Digital image and videos cannot be taken as authentic evidences, since their integrity is no longer trustworthy. Because of the uniqueness and peculiarities of video signals with respect to images, there are wider ranges of possible alterations that can be applied on video signals. So detection of video forgery has become a critical requirement to ensure integrity of video data. We examined currently available image and video forgery detection methods and introduced our own methods. This thesis presents our efforts to understand and improve video forgery detection methods. Firstly, a supervised image splicing detection method, which makes use of statistical moment features, is extended to find forgery in videos. Its limitations are discussed and a subset of those features is used for our next section of unsupervised methods. An unsupervised video forgery detection method based on suitable subset of statistical moment features and normalized cross correlation factor is proposed. The location of duplicated block is also found using the algorithm. Its advantages, limitations and comparisons with existing methods is also given. Finally, a semi-supervised video forgery detection method, which is based on feature similarity index, is also proposed. The index, primarily made for image quality assessment, is used for finding duplication among the frames in a video sequence. Experimental results for the three methods are discussed. Detailed literature review and the possible future work on the related research area are also discussed in the thesis.

Master of Engineering (SCE)