Video Matting, Watermarking and Forensics

The fast growth of social media networks with wide mobile usage causes the sharing of videos more frequently among the users. The wide spread of videos across the networks ends up with various video forgeries such as digital video theft, video adulteration, fake video creation and video tampering. The analysis of video for its originality should be performed in many situations like crime investigation for real-time evidence. For inspecting the video footage, video matting, watermarking and forensics techniques can be applied. Video matting is the task of extracting the foreground objects with alpha matte (opacity estimates) by removing the background video sequences. Digital video watermarking involves the addition of information in the form of a watermark at the background of videos to avoid video thefts. Video forensics is the process of examining the video in a scientific manner, assessing or comparing videos for legal matters. In this proposed work, an improved convolutional neural network based video watermark algorithm along with discrete wavelet transform and singular value decomposition are employed for embedding and extracting the watermarks in video.