基于特征部位和肤色的不良视频检测的算法研究
摘要
中国网民规模继续呈现持续快速发展的趋势,越来越多的居民认识到互联网的便捷作用。随着上网设备成本的下降和居民收入水平的提高,互联网已经走进千家万户。随着互联网的蓬勃发展,我们可以在网络上获得很多有益的信息。但是同时我们也不可避免地遭遇暴力、色情的信息,受到不良的影响。尤其是对于青少年,他们正处在身体和心理的未成熟期,在这个时候受到色情信息的刺激,会对他们形成自己的人生观世界观价值观造成不利的影响。因此,我们需要对网络上的不良信息进行检测识别和屏蔽。在本文中,我们主要的研究对象是不良视频,然后通过对不良视频的检测,希望能够有效的检测出确属色情视频,以利于将其屏蔽掉。
首先我们针对这个目的对各种学术文献和已有检测系统进行了分析,然后研究更有效的不良视频检测的算法。我们将不良视频的检测系统分为三个步骤：镜头分割,关键帧提取,不良图像的检测。我们用直方图的方法描述帧,通过对相邻帧直方图的差异取阈值,来判断是否出现镜头变换。然后通过局部光流法选取一个镜头内的关键帧,我们只需要判断这些关键帧,就能得到关于此镜头的判断结果。即是说,我们判断不良视频的单位是镜头,如果此镜头的关键帧有色情的嫌疑,那么就需要将其屏蔽。而判断每一帧图像是通过皮肤和特征部位联合的方法,我们用高斯混合模型来描述皮肤像素,然后利用adaboost(自举算法)算法识别特征部位,最后将两者联合起来判断图像的性质。本文最后实验部分,显示了我们提出的不良视频检测的算法在时间复杂度和检测的效果上都有令人满意的效果。
More and more people recognize the conveniences of the network as the scale of net citizen is developing rapidly. And the plenty of families get contact with the network since the decreasement of price of equipment of network and increasement of incoming of people. Nowadays, we can get some beneficial information from the network. However, the influence of the sexual and violent information is abstained impossibly. The stimulation of the sexual information will affect the attitude of the world and value, especially the attitude of the teenagers. Therefore, it is necessary to find the algorithm of the detection and prohibition of the sexual information. In our paper, the main object of our research is sexual video. Based on the detection of the video, we hope to ban it effectively.
Firstly, based on the investigation of all kinds of the temporal methods, an innovative and effective method of porn video detection is proposed. We divide the procedure of porn video detection into three steps:video shot detection, key frame extract, porn image detection. Differential of neighbor frames is used to find the location of the video shot. The shot of video is determined by the changes of the histograms of the neighbor frames. Then, the key frame is extracted by the method of the local optical flow. Whether the shot is porn or not can be convincing judged only by key frame. That also means shot is the unit of porn video detection. The shot needs to be forbidden if the key frame of the shot has the suspicion of porn. The method of the detection of frame combines the skin and feature (nipple). We used the GMM model to describe the skin pixel. Then the adaboost algorithm is to detection the feature (nipple).The frame is determined by the combination of the two above. At the final of our paper, the experiment shows that the time complexity and the result of the detection is satisfied.
首先我们针对这个目的对各种学术文献和已有检测系统进行了分析,然后研究更有效的不良视频检测的算法。我们将不良视频的检测系统分为三个步骤：镜头分割,关键帧提取,不良图像的检测。我们用直方图的方法描述帧,通过对相邻帧直方图的差异取阈值,来判断是否出现镜头变换。然后通过局部光流法选取一个镜头内的关键帧,我们只需要判断这些关键帧,就能得到关于此镜头的判断结果。即是说,我们判断不良视频的单位是镜头,如果此镜头的关键帧有色情的嫌疑,那么就需要将其屏蔽。而判断每一帧图像是通过皮肤和特征部位联合的方法,我们用高斯混合模型来描述皮肤像素,然后利用adaboost(自举算法)算法识别特征部位,最后将两者联合起来判断图像的性质。本文最后实验部分,显示了我们提出的不良视频检测的算法在时间复杂度和检测的效果上都有令人满意的效果。
More and more people recognize the conveniences of the network as the scale of net citizen is developing rapidly. And the plenty of families get contact with the network since the decreasement of price of equipment of network and increasement of incoming of people. Nowadays, we can get some beneficial information from the network. However, the influence of the sexual and violent information is abstained impossibly. The stimulation of the sexual information will affect the attitude of the world and value, especially the attitude of the teenagers. Therefore, it is necessary to find the algorithm of the detection and prohibition of the sexual information. In our paper, the main object of our research is sexual video. Based on the detection of the video, we hope to ban it effectively.
