基于多分类和ResNet的不良图片识别框架
|
摘要
针对实际应用中色情图片的复杂多样性问题,提出一种基于多分类和深度残差网络(ResNet)的不良图片识别框架.不同于已有的方法将色情图片识别作为二分类问题,该方法基于多样性特征将色情图片分为7个更细粒度的类别,并将正常图片分为是否包含人物2个类别,通过50层ResNet模型进行分类,再按照阈值计算是否属于不良图片.为了减少训练时间和挖掘优质特征,采用一种反馈修正的训练策略.提出一种单边滑动窗口的预处理方法以解决图片不同尺度的影响问题.测试结果表明,该方法在时间效率和识别准确率上效果良好.
To filter the variety of pornographic images in the reality Internet, the study proposed a Pornographic Images Recognition(PIR) framework based on multi-classification and deep Residual Network(ResNet). Traditional methods usually consider the PIR task as a binary classification, while the approach presented in this paper divides porno images into 7 detailed classes based on its variety features with 2 more benign image classes(with or without human in it). The approach relies on 50-ResNet to extract image features automatically, and then decides whether it belongs to porno images based on the highest score and gives threshold value. At training stage, a feedback-reconstruct training tactics is adopted for the network to collect better features. To deal with images in different scales, a monolateral sliding window method is taken to get better performance. After testing on the data set constructed with collected images from the Internet, the experimental result shows that the approach can reach high accuracy with lower time cost.
To filter the variety of pornographic images in the reality Internet, the study proposed a Pornographic Images Recognition(PIR) framework based on multi-classification and deep Residual Network(ResNet). Traditional methods usually consider the PIR task as a binary classification, while the approach presented in this paper divides porno images into 7 detailed classes based on its variety features with 2 more benign image classes(with or without human in it). The approach relies on 50-ResNet to extract image features automatically, and then decides whether it belongs to porno images based on the highest score and gives threshold value. At training stage, a feedback-reconstruct training tactics is adopted for the network to collect better features. To deal with images in different scales, a monolateral sliding window method is taken to get better performance. After testing on the data set constructed with collected images from the Internet, the experimental result shows that the approach can reach high accuracy with lower time cost.
引文
1凡阿杰. 网络视频色情信息的检测技术研究与应用[硕士学位论文]. 北京: 北方工业大学, 2017.
2裴向杰, 唐红昇, 陈鹏. 融合 YCb Cr 肤色分割的不良图像检测算法研究. 计算机技术与发展, 2015, 25(12): 80–84,90.
3王景中, 周靖. 基于比例特征的网络不良图像过滤算法研究. 计算机工程与科学, 2016, 38(3): 514–519. [doi: 10.3969/j.issn.1007-130X.2016.03.018]
4黄杰, 史啸. 一种基于人体裸露皮肤形状的不良图像过滤系统. 东南大学学报 (自然科学版), 2014, 44(6): 1111–1115.
5Ding M, Fan GL. Multi-layer joint gait-pose manifold forhuman motion modeling. IEEE Transactions on Cybernetics,2015, 45(11): 2413–2424. [doi: 10.1109/TCYB.2014.2373393]
6Shao H, Yu TS, Xu MJ, et al. Image region duplicationdetection based on circular window expansion and phasecorrelation. Forensic Science International, 2012, 222(1–3):71–82. [doi: 10.1016/j.forsciint.2012.05.002]
7Krizhevsky A, Sutskever I, Hinton GE. Image Netclassification with deep convolutional neural networks.Communications of the ACM, 2017, 60(6): 84–90. [doi:10.1145/3098997]
8He KM, Zhang XY, Ren SQ, et al. Deep residual learning forimage recognition. 2016 IEEE Conference on ComputerVision and Pattern Recognition. Las Vegas, NV, USA. 2016.770–778. [doi: 10.1109/CVPR.2016.90]
9Moustafa M. Applying deep learning to classifypornographic images and videos. ar Xiv preprint ar Xiv:1511.08899, 2015.
10Zhou KL, Zhuo L, Geng Z, et al. Convolutional neuralnetworks based pornographic image classification. 2016IEEE Second International Conference on Multimedia BigData. Taipei, China. 2016. 206–209. [doi: 10.1109/Big MM.2016.29]
11Huang Y, Kong AWK. Using a CNN ensemble for detectingpornographic and upskirt images. 2016 IEEE 8thInternational Conference on Biometrics Theory, Applicationsand Systems (BTAS). Niagara Falls, NY, USA. 2016. 1–7.
