﻿ 基于速度场的手持重拍摄视频检测算法 Analysis of Hand-Held Recaptured Video Detection Algorithm Exploiting Velocity Field

Computer Science and Application
Vol.06 No.12(2016), Article ID:19236,11 pages
10.12677/CSA.2016.612091

Analysis of Hand-Held Recaptured Video Detection Algorithm Exploiting Velocity Field

Zhuoyan Jiang, Tanfeng Sun, Xinghao Jiang

Shanghai Jiao Tong University, Shanghai

Received: Nov. 27th, 2016; accepted: Dec. 8th, 2016; published: Dec. 15th, 2016

ABSTRACT

Video recaptured detection is one of the important methods of detecting video infringement. It is a passive detection method and can be used for video copyright. In this paper, combining with the hand-held video characteristics and the velocity field algorithm, we calculate the velocity field between frames, according to the moving trend to judge whether velocity field changes or not. We extract a set of velocity change values and select the average and variance value as a two-dimen- sional characteristic, and finally we use support vector machine (SVM) for experiments. The experiment result shows that this algorithm has a higher accuracy.

Keywords:Passive Detection, Hand-Held Recapture, Velocity Field, Velocity Field Change Degree, SVM

1. 引言

2. 手持特征概述

2.1. 手持特征分析

2.2. 速度场理论

Figure 1. Hand-held recapture scene

2.3. 基于速度场的二维手持特征

3. 基于速度场的手持重拍摄视频检测算法

3.1. 算法框架

(a) (b)

Figure 2. Move direction area divide. (a) Area partition; (b) Vector example

Figure 3. Algorithm framework

3.2. 算法流程

3.3. 速度场计算

(1)

1) 对于静止的视频，只需计算相邻帧速度场就能得到比较清晰一致的运动轨迹；

2) 对于动态的视频而言，使用矢量图观测运动轨迹不是很清晰，很难鉴定该运动轨迹是由帧内运动造成还是由手持抖动造成。

Figure 4. Algorithm flow

3.4. 速度场变化程度提取

(2)

(a) (b) (c)

Figure 5. Area divide. (a) Pre-velocity field move direction; (b) Post-velocity field move direction 1; (c) Post-velocity field move direction 2

3.5. 手持二维特征提取

(3)

(4)

(a)(b)

Figure 6. VC value. (a) VC value of origin video; (b) VC value of recapture video

3.6. SVM分类

SVM即支持向量机，属于机器学习领域中一个有监督的学习模型，通常用于模式识别与分类。

SVM的主要思想概括来讲为以下两点：

1) 针对线性可分情况直接进行分析；

2) 针对线性不可分情况，通过非线性算法将低维输入空间的不可份样本转化至高维空间，使其线性可分。

4. 实验与分析

4.1. 数据集

4.2. 实验结果与分析

1) 从表中数据可以看出本文算法在正样本的检测上准确率达到了96.6%，在负样本的检测上达到了100%，平均准确率达到了98.3%，实验表明了本文算法中的二维手持特征能很好的区分手持重拍摄视频与原始视频。

2) 本文算法还与Marco Visentini-Scarzanella’s [3] 的算法进行了对比，Marco Visentini-Scarzanella’算法的准确率有文献 [3] 给出，本文算法在TPR上提高了6.6%，在TNR上提高了30%，在平均准确率上提高了18.3%。实验表明了本文算法对于手持重拍摄视频而言有着更好的检测准确率。

4.3. 误差分析

4.4. 算法效率统计

5. 总结

Figure 7. Video set

Figure 8. Misjudged origin video

Table 1. Experiment result

TPR：true positive rate，在本文算法中表示原始视频被判断为原始视频的概率；TNR：true negative rate，在本文算法中表示重拍摄视频被判断为重拍摄视频的概率。

Table 2. Running time

Analysis of Hand-Held Recaptured Video Detection Algorithm Exploiting Velocity Field[J]. 计算机科学与应用, 2016, 06(12): 761-771. http://dx.doi.org/10.12677/CSA.2016.612091