Firstly, based on the investigation of all kinds of the temporal methods, an innovative and effective method of porn video detection is proposed. We divide the procedure of porn video detection into three steps:video shot detection, key frame extract, porn image detection. Differential of neighbor frames is used to find the location of the video shot. The shot of video is determined by the changes of the histograms of the neighbor frames. Then, the key frame is extracted by the method of the local optical flow. Whether the shot is porn or not can be convincing judged only by key frame. That also means shot is the unit of porn video detection. The shot needs to be forbidden if the key frame of the shot has the suspicion of porn. The method of the detection of frame combines the skin and feature (nipple). We used the GMM model to describe the skin pixel. Then the adaboost algorithm is to detection the feature (nipple).The frame is determined by the combination of the two above. At the final of our paper, the experiment shows that the time complexity and the result of the detection is satisfied.
引文
[1]刘典,刘文萍.一种基于直方图的切变镜头自动检测算法[J].北方工业大学学报.2007，,(03)：16-20.
[2]任晓峰,吴巍.基于内容的视频检索算法研究[D].中国优秀博硕士学位论文全文数据库(硕士).2008，,(08).
[3]王伟强,高文,马继涌.一种高速切变镜头边界探测算法[J].计算机科学.2001,(07)：56-59.
[4]智敏,蔡安妮.基于基色调的镜头边界检测方法[J].自动化学报.2007,(06)：655-657.
[5]吕丽,杨树堂,陆松年等.基于光流法的不良视频检测算法[J].计算机工程.2007,(12)：220-221.
[6]刘晓楠.基于内容二次聚类的关键帧提取算法[11].计算机与数字工程.2010,(07)：26-29.
[7]吴静丽,马建斌,沈年凤等.基于视频关键帧的水印嵌入算法研究[J].电脑知识与技术.2010,(22)：6328-6330.
[8]王毅,李弼程.一种基于K-L变换和聚类的视频摘要方法[J].计算机应用研究.2010,(09)：3585-3587.
[9]王一丁.实际网络环境中不良图片的过滤方法[J].通信学报.2009,(S1)：103-106.
[10]赵更寅,邱兆文.基于SVM的不良图像检测[J].科技信息(科学教研).2007,(14)：278
[]1]谢志扬,史萍.网络不良视频信息过滤系统的研究与实现[J].中国传媒大学学报(自然科学版).2009,(07)：67-71
[12]吕丽,杨树堂,陆松年,李建华.基于光流法的不良视频检测算法[J].计算机工程.2007,(12)：220-221
[13]安红心,刘艳民.基于运动特征的不良视频检测算法[J].微计算机应用.2010,(07)：32-37.
[14]Fleck M, Forsyth D A, Bregler C. Find Naked People. In:Proc. of 4th European Conf.on Computer Vision[C].1996.592-602.
[15]V. Vezhnevets, V. Sazonov, A. Andreeva. A survey on pixel-based skin color detection techniques[C]. IN PROC. GRAPHICON-2003.2003.85-92.
[16]Hu W, Wu O, Chen Z, Fu Z, Maybank S. Recognition of pornographic web pages by classifying texts and images[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence.2007,29(6):1019-1034.
[17]DA Forsyth, MM Fleck. Automatic detection of human nudes[J]. International Journal of Computer Vision.1999, (32):63-77.
