一种基于改进的Xception网络的深度伪造视频检测模型

doi:10.3969/j.issn.1671-1122.2021.12.015

摘要/Abstract

摘要：

近年来,随着深度伪造技术的发展,深度伪造的内容变得更加难以识别,这给信息网络的安全带来了严峻的挑战。文章针对深度伪造篡改的内容不易识别以及现有深度伪造检测方法存在的面部特征提取不充分、参数量过大等问题,提出了一种融合Xception网络、SENet与WSDAN的深度伪造检测模型i_Xception。该模型将SE模块嵌入Xception网络中用来提取特征,再利用WSDAN模块对输入图片用注意力引导数据增强,把增强后的图像反馈回网络进行训练,以提高模型的检测精度。在此基础上,文章通过合理减小Xception网络的深度和宽度,设计了一个轻量级的网络模型i_miniXception,大大减少模型的参数。在目前深度伪造检测领域广泛使用的数据集FaceForensics++的两类子数据集FaceSwap和DeepFakes上验证,i_Xception检测的准确率分别达到99.50%和98.83%,i_miniXception检测的准确率分别达到99.17%和98.50%,优于现有的主流算法。

关键词: 深度伪造, 深度学习, Xception网络, 数据增强

Abstract:

In recent years, with the development of deep forgery technology, deep forged content have become more difficult to identify, which has brought severe challenges to the security of information network. Aiming at resolving the difficulty of identifying the content of deep forgery and tampering, as well as the insufficient facial featured extraction and excessive parameter amount in the existing deep forgery detection methods, this paper proposes a deep forgery detection model i_Xception that integrates Xception network, SENet and WSDAN. The model embeds the SE module in the Xception network to extract features, and then uses the WSDAN module to enhance the training images with the guidance of attention, and feeds the augmented images back to the network for training, which improves the detection accuracy of the model. On this basis, this paper designs a lightweight network model i_miniXception by reasonably reducing the depth and width of the Xception network and fusing the above methods, which greatly reduces the parameters of the model. It is verified on the two types of datasets FaceSwap and DeepFakes of FaceForensics++, which are currently widely used in the field of deep forgery detection. The accuracy of i_Xception detection reaches 99.50% and 98.83%, and the accuracy of i_miniXception detection reaches 99.17% and 98.50% respectively, which are better than existing main algorithms.

Key words: deep forgery, deep learning, XceptionNet, data augmentation

中图分类号:

TP309

马瑞, 蔡满春, 彭舒凡. 一种基于改进的Xception网络的深度伪造视频检测模型[J]. 信息网络安全, 2021, 21(12): 109-117.

MA Rui, CAI Manchun, PENG Shufan. A Deep Forgery Video Detection Model Based on Improved Xception Network[J]. Netinfo Security, 2021, 21(12): 109-117.

