view to full profile. Instead of running all detectors on all regions, a decision tree is trained to choose the appropriate detector for that viewpoint. However, the size of the training sets is as small as 2424 and the number of training images for profile face is about 1/3 of that for frontal face which results in the limited performance. VJ framework provides a fast solution with boosting cascade classifier using the Haar features for the frontal face detection. This paper proposed to use the slipping window optimized expansion scheme to transform the frontal face detector to be profile face detector by making use of the over-representation characteristic of the frontal face detector trained with Haar-like features. This paper is structured as follows: Section 2 introduces the related research in face detection and the contributions of the researches. Section 3 introduces our method for exploring the possibility of applying expansion scheme for profile face. Section 4 introduces the experimental results. The conclusion and future work is given in Section 5. begin{figure*}[tb] vspace{-20mm} begin{center} includegraphics[width=1.28textwidth,natwidth=800,natheight=182]{./figs/1.png} end{center} caption{Two cases of expanding profile faces to make new target images; (a),(d)ROI contains asymmetrical face, which have different size about the left and right sides; (b)is the result that expanded from (a); (d)is the result that expanded from (c).} label{fig1} end{figure*} section{Related Works} The research in 2000s by Viola and Jones~cite{viola2001rapid}~cite{viola2001robust} had great impact in the field of face detection. There are three main contribution of their researches. First, a new image representation which is called ''Integral Image'', allows the features used by the detector for quickly and efficiently computing the sum of values in a rectangle subset of a grid. They proposed the method called Haar feature whiched is very efficient to compute due to the integral image technique, and provide good performance for building frontal face detector. Second, they proposed the learning algorithm, based on AdaBoost. The AdaBoost (Adaptive Boosting) algorithm is generally considered as the first step towards more practical boosting algorithm~cite{freund1997decision}. There research selects a small number of critical visual features and yields extremely efficient classifiers. Third, they proposed the attentional cascade structure. In the research of~cite{viola2001rapid}, a degenerate decision tree, which is called a ''cascade'' was proposed. It allows background regions of the image to be quickly discarded while taking attention to more computation on promising object-like regions. Another method called SURF cascade(Speeded Up Robust Features~cite{bay2008speeded}) is also used in face detection~cite{li2011face}. Since Haar features have limited representation capacity, that it has difficulty to deal with variations due to pose and illumination. In order to improve the detection accuracy, some researches extended Haar features to Haar-like features, joint Haar features, sparse features~cite{huang2006learning}, polygon features~cite{pham2010fast}, etc. Actually, Haar-like features improved the accuracy, on the other hand, they showed slower detection than Haar feature. SURF cascade was presented in ~cite{li2013learning}for fast and accurate object detectors. This research compared SURF cascade detector with the other algorithms on detection accuracy and speed. The experiments of this research showed the result that although SURF feature is smaller than that of VJ framework in model-size due to short cascade but also achieve the results on par with state-of-art detectors. This paper explores the potential of existing frontal face detector of Haar-like features cascade training with VJ framework. A novel idea is proposed to verify that the existing frontal face detector can be used to detect profile faces. section{Method} subsection{Expansion Scheme} There are a number of techniques for quickly and accurately detecting frontal faces, which is symmetrical horizontally. The actual situation is that faces are always rotated while being captured into an image. This paper focuses on exploring of the detection of asymmetrical face, which have different width between the left side and the right side. Firstly, we expanded the narrower half of the face manually and make the profile face look like a frontal face. The existing Haar-like feature cascade detector is applied to test whether the modified face can be detected. Various profile views of human face are tested to draw the reliable conclusion. However, when we input a similar frontal face which had been expanded, the cascade detector can still recognize it as a frontal face, see Figure 1. The experimental result showed that the profile face can be detected by the detection technical of frontal face. The contribution of this paper is that a novel expansion scheme is proposed to detect the profile face using a frontal face detector. We call this method ''Expansion scheme". subsection{Sliding ROI} The sliding window method plays an integral role in object detection, as it allow us to localize exactly ''where''in an image an object resides. The procedures of sliding window method can be illustrated as follows: begin{itemize} item Firstly, select the ROI (region of interest), and let it move in