[18]P. Fuangkhon, T. Tanprasert. Nipple detection for obscene pictures[C]. In Proc. of WSEAS Int. Conf. on Signal, Speech and Image Processing.2005.315-320.
[19]Yue Wang, Jun Li, Heelin Wang, and Zujun Hou. Automatic nipple detection using shape and statistical skin color information[C]. Proceeding in Advances in Multimedia Modeling.2010.644-649.
[20]Cybersitter, http://www.solidoak.com, visited on 2010-10-9.
[21]Thomas M. Chen, Victoria Wang, Web Filtering and Censoring[J], IEEE Computer Magazine.2010(3):94-97.
[22]Jerry Ropelato. Internet Pornograghy Statistics[OL]. http://internet-filter-review. toptenreviews.com/intenet-pornography-statistics.html.
[23]T. Ohtsuki and G. Healey. Using color and geometric models for extracting facial features[J]. Journal of Imaging Science and Technology.1998,42(6):554-561.
[24]Bosson A, Cawley G, Chan Y, et al. Non-retrieval:blocking pornographic images[C]. In:Lew MS, Sebe N, Eakins JP (Eds.). Proc. Int.Conf.on Image and Video Retrieval. London, UK:Springer,2002.50-60.
[25]Yoav Freund and Robert E. Schapire. Experiments with a new boosting algorithm[C]. Machine Learning:Proceedings of the Thirteenth International Conference.1996. 148-156.
[26]Viola Jones. Rapid object detection using a boosted cascade of simple features[C]. IEEE Conference on Computer Vision and Pattern Recognition.2001.511-518.
[27]Dempster, A.P, Laird, N.M, Rubin, D.B. Maximum likelihood from incomplete data via the EM algorithm[J]. Journal of the Royal Statistical Society.1977, Series B 39, pp 1-38,.
[28]Cover TM, Hart PE, Nearest neighbor pattern classification[J], IEEE Transactions on Information Theory 1967,13(1):21-27,.
[29]D.H. Ballard, Generalizing the Hough transform to detect arbitrary shapes[J], Pattern Recognition.1981,Vol.13, No.2:111-122,.
[30]J. Agbinya, B. Lok, Y. Sze Wong, et al. Automatic online porn detection and tracking[C]. in Proc. of 12th IEEE Int. Conf. on Telecommunications.2005, May, pp 1-5.
[31]陆燕,陈福生.基于内容的视频检索技术[J].计算机应用研究.2003,(11)：1-4.
[32]Andrew Burton and John Radford. Thinking in Perspective:Critical Essays in the Study of Thought Processes[M]. Routledge. ISBN 0416858406.
[33]David H. Warren and Edward R. Strelow. Electronic Spatial Sensing for the Blind: Contributions from Perception[M]. Springer. ISBN 9024726891.
[34]S.S. Beauchemin, J. L. Barron. The computation of optical flow[J].ACM Computing Surveys.1995, Volume 27 Issue 3, Sept.
[35]维基主成分分析[OL], http://en.wikipedia.org/wiki/Principal_component_analysis. visited on 2011-5-16.
[36]Pearson, K. On Lines and Planes of Closest Fit to Systems of Points in Space[J]. Philosophical Magazine 2 (6):559-572.
[37]Michael J. Jones, James M. Rehg. Statistical Color Models with Application to Skin Detection[J]. International Journal of Computer Vision,1999,(46):274-280.
[38]Opencv 中文网站[OL]. http://www.opencv.org.cn. visited on 2011-5-16.
[39]DH Ballard. Generalizing the Hough transform to detect arbitrary shapes [J]. Pattern recognition.1981, Volume 13, Issue 2, Pages 111-122.
[40]贾慧星,章毓晋.基于动态权重裁剪的快速Adaboost训练算法[J].计算机学报.2009,32(2)：336-341.
[41]杨琳,管业鹏.基于肤色分割与Adaboost融合鲁棒人脸检测方法[J]电子器件.2007,30(5)：1716-1719.
[42]刘晓雯,章毓晋,谭华春.基于Hausdorf距离的相似性和对称性度量及在人脸定位中的应用[J].信号处理.2008，,24(1)：118-121.
