Video Authentication Using Relative Correlation Information and SVM
Video authentication is often presented as evidence in many criminal cases. Therefore, the authenticity of the video data is of paramount interest. This paper presents an intelligent video authentication algorithm using support vector machine. The propose
- PDF / 6,838,336 Bytes
- 19 Pages / 595 x 842 pts (A4) Page_size
- 45 Downloads / 149 Views
2
West Virginia University, USA [email protected], [email protected] Purvanchal University, India [email protected]
Summary. Video authentication is often presented as evidence in many criminal cases. Therefore, the authenticity of the video data is of paramount interest. This paper presents an intelligent video authentication algorithm using support vector machine. The proposed algorithm does not require the computation and storage of secret key or embedding of watermark. It computes the local relative correlation information and classifies the video as tampered or non-tampered. Performance of the proposed algorithm is not affected by acceptable video processing operations such as compression and scaling and effectively classifies the tampered videos. On a database of 795 videos, the proposed algorithm outperforms the existing algorithm by 18.5%.
1 Introduction In today’s digital era, communication and compression techniques facilitate sharing of multimedia data such as image and video. However, multimedia editing tools can be used to efficiently and seamlessly alter the content of digital data thus compromising the reliability. In some applications, the reliability of video data is of paramount interest such as in video surveillance, forensics, law enforcement, and content ownership. For example, in court of law, it is important to establish the trustworthiness of any video that is used as evidence. So, video authentication is a process which ascertains that the content in a given video is authentic and exactly same as when captured. It also detects the type and location of malicious tampering. To accomplish this task automatically, several algorithms have been proposed which extract unique and resilient features from video and generate an authentication data. This authentication data is further used to establish the authenticity of the video content. There are several possible attacks that can be applied to alter the contents of a video data. These attacks can be classified into five classes. M. Vatsa et al.: Video Authentication Using Relative Correlation Information and SVM, Studies in Computational Intelligence (SCI) 96, 511–529 (2008) c Springer-Verlag Berlin Heidelberg 2008 www.springerlink.com
512
M. Vatsa et al.
Fig. 1. Example of frame addition attack. Top row shows the original frame sequence with frames 10 and 18. Bottom row shows the frame sequence after attack in which a new frame is inserted between 10 and 18 and frame 18 becomes frame 19
1. Frame addition attack. In frame addition attack, additional frames are deliberately inserted at some position in a given video. This attack is intended to camouflage the actual content and provide incorrect information. A simple example of frame addition attack is shown in Fig. 1. 2. Frame removal attack. In frame removal attack, frames are intentionally removed from the video. This attack is common in criminal investigation where an intruder wants to remove his/her presence from a surveillance video. Figure 2 shows an example of the frame removal attack. 3
Data Loading...