the target image from the left to right, from top to bottom. item Secondly, a cascade detector is used to make sure if there are frontal faces from each ROI. If the detector reports a "positive", the coordinates will be recorded as the result. And it means there are faces found. item Thirdly, if there are no face found after checking sliding ROI, the searching will stop and report that there are no face. It will slide to another ROI until all the possible regions are checked. end{itemize} subsection{Image Feature} As mentioned in 3.1, we used the expansion scheme to expand the asymmetrical face, then using the sliding window method to search in the target image. Then Haar-like feature cascade is used to make sure if there are frontal faces from each ROI. Haar-like features are digital image features used in object recognition. They owe their name to their intuitive similarity with Haar wavelets and were used in the first real-time face detector~cite{viola2001rapid}. Haar-like features can efficiently reduce/increase the in-class/out-of-class variability, thus making the classification easier~cite{lienhart2002extended}. The Haar-like features can describe the ratio between the dark and bright areas within a kernel. One typical example is that the eye region on the human face is darker than the cheek region, and the method of Haar-like feature can efficiently catch that characteristic effectively~cite{chen2008hand}. begin{figure*}[tb] vspace{-20mm} begin{center} includegraphics[width=1.28textwidth,natwidth=800,natheight=551]{figs/2.png} end{center} caption{(a)and(c)are the result images that not being expanded; (b)and(d)are the result images that used expansion scheme} label{fig3} end{figure*} section{Experimental results} To verify whether the expansion scheme would work, the experiment is designed as follows: Firstly, various profile views of human face images are selected from the Pointing'04 face database. Secondly, these images are manually modified according the expansion scheme. Thirdly, both the original images and the modified images are tested using the frontal face detector. In most cases, the frontal face detector cannot find any faces in the original images since most images contains a profile face. However, the frontal face detector was able to find faces in the modified images, which means the proposed expansion scheme worked. In other words, the proposed expansion scheme has made a profile face detected by a frontal face detector. Figure 2 displays the asymmetrical profile face detected by Haar-like features on database Pointing'04 ~cite{gourier2004estimating}. Without the expansion scheme, the profile face couldn't detected by the Haar-like feature method. After expansion, our method correctly detects the best match of the profile face in 5 of the images. section{Conclusion} Experimental results proved that it is possible to detect the profile face by the method of frontal face detection using expansion scheme. And expansion scheme is effective in the detection of profile face. This paper developed expansion scheme to transform the frontal face detector to be the profile face detector without any more processes of data collection or training. Haar-like features, which we used for the profile face detector provides a strong performance both on the accuracy and on the process of the speed. In the future, will expand the profile face automatically and try other methods to improve the speed and accuracy of profile face detector, and we will take measures to detect multiple faces using the proposed method. bibliographystyle{ieicetr}bibliography{template}end{document} |
| キーワード(和) | 多視点顔検出 / 正面顔検出 / 拡張方法 / コンピュータビジョン |
| キーワード(英) | Multi-view Face Detection / Frontal Face Detection / Expansion Scheme / Computer Vision |
| 資料番号 | ET2019-16 |
| 発行日 | 2019-06-29 (ET) |
| 研究会情報 | |
| 研究会 | ET |
|---|---|
| 開催期間 | 2019/7/6(から1日開催) |
| 開催地（和） | 岩手県立大学 |
| 開催地（英） | Iwate Prefectural University |
| テーマ（和） | 学習データ蓄積・分析・視覚化／一般 |
| テーマ（英） | Storage, Analysis, and Visualization of Learning Data, etc. |
| 委員長氏名（和） | 鈴木 栄幸(茨城大) |
| 委員長氏名（英） | Hideyuki Suzuki(Ibaraki Univ.) |
| 副委員長氏名（和） | 鷹岡 亮(山口大) |
| 副委員長氏名（英） | Ryo Takaoka(Yamaguchi Univ.) |
| 幹事氏名（和） | 中山 祐貴(早大) / 舟生 日出男(創価大) |
| 幹事氏名（英） | Hiroki Nakayama(Waseda Univ.) / Hideo Funaoi(Soka Univ.) |
| 幹事補佐氏名（和） | 倉山 めぐみ(函館高専) / 大沼 亮(福島大) |
| 幹事補佐氏名（英） | Megumi Kurayama(National Inst. of Tech., Hakodate College) / Ryo Oonuma(Fukushima Univ.) |
| 講演論文情報詳細 | |
| 申込み研究会 | Technical Committee on Educational Technology |
|---|---|
| 本文の言語 | ENG-JTITLE |
| タイトル（和） | プロファイル顔検出のために拡張方式を適用する可能性を探索する |
| サブタイトル（和） | |
| タイトル（英） | Exploring the Possibility of Applying Expansion Scheme for Profile Face Detection |
| サブタイトル（和） | |
| キーワード(1)（和/英） | 多視点顔検出 / Multi-view Face Detection |
| キーワード(2)（和/英） | 正面顔検出 / Frontal Face Detection |
| キーワード(3)（和/英） | 拡張方法 / Expansion Scheme |
| キーワード(4)（和/英） | コンピュータビジョン / Computer Vision |
| 第 1 著者 氏名（和/英） | 鄒 敏 / Min Zou |
| 第 1 著者 所属（和/英） | 岩手大学(略称：岩手大) Iwate University(略称：Iwate Univ.) |
| 第 2 著者 氏名（和/英） | 游 夢博 / Mengbo You |
| 第 2 著者 所属（和/英） | 西北農林科技大学(略称：西北農林科技大) Northwest A&F University(略称：NWAFU) |
| 第 3 著者 氏名（和/英） | 明石 卓也 / Takuya Akashi |
| 第 3 著者 所属（和/英） | 岩手大学(略称：岩手大) Iwate University(略称：Iwate Univ.) |
| 発表年月日 | 2019-07-06 |
| 資料番号 | ET2019-16 |
| 巻番号（vol） | vol.119 |
| 号番号（no） | ET-105 |
| ページ範囲 | pp.7-10(ET), |
| ページ数 | 4 |
| 発行日 | 2019-06-29 (ET) |