具有中心窝透镜组的视觉识别系统研究
|
摘要
具有中心窝透镜组的视觉识别系统是现代科技与仿生学相结合的产物,该系统的核心元件中心窝透镜组是模仿人眼的生理结构设计的,它具有两种视野区域,中心视野分辨率高但视觉范围较小,与之相反,周边视野的分辨率较低但视角范围大。系统采用能动视觉的工作原理,用大视角的周边视野检出移动物体,然后控制摄像机,将其中心视野区域对准移动物体,用高分辨率的中心视野对移动物体的形状和动向进行识别。这种设计可削减需处理的信息量,使实时地高速识别成为可能。
论文的研究是以具有中心窝透镜组的视觉识别系统在安全保障领域中地应用为背景展开的,为使该系统能不遗漏地识别出犯罪行为,且实时地对嫌疑人的行为进行跟踪监测并拍下高分辨率图像作为犯罪证据,论文采用了模式识别算法,使用一类支持向量机(OCSVM)设计了线性分类器,用人物的形状特征作为模式识别特征量,由于中心窝透镜组是具有特殊放大率的光学元件,以往研究中所采用的筛网特征和边缘特征在本研究中并不适用,因此,本文采用了以人像对透镜组的入射角为参数来表征人物形状识别特征量的方法,对人物的形状信息进行提取。
论文通过具体的系统测试实验,验证了本文提出的采用人像对透镜系的入射角为人物形状识别特征量的可行性,对具体实验数据的分析表明,采用本文中的方法,中心窝透镜组视觉识别系统能以较高的识别率进行人物行为识别,并对人物进行实时的锁定追踪,该系统能较可靠地工作于安全保障领域。论文最后还针对研究中的不足之处给出了自己的改进建议。
The imagery processing technology was born in the 1960s, it was widely used in each domain of social life now. In recent years, the development trends of this technology is the research of time-varying image processing. The so-called time-varying image processing is a technology aims on capturing the moving object whose position is changing along with time variation and identify the shape and trends of the moving object in real time. Recently, the research to apply time-varying imagery processing technology in robot vision and safe guard field is in the ascendant.
The foveal lens visual recognition system is a tipical application of time-varying imagery processing technology in safe guard field. Foveal lens visual recognition system has a ken about 120 degree, and can be divided into peripheral regards which is in low resolution but have a wide visual angle, and the center regards which is in high resolution but the visual angle is much smaller. The principle of this system is dynamic visual, it use peripheral regards to detect moving object and move the visual line of camera, using center regards capture the detected moving object and recognise it’s shape and trends.This is an imitation of human eye’s visual function. Human eye can select and accept the information actively. Usualy the movements of visual line perform as saccade which is quick and controll both eyes to do simultaneous movements in the same direction, it’s a jump-style scan that suit to see some irregular drifts and movements. Only when pursuing the moving object in field of vision the smooth pursuit movement will occur, and it’s a continuous movement. The movements of visual line will switch between these two functions when necessary. Thus , this dynamic visual could keep a large ken and select necessary image information at the same time so that it could put down amount of information and make rapid visual recognition possible.
The research in this article aims at using foveal lens visual recognition system in safe guard domain. As an accessorial device of safe guard, we hope the surveillance cameras should possess the characteristics as follows: (1st) it may capture the image of criminal exhaustively. (2nd) to have a large ken and an excellent performance in pursuiting, and it’s better to be able to recognise the criminal behavior and take a high resolution photo of the criminal as evidence. (3rd) easy to be managed and conductive to improving the working efficiency. In order to make the foveal lens visual recognition system to be an intelligent surveillance system like that, we do the research work as follow:
1st, For the Fovea lens Recongnization System is designed to be able to lock in the intruder and pursue it real-time, velocity parameter of the intruder is needed. So we addopt Optical Flow Methord which can detect the moving field of a moving object in the research. The theory of Optical Flow Methord evaluate detailed in this paper, for considering the reliability of optical flow in using, we solve this problem as a hypothesis in Statistical Theory.
2nd, We use pattern recognition method to carry on the criminality recognition: Pattern recognition is the act of taking in raw data and taking an action based on the category of the data. Pattern recognition aims to classify data based either on a priori-knowledge or on statistical information extracted from the patterns. The patterns to be classified are usually groups of measurements or observations, defining points in an appropriate multidimensional space.This research regards walking and standing as the normal behavior and gesture, adopting pattern recognition method based on statistic, we design a linear classifier by using one-class support vector machine. Some violence movement and abnormal behaviour which can obtain in daily life are used as the learning pattern, such as strokes, kicks tramples, picks up or throws down the thing and so on.