12Simonyan K, Zisserman A. Very deep convolutionalnetworks for large-scale image recognition. ComputerScience, 2014: 1–14. [doi: 10.5121/csit.2014.41000]
13Szegedy C, Liu W, Jia YQ, et al. Going deeper withconvolutions. 2015 IEEE Conference on Computer Visionand Pattern Recognition. Boston, MA, USA. 2015. 1–9. [doi:10.1109/CVPR.2015.7298594]
14Wang YH, Jin X, Tan XY. Pornographic image recognitionby strongly-supervised deep multiple instance learning. 2016IEEE International Conference on Image Processing (ICIP).Phoenix, AZ, USA. 2016. 4418–4422. [doi: 10.1109/ICIP.2016.7533195]
15Nian FD, Li T, Wang Y, et al. Pornographic image detectionutilizing deep convolutional neural networks.Neurocomputing, 2016, 210: 283–293. [doi: 10.1016/j.neucom.2015.09.135]
16Connie T, Al-Shabi M, Goh M. Smart content recognitionfrom images using a mixture of convolutional neuralnetworks. IT Convergence and Security 2017. Springer.Singapore. 2018. 11–18.
17Zhuo L, Geng Z, Zhang J, et al. ORB feature based webpornographic image recognition. Neurocomputing, 2016,173: 511–517. [doi: 10.1016/j.neucom.2015.06.055]
2裴向杰, 唐红昇, 陈鹏. 融合 YCb Cr 肤色分割的不良图像检测算法研究. 计算机技术与发展, 2015, 25(12): 80–84,90.
3王景中, 周靖. 基于比例特征的网络不良图像过滤算法研究. 计算机工程与科学, 2016, 38(3): 514–519. [doi: 10.3969/j.issn.1007-130X.2016.03.018]
4黄杰, 史啸. 一种基于人体裸露皮肤形状的不良图像过滤系统. 东南大学学报 (自然科学版), 2014, 44(6): 1111–1115.
5Ding M, Fan GL. Multi-layer joint gait-pose manifold forhuman motion modeling. IEEE Transactions on Cybernetics,2015, 45(11): 2413–2424. [doi: 10.1109/TCYB.2014.2373393]
6Shao H, Yu TS, Xu MJ, et al. Image region duplicationdetection based on circular window expansion and phasecorrelation. Forensic Science International, 2012, 222(1–3):71–82. [doi: 10.1016/j.forsciint.2012.05.002]
7Krizhevsky A, Sutskever I, Hinton GE. Image Netclassification with deep convolutional neural networks.Communications of the ACM, 2017, 60(6): 84–90. [doi:10.1145/3098997]
8He KM, Zhang XY, Ren SQ, et al. Deep residual learning forimage recognition. 2016 IEEE Conference on ComputerVision and Pattern Recognition. Las Vegas, NV, USA. 2016.770–778. [doi: 10.1109/CVPR.2016.90]
9Moustafa M. Applying deep learning to classifypornographic images and videos. ar Xiv preprint ar Xiv:1511.08899, 2015.
10Zhou KL, Zhuo L, Geng Z, et al. Convolutional neuralnetworks based pornographic image classification. 2016IEEE Second International Conference on Multimedia BigData. Taipei, China. 2016. 206–209. [doi: 10.1109/Big MM.2016.29]
11Huang Y, Kong AWK. Using a CNN ensemble for detectingpornographic and upskirt images. 2016 IEEE 8thInternational Conference on Biometrics Theory, Applicationsand Systems (BTAS). Niagara Falls, NY, USA. 2016. 1–7.
12Simonyan K, Zisserman A. Very deep convolutionalnetworks for large-scale image recognition. ComputerScience, 2014: 1–14. [doi: 10.5121/csit.2014.41000]
13Szegedy C, Liu W, Jia YQ, et al. Going deeper withconvolutions. 2015 IEEE Conference on Computer Visionand Pattern Recognition. Boston, MA, USA. 2015. 1–9. [doi:10.1109/CVPR.2015.7298594]
14Wang YH, Jin X, Tan XY. Pornographic image recognitionby strongly-supervised deep multiple instance learning. 2016IEEE International Conference on Image Processing (ICIP).Phoenix, AZ, USA. 2016. 4418–4422. [doi: 10.1109/ICIP.2016.7533195]
15Nian FD, Li T, Wang Y, et al. Pornographic image detectionutilizing deep convolutional neural networks.Neurocomputing, 2016, 210: 283–293. [doi: 10.1016/j.neucom.2015.09.135]
16Connie T, Al-Shabi M, Goh M. Smart content recognitionfrom images using a mixture of convolutional neuralnetworks. IT Convergence and Security 2017. Springer.Singapore. 2018. 11–18.
17Zhuo L, Geng Z, Zhang J, et al. ORB feature based webpornographic image recognition. Neurocomputing, 2016,173: 511–517. [doi: 10.1016/j.neucom.2015.06.055]