Detecting doctored video footages using deep neural networks

Innovations in deep learning has already accelerated the developing eld of
computer vision. Convolutional neural network (CNN) is a newfangled deep
learning tool that learns high level features directly from a huge data set of
labelled images. With break neck advances in digital information processing
systems, and more specically in digital image processing software, there is
a far
ung development of advanced tools and techniques for digital image
forgery as well as video and audio forgery. The prevalent copy-move forgery
methods mostly make use of hand-crafted features which imposes stipulations
on the performance of copy-move forgery detection. This report presents a
video forgery detection method based on deep learning technique which utilizes
a convolutional neural network (CNN) to automatically learn the stratied
frames in a video which represents the input RGB color images. The
proposed CNN is specically designed for image splicing and copy-move detection
applications. This method proposes to perform forgery detection by
means of deep learning using an architecture based on ConvNet. A training
phase on a few pristine frames allows the ConvNet to learn an intrinsic model
of the source. This work focuses on a deep learning approach containing network
layers to concentrate on the in-depth properties of the input RGB video
frames. By conducting experiments on challenging video data sets containing
fast camera motion and forged videos, the experimental results clearly
demonstrate the adequacy of the proposed approach.

MSC CS 2017-2019 INT Pradeep Kumar K