3rd ,We use shape characteristic of human as behavior recognition feature quantity. For the magnification of foveal lens is special, the mesh feature and edge feature which used in the research before don’t fit for fovea lens.So we suggest using incidence angle as parameter to characterize human shape. Experiment prove that the recognition accuracy is fairly high by using incidence angle parameter as behavior recognition feature quantity. And we can see that Fovea Lens Recognition System can be used dependable in Safe Guard area by analyzing specific data obtained in the experiment.
论文的研究是以具有中心窝透镜组的视觉识别系统在安全保障领域中地应用为背景展开的,为使该系统能不遗漏地识别出犯罪行为,且实时地对嫌疑人的行为进行跟踪监测并拍下高分辨率图像作为犯罪证据,论文采用了模式识别算法,使用一类支持向量机(OCSVM)设计了线性分类器,用人物的形状特征作为模式识别特征量,由于中心窝透镜组是具有特殊放大率的光学元件,以往研究中所采用的筛网特征和边缘特征在本研究中并不适用,因此,本文采用了以人像对透镜组的入射角为参数来表征人物形状识别特征量的方法,对人物的形状信息进行提取。
论文通过具体的系统测试实验,验证了本文提出的采用人像对透镜系的入射角为人物形状识别特征量的可行性,对具体实验数据的分析表明,采用本文中的方法,中心窝透镜组视觉识别系统能以较高的识别率进行人物行为识别,并对人物进行实时的锁定追踪,该系统能较可靠地工作于安全保障领域。论文最后还针对研究中的不足之处给出了自己的改进建议。
The imagery processing technology was born in the 1960s, it was widely used in each domain of social life now. In recent years, the development trends of this technology is the research of time-varying image processing. The so-called time-varying image processing is a technology aims on capturing the moving object whose position is changing along with time variation and identify the shape and trends of the moving object in real time. Recently, the research to apply time-varying imagery processing technology in robot vision and safe guard field is in the ascendant.
The foveal lens visual recognition system is a tipical application of time-varying imagery processing technology in safe guard field. Foveal lens visual recognition system has a ken about 120 degree, and can be divided into peripheral regards which is in low resolution but have a wide visual angle, and the center regards which is in high resolution but the visual angle is much smaller. The principle of this system is dynamic visual, it use peripheral regards to detect moving object and move the visual line of camera, using center regards capture the detected moving object and recognise it’s shape and trends.This is an imitation of human eye’s visual function. Human eye can select and accept the information actively. Usualy the movements of visual line perform as saccade which is quick and controll both eyes to do simultaneous movements in the same direction, it’s a jump-style scan that suit to see some irregular drifts and movements. Only when pursuing the moving object in field of vision the smooth pursuit movement will occur, and it’s a continuous movement. The movements of visual line will switch between these two functions when necessary. Thus , this dynamic visual could keep a large ken and select necessary image information at the same time so that it could put down amount of information and make rapid visual recognition possible.
The research in this article aims at using foveal lens visual recognition system in safe guard domain. As an accessorial device of safe guard, we hope the surveillance cameras should possess the characteristics as follows: (1st) it may capture the image of criminal exhaustively. (2nd) to have a large ken and an excellent performance in pursuiting, and it’s better to be able to recognise the criminal behavior and take a high resolution photo of the criminal as evidence. (3rd) easy to be managed and conductive to improving the working efficiency. In order to make the foveal lens visual recognition system to be an intelligent surveillance system like that, we do the research work as follow:
1st, For the Fovea lens Recongnization System is designed to be able to lock in the intruder and pursue it real-time, velocity parameter of the intruder is needed. So we addopt Optical Flow Methord which can detect the moving field of a moving object in the research. The theory of Optical Flow Methord evaluate detailed in this paper, for considering the reliability of optical flow in using, we solve this problem as a hypothesis in Statistical Theory.
2nd, We use pattern recognition method to carry on the criminality recognition: Pattern recognition is the act of taking in raw data and taking an action based on the category of the data. Pattern recognition aims to classify data based either on a priori-knowledge or on statistical information extracted from the patterns. The patterns to be classified are usually groups of measurements or observations, defining points in an appropriate multidimensional space.This research regards walking and standing as the normal behavior and gesture, adopting pattern recognition method based on statistic, we design a linear classifier by using one-class support vector machine. Some violence movement and abnormal behaviour which can obtain in daily life are used as the learning pattern, such as strokes, kicks tramples, picks up or throws down the thing and so on.