[43]Yue Feng, Reede Ren, Joemon Jose. Rule-Based Scene Boundary Detection for Semantic Video Segmentation[C]. Visual Information Engineering,2008. VIE 2008. 5th International Conference on.2008, pp 667-672.
[44]Paul voila, Micheal jones. Robust Real-time Face Detection[J]. International Journal of Computer Vision.2004, Volume 57, Number 2,137-154.
[45]Jerome Friedman, Trevor Hastie, Robert Tibshirani. Additive logistic regression:a statistical view of boosting[J]. Ann. Statist.2000, Volume 28, Number 2,337-407.
[46]Lienhart, R., Maydt, J. An extended set of Haar-like features for rapid object detection[C]. International Conference on Image Processing,2002,1-900-I-903 vol.1.
[2]任晓峰,吴巍.基于内容的视频检索算法研究[D].中国优秀博硕士学位论文全文数据库(硕士).2008，,(08).
[3]王伟强,高文,马继涌.一种高速切变镜头边界探测算法[J].计算机科学.2001,(07)：56-59.
[4]智敏,蔡安妮.基于基色调的镜头边界检测方法[J].自动化学报.2007,(06)：655-657.
[5]吕丽,杨树堂,陆松年等.基于光流法的不良视频检测算法[J].计算机工程.2007,(12)：220-221.
[6]刘晓楠.基于内容二次聚类的关键帧提取算法[11].计算机与数字工程.2010,(07)：26-29.
[7]吴静丽,马建斌,沈年凤等.基于视频关键帧的水印嵌入算法研究[J].电脑知识与技术.2010,(22)：6328-6330.
[8]王毅,李弼程.一种基于K-L变换和聚类的视频摘要方法[J].计算机应用研究.2010,(09)：3585-3587.
[9]王一丁.实际网络环境中不良图片的过滤方法[J].通信学报.2009,(S1)：103-106.
[10]赵更寅,邱兆文.基于SVM的不良图像检测[J].科技信息(科学教研).2007,(14)：278
[]1]谢志扬,史萍.网络不良视频信息过滤系统的研究与实现[J].中国传媒大学学报(自然科学版).2009,(07)：67-71
[12]吕丽,杨树堂,陆松年,李建华.基于光流法的不良视频检测算法[J].计算机工程.2007,(12)：220-221
[13]安红心,刘艳民.基于运动特征的不良视频检测算法[J].微计算机应用.2010,(07)：32-37.
[14]Fleck M, Forsyth D A, Bregler C. Find Naked People. In:Proc. of 4th European Conf.on Computer Vision[C].1996.592-602.
[15]V. Vezhnevets, V. Sazonov, A. Andreeva. A survey on pixel-based skin color detection techniques[C]. IN PROC. GRAPHICON-2003.2003.85-92.
[16]Hu W, Wu O, Chen Z, Fu Z, Maybank S. Recognition of pornographic web pages by classifying texts and images[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence.2007,29(6):1019-1034.
[17]DA Forsyth, MM Fleck. Automatic detection of human nudes[J]. International Journal of Computer Vision.1999, (32):63-77.
[18]P. Fuangkhon, T. Tanprasert. Nipple detection for obscene pictures[C]. In Proc. of WSEAS Int. Conf. on Signal, Speech and Image Processing.2005.315-320.
[19]Yue Wang, Jun Li, Heelin Wang, and Zujun Hou. Automatic nipple detection using shape and statistical skin color information[C]. Proceeding in Advances in Multimedia Modeling.2010.644-649.
[20]Cybersitter, http://www.solidoak.com, visited on 2010-10-9.
[21]Thomas M. Chen, Victoria Wang, Web Filtering and Censoring[J], IEEE Computer Magazine.2010(3):94-97.
[22]Jerry Ropelato. Internet Pornograghy Statistics[OL]. http://internet-filter-review. toptenreviews.com/intenet-pornography-statistics.html.
[23]T. Ohtsuki and G. Healey. Using color and geometric models for extracting facial features[J]. Journal of Imaging Science and Technology.1998,42(6):554-561.