1. 1. Hajj-Ahmad, A, Baudry, S., Chupeau, B., et al. (2015) Flicker Forensics for Pirate Device Identification. Proceedings of the 3rd ACM Workshop on Information Hiding and Multimedia Security, Portland, 17-19 June 2015, 75-84. https://doi.org/10.1145/2756601.2756612

2. 2. Rolland-Neviere, X., Chupeau, B., Doerr, G., et al. (2012) Forensic Characterization of Camcorded Movies: Digital Cinema vs. Celluloid Film Prints. Proceedings of SPIE—The International Society for Optical Engineering, Burlin-game, 9 February 2012, 83030R-83030R-11.

3. 3. Lee, J.W., Lee, M.J., Lee, H.Y., et al. (2012) Screenshot Identification by Analysis of Directional Inequality of Interlaced Video. EURASIP Journal on Image and Video Processing, No. 1, 1-15. https://doi.org/10.1186/1687-5281-2012-7

4. 4. Visentini-Scarzanella, M. and Dragotti, P.L. (2012) Video Jitter Analysis for Automatic Bootleg Detection. 2012 IEEE 14th International Workshop on Multimedia Signal Processing (MMSP), Banff, 17-19 September 2012, 101-106. https://doi.org/10.1109/MMSP.2012.6343423

5. 5. Bestagini, P., Visentini-Scarzanella, M., Tagliasacchi, M., et al. (2013) Video Recapture Detection Based on Ghosting Artifact Analysis. 2013 IEEE International Conference on Image Processing, Melbourne, 15-18 September 2013, 4457-4461. https://doi.org/10.1109/ICIP.2013.6738918

6. 6. Moreira-Perez, J.J., Chupeau, B., Doerr, G., et al. (2013) Ex-ploring Color Information to Characterize Camcorder Piracy. 2013 IEEE International Workshop on Information Forensics and Security (WIFS), Guangzhou, 18-21 November 2013, 132-137. https://doi.org/10.1109/WIFS.2013.6707807

7. 7. Thongkamwitoon, T., Muammar, H. and Dragotti, P.L. (2015) An Image Recapture Detection Algorithm Based on Learning Dictionaries of Edge Profiles. IEEE Transactions on Information Forensics and Security, 10, 953-968. https://doi.org/10.1109/TIFS.2015.2392566

8. 8. Mahdian, B., Novozámský, A. and Saic, S. (2015) Identification of Aliasing-Based Patterns in Re-Captured LCD Screens. 2015 IEEE International Conference on Image Processing (ICIP), Quebec City, 27-30 September 2015, 616-620. https://doi.org/10.1109/ICIP.2015.7350872

9. 9. 吴俞醒. 基于连续性特征的视频帧间篡改检测算法的研究与实现[D]: [硕士学位论文]. 上海: 上海交通大学, 2015.

10. 10. Chupeau, B., Baudry, S. and Doerr, G. (2014) Forensic Characterization of Pirated Movies: Digital Cinema Cam vs. Optical Disc Rip. 2014 IEEE International Workshop on Information Forensics and Security (WIFS), Atlanta, 3-5 December 2014, 155-160. https://doi.org/10.1109/WIFS.2014.7084320

11. 11. Schaber, P., Kopf, S., Wetzel, S., et al. (2015) Cammark: Analyzing, Modeling, and Simulating Artifacts in Camcorder Copies. ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM), 11, 42. https://doi.org/10.1145/2700295

12. 12. Schaber, P., Dong, S., Guthier, B., et al. (2015) Modeling Temporal Effects in Re-Captured Video. Proceedings of the 23rd ACM international conference on Multimedia, Brisbane, 26-30 October 2015, 1279-1282. https://doi.org/10.1145/2733373.2806405

13. 13. Hajj-Ahmad, A., Doerr, G., Wu, M., et al. (2016) Flicker Forensics for Camcorder Piracy. IEEE Transactions on Information Forensics and Security, 12, 89-100. https://doi.org/10.1109/TIFS.2016.2603603