3rd ,We use shape characteristic of human as behavior recognition feature quantity. For the magnification of foveal lens is special, the mesh feature and edge feature which used in the research before don’t fit for fovea lens.So we suggest using incidence angle as parameter to characterize human shape. Experiment prove that the recognition accuracy is fairly high by using incidence angle parameter as behavior recognition feature quantity. And we can see that Fovea Lens Recognition System can be used dependable in Safe Guard area by analyzing specific data obtained in the experiment.
引文
[1]増田功,セキュリティサービスにおける認識?理解技術への期待,電子情報通信学会技術研究報告，1997,PRMU 97-54:49-55.
[2]越後富夫,岩井儀雄,森島繁生,鷲見和彦,井岡幹博,八木康史.人图像処理[M].オーム社,2007
[3]大城尚紀,丸典明,西川敦,宮崎文夫.網膜と皮質間の写像関係に基づく両眼追跡[C].システム制御情報学会論文誌,1997,No.10,No.6:287-296.
[4]小渡悟,星野聖.対数極座標変換された時系列图像中の注視物体追跡[R].電子情報通信学会技術研究報告，1999,PRMU 99-161:79-83.
[5]近藤真悟,佐藤幸男.人間の視解像度を模擬した視覚システムの構築[R].電子情報通信学会技術研究報告，1998,PRMU 98-17:31-36.
[6]大和淳司,大谷淳,石井健一郎.隠れマルコフモデルを用いた動图像からの人物の行動認識[C].電子情報通信学会論文誌,1993,D-ⅡVol.J76-D-Ⅱ:2556-2563.
[7]高橋勝彦,関進,小島浩,岡隆一.ジェスチャー動图像のスポッテ?螗罢J識[C] .電子情報通信学会論文誌, 1994,D-ⅡVol.J77-D-Ⅱ:1552-1561.
[8]福島邦彦.視覚の生理とバイオニクス.コロナ社,1978
[9] M.A.フィシュラー,O.ファーシャイン.人と機械の知能[M].トッパン,1989
[10]大頭仁,行田尚義.視覚と图像[M].森北出版,1994
[11]原島博.3次元图像と人間の科学.オーム社,2000
[12]西本義文.超音波モータを応用した旋回台システム[J].超音THECNO,2001,No.7:35-37,
[13]浅田稔,谷内田正彦,辻三郎.運動物体の検出と追跡[C].電子情報通信学会論文誌,Vol.J62-D 1979,No.6:395-402
[14]高橋政雄,北村忠明,小林芳樹.空間微分および差分処理を用いた車両抽出法[C].電子情報通信学会論文誌,D-ⅡVol.J80-D-Ⅱ1997,No.11:2976-2985
[15]上田浩次,堀場勇夫，池谷和夫，小野寺浩.图像処理を用いた駐車車両検出アルゴリズム[C].電子情報通信学会論文誌,D-ⅡVol.J74-D-Ⅱ1991,No.10:1379-1389
[16]和田俊和,浮田宗伯,松山隆司.視点固定型パンチルトズームカメラとその応用[C],電子情報通信学会論文誌,D-ⅡVol.J81-D-Ⅱ1998,No.6:1182-1193
[17]松山隆司,和田俊和,物部祐亮.視点固定型パン?チルト?ズームカメラを用いた実時間対象検出?追跡[C].情報処理学会論文誌,Vol.40,1999,No.8
[18] Berthold K.P. Horn and Brian G. Schunck, Determining Optical Flow, ARTIFICIAL INTELLIGENCE, No.17, pp.185-204, 1981
[19] Bruce D. Lucas and Takeo Kanade, An Iterative Image Registration Technique with an Application to Stereo Vision, Proceedings of Imaging Understanding Workshop, pp.121-130,1981
[20]马颂德.计算机视觉[M].科学出版社.1998
[21]杨文杰,胡明昊,杨静宇.一种基于光流的障碍物估计算法[J].计算机工程与应用,2006,05:80-81
[22]胡以静,李政访,胡跃明.基于光流的运动分析理论及应用[J].计算机测量与控制,2007,15(2):219-221
[23] Iketani A , Kuno Y, et al . Real time surveillance system detecting persons in complex scenery [A] . Proc. of Image Analysis and Processing [C] . 1999.