图/表 6

图1

图2

图3

表1

图4

表2

参考文献 27

[1]	SHU Kai, SLIVA A, WANG Suhang, et al. Fake News Detection on Social Media: A Data Mining Perspective[J]. ACM SIGKDD Explorations Newsletter, 2017, 19(1):22-36. doi: 10.1145/3137597.3137600 URL
[2]	KINGMA D P, WELLING M. Auto-encoding Variational Bayes[EB/OL]. https://arxiv.org/abs/1312.6114, 2013-12-20.
[3]	REZENDE D J, MOHAMED S, WIERSTRA D. Stochastic Backpropagation and Approximate Inference in Deep Generative Models[C]// ICML. International Conference on Machine Learning, June 21-26,2014, Beijing, China. New York: PMLR, 2014: 1278-1286.
[4]	GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al. Generative Adversarial Networks[J]. Communications of the ACM, 2020, 63(11):139-144. doi: 10.1145/3422622 URL
[5]	AGARWAL S, FARID H, GU Y, et al. Protecting World Leaders against Deep Fakes[C] // IEEE. In: Proc. of the IEEE Conf. on Computer Vision and Pattern Recognition Workshops, June 16-20, 2019, Long Beach, CA, USA. New York: IEEE, 2019: 38-45.
[6]	WANG Run, MA Lei, JUEFEI-XU F, et al. Fakespotter: A Simple Yet Robust Baseline for Spotting Ai-synthesized Fake Faces[EB/OL]. https://arxiv.org/abs/1909.06122, 2019-09-13.
[7]	AMOS B, LUDWICZUK B, SATYANARAYANAN M. Openface: A General-purpose Face Recognition Library with Mobile Applications[EB/OL]. http://reports-archive.adm.cs.cmu.edu/anon/anon/usr0/ftp/2016/CMU-CS-16-118.pdf, 2021-10-06.
[8]	SCHROFF F, KALENICHENKO D, PHILBIN J. Facenet: A Unified Embedding for Face Recognition and Clustering[C]// IEEE. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, June 7-12, 2015, Boston, MA, USA. New York: IEEE, 2015: 815-823.
[9]	AMERINI I, GALTERI L, CALDELLI R, et al. Deepfake Video Detection through Optical Flow Based CNN[C]// IEEE. In: Proc. of the IEEE International Conference. on Computer Vision Workshops, October 27-28, 2019, Seoul, Korea (South). New York: IEEE, 2019: 1205-1207.
[10]	SABIR E, CHENG Jiaxin, JAISWAL A, et al. Recurrent Convolutional Strategies for Face Manipulation Detection in Videos[J]. Interfaces (GUI), 2019, 3(1):80-87.
[11]	MITTAL T, BHATTACHARYA U, CHANDRA R, et al. Emotions Don’t Lie: An Audio-uisual Deepfake Detection Method Using Affective Cues[C]// ACM. Proceedings of the 28th ACM International Conference on Multimedia, October 12-16, 2020, Seattle, WA, USA. New York: ACM, 2020: 2823-2832.
[12]	LIMA O D, FRANKLIN S, BASU S, et al. Deepfake Detection Using Spatiotemporal Convolutional Networks[EB/OL]. https://arxiv.org/abs/2006.14749, 2020-06-26.
[13]	AFCHAR D, NOZICK V, YAMAGISHI J, et al. Mesonet: A Compact Facial Video Forgery Detection Network[C]// IEEE. In: Proc. of the IEEE Int’l Workshop on Information Forensics and Security (WIFS), December 11-13, 2018, Hong Kong, China. New York: IEEE, 2018: 1-7.
[14]	NGUYEN HH, YAMAGISHI J, ECHIZEN I. Capsule-forensics: Using Capsule Networks to Detect Forged Images and Videos[C]// IEEE. In: Proc. Of the IEEE Int’l Conf. on Acoustics, Speech and Signal Processing (ICASSP 2019), May 12-17, 2019, Brighton, United Kingdom. New York: IEEE, 2019: 2307-2311.
[15]	DANG Hao, LIU Feng, STEHOUWER J, et al. On the Detection of Digital Face Manipulation[C]// IEEE. In:Proc. of the IEEE Conference. on Computer Vision and Pattern Recognition (CVPR), May 12-17, 2019, Brighton, United Kingdom. New York: IEEE, 2020: 5780-5789.
[16]	TAN Mingxing, LE Q. Efficientnet: Rethinking Model Scaling for Convolutional Neural Networks[C]// ICML. International Conference on Machine Learning. June 9-15, 2019, Long Beach, California, USA. New York: PMLR, 2019: 6105-6114.
[17]	ZHU Xintong, TANG Yunqi, GENG Pengzhi. Detection Algorithm of Tamper and Deepfake Image Based on Feature Fusion[J]. Netinfo Security, 2021, 21(8):70-81.
	朱新同, 唐云祁, 耿鹏志. 基于特征融合的篡改与深度伪造图像检测算法[J]. 信息网络安全, 2021, 21(8):70-81.
[18]	CHOLLET F. Xception: Deep Learning with Depthwise Separable Convolutions[C]// IEEE. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, July 21-26, 2017, Honolulu, HI, USA. NewYork: IEEE, 2017: 1251-1258.
[19]	CHEN Shuping, CHEN Bingcai. Research on Image Classification Method Based on Improved Xception Model[M]// LIANG Qilian,WANG Wei, MU Jiasong,et al.Artificial Intelligence in China. Singapore: Springer, 2021: 538-547.
[20]	WOO S, PARK J, LEE J Y, et al. Cbam: Convolutional Block Attention Module[C]// ICAIS. Proceedings of the European Conference on Computer Vision (ECCV), July 26-28, 2019, New York, NY, USA. New York: Springer, 2019: 3-19.
[21]	LIN Hao, LUO Weiqi, WEI Kangkang, et al. Improved Xception with Dual Attention Mechanism and Feature Fusion for Face Forgery Detection[EB/OL]. https://arxiv.org/abs/2109.14136, 2021-9-29.
[22]	HU Jie, SHEN Li, SUN Gang. Squeeze-and-excitation networks[C]// IEEE. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, July 26-28, 2019, New York, USA. New York: IEEE, 2018: 7132-7141.
[23]	HU Tao, QI Honggang, HUANG Qingming, et al. See Better Before Looking Closer: Weakly Supervised Data Augmentation Network for Fine-grained Visual Classification[EB/OL]. https://arxiv.org/abs/1901.09891, 2019-01-26.
[24]	DENG Jiankang, GUO Jia, ZHOU Yuxiang, et al. Retinaface: Single-stage Dense Face Localisation in the Wild[EB/OL]. https://arxiv.org/abs/1905.00641, 2019-05-02.
[25]	SHORTEN C, KHOSHGOFTAAR T M. A Survey on Image Data Augmentation for Deep Learning[J]. Journal of Big Data, 2019, 6(1):1-48. doi: 10.1186/s40537-018-0162-3 URL
[26]	CUBUK E D, ZOPH B, MANE D, et al. Autoaugment: Learning Augmentation Policies from Data[EB/OL]. https://arxiv.org/abs/1805.09501, 2018-05-24.
[27]	ROSSLER A, COZZOLINO D, VERDOLIVA L, et al. Faceforensics++: Learning to Detect Manipulated Facial Images[C]// IEEE. Proceedings of the IEEE/CVF International Conference on Computer Vision, October 27-November 2, 2019, Seoul, Korea (South). New York: IEEE, 2019: 1-11.

Model				Acc		CrossEntropy Loss		Param-eters/MB
Xception	Mini_Xception	SENet	WSDAN	FS	DP	FS	DP	Param-eters/MB
√	—	—	—	98.17%	97.50%	0.0863	0.1027	21.80
√	—	√	—	99.17%	98.33%	0.0752	0.0863	22.30
√	—	√	√	99.50%	98.83%	0.0628	0.0786	20.53
—	√	—	—	97.67%	97.16%	0.0893	0.1063	1.96
—	√	√	—	98.50%	98.16%	0.0796	0.0886	2.37
—	√	√	√	99.17%	98.50%	0.0746	0.0832	2.05

方法	Acc		Parameters/MB
方法	FaceSwap	Deepfakes	Parameters/MB
文献[10]方法	96.30%	96.50%	-
Mesonet	93.43%	95.26%	0.11
Efficientnet b3	98.00%	97.60%	12.23
XceptionNet	98.17%	97.50%	21.80
i_Xception(ours)	99.50%	98.83%	20.53
I_miniXception(ours)	99.17%	98.50%	2.05