Loading...

Blind Detection of Processes Applied to Video Via Action Effects Analysis

Amanipour, Vahideh | 2019

506 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 51812 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Gaemmaghami, Shahrokh
  7. Abstract:
  8. With quick development of inexpensive video acquisition tools, fast progress of digital technology and growing availability of video editing software tools, editing and tampering of video signal has become easy and straightforward. This fact highlights the need for tools and methods that recover the editing history of digital data which has, for instance, motivated a joint project between several European universities and institutions called REWIND. Our main goal in this thesis is the blind (partial) detection of editing history for video signals without access to the original signal. The video is processed in the acquisition, coding and editing phases. Our focus in this thesis is on the detection problem in the editing phase. Editing detection is widely studied in the literature for image signals, while less attention is paid to designing detection schemes for video. This detection problem is approached in the literature either by embedding a suitable watermark, or through detection of the footprints of editing phases. Reconstruction of the original content (of acceptable quality) for highly modified videos is an important challenge in malicious editing detection. In light of previous approaches to the study of this problem, this leads us to employ the watermark embedding approach for detection of malicious editing, while the second approach is used for detection of benign editing for purposes such as enhancement and geometric modification. Editing detection in the compressed domain is considered a major challenge.
    Adapting a self-embedding scheme for the video signal, we first introduce a method for detection of editing history, localization of tampering and successful reconstruction of the original content for tampering rate as high as 67%. The watermark payload, which consists of compressed versions of keyframes and localization information, is embedded in the video using fountain coding. Moreover, we propose a detection scheme in compressed domain based on the singular value decomposition (SVD) of the process matrix, which is a linear approximation of the editing process . Projection over some of the eigen-spaces of the process matrix gives a set of features of small dimension , even as small as 3 , making the proposed scheme a fast and suitable detector . When editing manipulation is well-approximated by the process matrix, as in the case of median filter, the proposed scheme performs well over compressed video signals. The experimental evaluations show that the proposed method outperforms the state of the art detectors of median filtering , and its edge becomes significant for localized detection over small blocks within frames
  9. Keywords:
  10. Video Signal Processing ; Blind Detection ; Video Watermaking ; Video Forensics ; Self Embedding

 Digital Object List

 Bookmark

No TOC