[24] Barron J , Fleet D , Beauchemin S. Performance of optical flow techniques [J ] . InternationalJournal of Computer Vision , 1994 ,12 (1) : 42 - 77.
[25]卢宗庆.运动图象分析中的光流计算方法研究[D].西安:西安电子科技大学,2007
[26]井上博充,稲葉雅幸,森武俊,立川哲也.局所相関演算に基づく実時間ビジョンシステムの開発[C].日本ロボット学会誌,Vol.13,1995,No.1:134-140
[27]出口光一郎,鏡慎吾,嵯峨智,本谷秀堅.能動カメラによる運動物体追跡と実時間形状復元システム[C].計測自動制御学会論文集,Vol35,1999,No.5:675-683
[28] J.L. Barron and D.J. Fleet and S.S. Beauchemin. Performance of Optical Flow Techniques[J]. International Journal of Computer Vision, 1994,12(1):43-77,
[29]太田直哉.信頼性指標を持つ移動ベクトルの検出[C].コンピュータビジョン‘90情報処理学会シンポジウム論文集,Vol.90,1990,No.2:21-30
[30]李衍达,郑大钟.模式识别[M].北京.清华大学出版社.1999
[31]龚声容,刘纯平,王强等.数字图像处理与分析[M].清华大学出版社.2006
[32]王珏,陈汝铃,周傲英,周志华.机器学习及其应用[M].清华大学出版社.2006
[33]中村克行,趙卉菁?柴崎亮介.マルチセンサを用いた群集の行動認識.東京大学空間情報科学研究センター
[34]布施孝志,清水英範,前田亮.高度撮影時系列图像を用いた車両動態認識手法の構築[C].土木学会論文集,2003,No.735,6-59:157-170
[35] Rong-En Fan?Pai-Hsuen Chen?Chih-Jen Lin,Working Set Selection Using Second Order Information for Trainig Support Vector Machines,Journal of Machine Learning Research,No.6,pp.1889-1918,2005
[36] Nello Cristianini,John Shawe-Taylor. An Introduction to Support Victor Machines and Kernel-based Learning Methods .1992
[37]张学工.关于统计学习理论与支持向量机[J].自动化学报,2000,26(1):32-42
[38]高島慧,香田正人.周辺密度を用いたOne-Class SVMの提案[D],システム情報工学研究科修士論文,筑波大学大学院
[39] C.M.ビショップ,パターン認識と機械学習上?下[M].シュプリンガージャパン株式会社,2008
[40] Larry M. Manevitz, Malik Yousef. One-Class SVMs for Document Classification[J]. Journal of Machine Learning Research 2 (2001) 139-154
[41]黄谦,王震,韦韬,陈昱.基于One-Class SVM的实时入侵检测系统.计算机工程. 2006(8):127-129
[42] Yunqiang Chen, Xiang Zhou, and Thomas S. Huang. ONE-CLASS SVM FOR LEARNING IN IMAGE RETRIEVAL[C]. In Proc. IEEE Int'l Conf. on Image Processing 2001, Thessaloniki, Greece
[43] G. R?tsch, B. Sch?lkopf, S. Mika, and K. R. Müller.“SVM and boosting: One class”, Technical Report 119, GMD FIRST, Berlin, November 2000.
[44] CHEN Y, ZHOU X S, HUANG T S.One- class SVM for learning in image retrieval[C] / /IEEE Int’l Conf on Image Proc( ICIP’2001) ,Thessaloniki, Greece.
[45]钟清流,蔡自兴.基于一类支持向量机的传感器故障诊断[J].计算机工程与应用,2006,19:1-3、7
[46]韩鹏,杨晓元,唐玉华.基于一类支持向量机的隐秘图象盲检测算法[J].计算机工程与应用,2006,35:45-47
[47] Tax D M J , Duin R P W. Support vector domain description[J ] . Pattern Recognition Letters , 1999 , 20 (11213) : 1191-1199.
[48] Sch¨olkopf B , Williamson R C , Smola A J , et al.Support vector method for novelty detection [ J ] .Advances in Neural Information Processing Systems,1999 , 12 : 5262532.
[49]徐磊,赵光宙.基于模糊一类支持向量机的核聚类算法[J].控制与决策,2008,9:1030-1034
[50] V.Vapnik.The Nature of Statistical Learning Theory[M].Springer Verlag,1995
[51] SCHOLKOPF B, SMOLA A,W ILL IAMSON R, et al. New support vector algorithms [ J ]. Neural Computation, 2000, 12 (5 ) : 1207 -1245.
