在過去不良圖片的辨識通常利用皮膚特徵或是圖片附註的關鍵字描述搭配多個過濾器進行過濾。深度學習通常都需要花費大量的時間來訓練，。隨著硬體的進步以及新的演算法不斷提出，訓練時間問題漸漸緩解。深度卷積神經網路在圖片的分類上有非常良好的效果。在此篇論文中我們利用深度卷積神經網路來解決不良圖片分類的問題。這也是我們所知第一篇使用深度卷積神經網路來學習以及分類不良圖片的論文。我們比較不同模型的深度卷積神經網路的效能，使用八萬張圖片在我們的圖形處理器環境下，大約花費2.5小時就能取得97%左右的分類精確度，這是非常有競爭力的結果。

In the past, people usually use the skin feature extract or the keywords for text contents combined many filters to recognize the objectionable images. Deep learning usually takes lots of time to train. With the advances in hardware and new algorithm proposed, the training time problem is gradually alleviated. The deep convolutional neural networks have good effect on images classification. In this paper, we use deep convolutional neural networks to solve classification of the objectionable images problem. To the best of our knowledge, this study is the first to use deep convolutional neural networks to learn and classify objectionable images. We compare the effects with different deep convolutional neural network models. Using 80000 training images in our GPU environment, we can get 97% classification accuracy with training cost of only 2.5 hours. This is a very competitive result.

[1] Abadpour, A., & Kasaei, S. (2005). Pixel-based skin detection for pornography filtering. Iranian Journal of Electrical and Electronic Engineering, 1(3), 21-41.
[2] Fleck, M. M., Forsyth, D. A., & Bregler, C. (1996). Finding naked people. In Computer Vision—ECCV'96 (pp. 593-602). Springer Berlin Heidelberg.
[3] Forsyth, D. A., & Fleck, M. M. (1996, December). Identifying nude pictures. In Applications of Computer Vision, 1996. WACV'96., Proceedings 3rd IEEE Workshop on (pp. 103-108). IEEE.
[4] Forsyth, D. A., & Fleck, M. M. (1999). Automatic detection of human nudes. International Journal of Computer Vision, 32(1), 63-77.
[5] Ioffe, S., & Forsyth, D. (1999). Finding people by sampling. In Computer Vision, 1999. The Proceedings of the Seventh IEEE International Conference on (Vol. 2, pp. 1092-1097). IEEE.
[6] Wu, Y. L., Goh, K. S., Li, B., You, H., & Chang, E. Y. (2003, August). The anatomy of a multimodal information filter. In Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 462-471). ACM.
[7] Wu, Y. L., Chang, E., Cheng, K. T., Chang, C. W., Hsu, C. C., Lai, W. C., & Wu, C. T. (2002). MORF: A distributed multimodal information filtering system. Advances in Multimedia Information Processing—PCM 2002, 97-104.
[8] Kia, S. M., Rahmani, H., Mortezaei, R., Moghaddam, M. E., & Namazi, A. (2014). A Novel Scheme for Intelligent Recognition of Pornographic Images. arXiv preprint arXiv:1402.5792.
[9] Marcial-Basilio, J. A., Aguilar-Torres, G., Sánchez-Pérez, G., ToscanoMedina, L. K., & Pérez-Meana, H. M., (2011). Detection of Pornographic Digital Images . International Journal of Computers (Vol. 5)
[10] Wang, J. Z., Li, J., Wiederhold, G., & Firschein, O. (1998). System for screening objectionable images. Computer Communications, 21(15), 1355-1360.
[11] GPU development in recent years, “http://bkultrasound.com/blog/the-next-generation-of-ultrasound-technology”, Referenced on May 15 th, 2015
[12] Bengio, Y. (2009). Learning deep architectures for AI. Foundations and trends® in Machine Learning, 2(1), 1-127.
[13] Lee, T. H., (2015) A Study of General Fine-Grained Classification with Deep Convolutional Neural Networks, National Taiwan University of Science and Technology Department of Computer Science and Information Engineering
[14] Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., ... & Rabinovich, A. (2015). Going deeper with convolutions. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 1-9).
[15] LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
[16] Typical CNN architecture,
“https://en.wikipedia.org/wiki/Convolutional_neural_network”
Referenced on May 31 th, 2016
[17] Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., ... & Darrell, T. (2014, November). Caffe: Convolutional architecture for fast feature embedding. In Proceedings of the ACM International Conference on Multimedia (pp. 675-678). ACM.
[18] Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. In Advances in neural information processing systems (pp. 1097-1105).
[19] Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., ... & Darrell, T. (2014, November). Caffe: Convolutional architecture for fast feature embedding. In Proceedings of the ACM International Conference on Multimedia (pp. 675-678). ACM. \
[20] Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., ... & Berg, A. C. (2015). Imagenet large scale visual recognition challenge. International Journal of Computer Vision, 115(3), 211-252.
[21] Nair, V., & Hinton, G. E. (2010). Rectified linear units improve restricted boltzmann machines. In Proceedings of the 27th International Conference on Machine Learning (ICML-10) (pp. 807-814).
[22] Boureau, Y. L., Bach, F., LeCun, Y., & Ponce, J. (2010, June). Learning mid-level features for recognition. In Computer Vision and Pattern Recognition (CVPR), 2010 IEEE Conference on (pp. 2559-2566). IEEE.
[23] Hinton, G. E., Srivastava, N., Krizhevsky, A., Sutskever, I., & Salakhutdinov, R. R. (2012). Improving neural networks by preventing co-adaptation of feature detectors. arXiv preprint arXiv:1207.0580.