[24]Bosson A, Cawley G, Chan Y, et al. Non-retrieval:blocking pornographic images[C]. In:Lew MS, Sebe N, Eakins JP (Eds.). Proc. Int.Conf.on Image and Video Retrieval. London, UK:Springer,2002.50-60.
[25]Yoav Freund and Robert E. Schapire. Experiments with a new boosting algorithm[C]. Machine Learning:Proceedings of the Thirteenth International Conference.1996. 148-156.
[26]Viola Jones. Rapid object detection using a boosted cascade of simple features[C]. IEEE Conference on Computer Vision and Pattern Recognition.2001.511-518.
[27]Dempster, A.P, Laird, N.M, Rubin, D.B. Maximum likelihood from incomplete data via the EM algorithm[J]. Journal of the Royal Statistical Society.1977, Series B 39, pp 1-38,.
[28]Cover TM, Hart PE, Nearest neighbor pattern classification[J], IEEE Transactions on Information Theory 1967,13(1):21-27,.
[29]D.H. Ballard, Generalizing the Hough transform to detect arbitrary shapes[J], Pattern Recognition.1981,Vol.13, No.2:111-122,.
[30]J. Agbinya, B. Lok, Y. Sze Wong, et al. Automatic online porn detection and tracking[C]. in Proc. of 12th IEEE Int. Conf. on Telecommunications.2005, May, pp 1-5.
[31]陆燕,陈福生.基于内容的视频检索技术[J].计算机应用研究.2003,(11)：1-4.
[32]Andrew Burton and John Radford. Thinking in Perspective:Critical Essays in the Study of Thought Processes[M]. Routledge. ISBN 0416858406.
[33]David H. Warren and Edward R. Strelow. Electronic Spatial Sensing for the Blind: Contributions from Perception[M]. Springer. ISBN 9024726891.
[34]S.S. Beauchemin, J. L. Barron. The computation of optical flow[J].ACM Computing Surveys.1995, Volume 27 Issue 3, Sept.
[35]维基主成分分析[OL], http://en.wikipedia.org/wiki/Principal_component_analysis. visited on 2011-5-16.
[36]Pearson, K. On Lines and Planes of Closest Fit to Systems of Points in Space[J]. Philosophical Magazine 2 (6):559-572.
[37]Michael J. Jones, James M. Rehg. Statistical Color Models with Application to Skin Detection[J]. International Journal of Computer Vision,1999,(46):274-280.
[38]Opencv 中文网站[OL]. http://www.opencv.org.cn. visited on 2011-5-16.
[39]DH Ballard. Generalizing the Hough transform to detect arbitrary shapes [J]. Pattern recognition.1981, Volume 13, Issue 2, Pages 111-122.
[40]贾慧星,章毓晋.基于动态权重裁剪的快速Adaboost训练算法[J].计算机学报.2009,32(2)：336-341.
[41]杨琳,管业鹏.基于肤色分割与Adaboost融合鲁棒人脸检测方法[J]电子器件.2007,30(5)：1716-1719.
[42]刘晓雯,章毓晋,谭华春.基于Hausdorf距离的相似性和对称性度量及在人脸定位中的应用[J].信号处理.2008，,24(1)：118-121.
[43]Yue Feng, Reede Ren, Joemon Jose. Rule-Based Scene Boundary Detection for Semantic Video Segmentation[C]. Visual Information Engineering,2008. VIE 2008. 5th International Conference on.2008, pp 667-672.
[44]Paul voila, Micheal jones. Robust Real-time Face Detection[J]. International Journal of Computer Vision.2004, Volume 57, Number 2,137-154.
[45]Jerome Friedman, Trevor Hastie, Robert Tibshirani. Additive logistic regression:a statistical view of boosting[J]. Ann. Statist.2000, Volume 28, Number 2,337-407.
[46]Lienhart, R., Maydt, J. An extended set of Haar-like features for rapid object detection[C]. International Conference on Image Processing,2002,1-900-I-903 vol.1.