[52]唐玉华,杨晓元,张敏情,韩鹏.多超球面OC-SVM算法在隐秘图像检测中的应用[J].计算机应用,2006,12:2887-2889
[53] LIBSVM,Chih-Chung Chang.Chih-Jen Lin,http://www.csie.ntu.edu.tw/~cjlin/libsvm/
[2]越後富夫,岩井儀雄,森島繁生,鷲見和彦,井岡幹博,八木康史.人图像処理[M].オーム社,2007
[3]大城尚紀,丸典明,西川敦,宮崎文夫.網膜と皮質間の写像関係に基づく両眼追跡[C].システム制御情報学会論文誌,1997,No.10,No.6:287-296.
[4]小渡悟,星野聖.対数極座標変換された時系列图像中の注視物体追跡[R].電子情報通信学会技術研究報告，1999,PRMU 99-161:79-83.
[5]近藤真悟,佐藤幸男.人間の視解像度を模擬した視覚システムの構築[R].電子情報通信学会技術研究報告，1998,PRMU 98-17:31-36.
[6]大和淳司,大谷淳,石井健一郎.隠れマルコフモデルを用いた動图像からの人物の行動認識[C].電子情報通信学会論文誌,1993,D-ⅡVol.J76-D-Ⅱ:2556-2563.
[7]高橋勝彦,関進,小島浩,岡隆一.ジェスチャー動图像のスポッテ?螗罢J識[C] .電子情報通信学会論文誌, 1994,D-ⅡVol.J77-D-Ⅱ:1552-1561.
[8]福島邦彦.視覚の生理とバイオニクス.コロナ社,1978
[9] M.A.フィシュラー,O.ファーシャイン.人と機械の知能[M].トッパン,1989
[10]大頭仁,行田尚義.視覚と图像[M].森北出版,1994
[11]原島博.3次元图像と人間の科学.オーム社,2000
[12]西本義文.超音波モータを応用した旋回台システム[J].超音THECNO,2001,No.7:35-37,
[13]浅田稔,谷内田正彦,辻三郎.運動物体の検出と追跡[C].電子情報通信学会論文誌,Vol.J62-D 1979,No.6:395-402
[14]高橋政雄,北村忠明,小林芳樹.空間微分および差分処理を用いた車両抽出法[C].電子情報通信学会論文誌,D-ⅡVol.J80-D-Ⅱ1997,No.11:2976-2985
[15]上田浩次,堀場勇夫，池谷和夫，小野寺浩.图像処理を用いた駐車車両検出アルゴリズム[C].電子情報通信学会論文誌,D-ⅡVol.J74-D-Ⅱ1991,No.10:1379-1389
[16]和田俊和,浮田宗伯,松山隆司.視点固定型パンチルトズームカメラとその応用[C],電子情報通信学会論文誌,D-ⅡVol.J81-D-Ⅱ1998,No.6:1182-1193
[17]松山隆司,和田俊和,物部祐亮.視点固定型パン?チルト?ズームカメラを用いた実時間対象検出?追跡[C].情報処理学会論文誌,Vol.40,1999,No.8
[18] Berthold K.P. Horn and Brian G. Schunck, Determining Optical Flow, ARTIFICIAL INTELLIGENCE, No.17, pp.185-204, 1981
[19] Bruce D. Lucas and Takeo Kanade, An Iterative Image Registration Technique with an Application to Stereo Vision, Proceedings of Imaging Understanding Workshop, pp.121-130,1981
[20]马颂德.计算机视觉[M].科学出版社.1998
[21]杨文杰,胡明昊,杨静宇.一种基于光流的障碍物估计算法[J].计算机工程与应用,2006,05:80-81
[22]胡以静,李政访,胡跃明.基于光流的运动分析理论及应用[J].计算机测量与控制,2007,15(2):219-221
[23] Iketani A , Kuno Y, et al . Real time surveillance system detecting persons in complex scenery [A] . Proc. of Image Analysis and Processing [C] . 1999.
[24] Barron J , Fleet D , Beauchemin S. Performance of optical flow techniques [J ] . InternationalJournal of Computer Vision , 1994 ,12 (1) : 42 - 77.
[25]卢宗庆.运动图象分析中的光流计算方法研究[D].西安:西安电子科技大学,2007
[26]井上博充,稲葉雅幸,森武俊,立川哲也.局所相関演算に基づく実時間ビジョンシステムの開発[C].日本ロボット学会誌,Vol.13,1995,No.1:134-140
[27]出口光一郎,鏡慎吾,嵯峨智,本谷秀堅.能動カメラによる運動物体追跡と実時間形状復元システム[C].計測自動制御学会論文集,Vol35,1999,No.5:675-683
[28] J.L. Barron and D.J. Fleet and S.S. Beauchemin. Performance of Optical Flow Techniques[J]. International Journal of Computer Vision, 1994,12(1):43-77,
[29]太田直哉.信頼性指標を持つ移動ベクトルの検出[C].コンピュータビジョン‘90情報処理学会シンポジウム論文集,Vol.90,1990,No.2:21-30
[30]李衍达,郑大钟.模式识别[M].北京.清华大学出版社.1999
[31]龚声容,刘纯平,王强等.数字图像处理与分析[M].清华大学出版社.2006
[32]王珏,陈汝铃,周傲英,周志华.机器学习及其应用[M].清华大学出版社.2006
[33]中村克行,趙卉菁?柴崎亮介.マルチセンサを用いた群集の行動認識.東京大学空間情報科学研究センター
[34]布施孝志,清水英範,前田亮.高度撮影時系列图像を用いた車両動態認識手法の構築[C].土木学会論文集,2003,No.735,6-59:157-170
[35] Rong-En Fan?Pai-Hsuen Chen?Chih-Jen Lin,Working Set Selection Using Second Order Information for Trainig Support Vector Machines,Journal of Machine Learning Research,No.6,pp.1889-1918,2005
[36] Nello Cristianini,John Shawe-Taylor. An Introduction to Support Victor Machines and Kernel-based Learning Methods .1992
[37]张学工.关于统计学习理论与支持向量机[J].自动化学报,2000,26(1):32-42
[38]高島慧,香田正人.周辺密度を用いたOne-Class SVMの提案[D],システム情報工学研究科修士論文,筑波大学大学院
[39] C.M.ビショップ,パターン認識と機械学習上?下[M].シュプリンガージャパン株式会社,2008
[40] Larry M. Manevitz, Malik Yousef. One-Class SVMs for Document Classification[J]. Journal of Machine Learning Research 2 (2001) 139-154
[41]黄谦,王震,韦韬,陈昱.基于One-Class SVM的实时入侵检测系统.计算机工程. 2006(8):127-129
[42] Yunqiang Chen, Xiang Zhou, and Thomas S. Huang. ONE-CLASS SVM FOR LEARNING IN IMAGE RETRIEVAL[C]. In Proc. IEEE Int'l Conf. on Image Processing 2001, Thessaloniki, Greece
[43] G. R?tsch, B. Sch?lkopf, S. Mika, and K. R. Müller.“SVM and boosting: One class”, Technical Report 119, GMD FIRST, Berlin, November 2000.
[44] CHEN Y, ZHOU X S, HUANG T S.One- class SVM for learning in image retrieval[C] / /IEEE Int’l Conf on Image Proc( ICIP’2001) ,Thessaloniki, Greece.
[45]钟清流,蔡自兴.基于一类支持向量机的传感器故障诊断[J].计算机工程与应用,2006,19:1-3、7
[46]韩鹏,杨晓元,唐玉华.基于一类支持向量机的隐秘图象盲检测算法[J].计算机工程与应用,2006,35:45-47
[47] Tax D M J , Duin R P W. Support vector domain description[J ] . Pattern Recognition Letters , 1999 , 20 (11213) : 1191-1199.
[48] Sch¨olkopf B , Williamson R C , Smola A J , et al.Support vector method for novelty detection [ J ] .Advances in Neural Information Processing Systems,1999 , 12 : 5262532.
[49]徐磊,赵光宙.基于模糊一类支持向量机的核聚类算法[J].控制与决策,2008,9:1030-1034
[50] V.Vapnik.The Nature of Statistical Learning Theory[M].Springer Verlag,1995
[51] SCHOLKOPF B, SMOLA A,W ILL IAMSON R, et al. New support vector algorithms [ J ]. Neural Computation, 2000, 12 (5 ) : 1207 -1245.
[52]唐玉华,杨晓元,张敏情,韩鹏.多超球面OC-SVM算法在隐秘图像检测中的应用[J].计算机应用,2006,12:2887-2889
[53] LIBSVM,Chih-Chung Chang.Chih-Jen Lin,http://www.csie.ntu.edu.tw/~cjlin/libsvm/