A novel approach to video matting using automated scribbling by motion analysis

Abstract

Video Matting can be described as a method of extracting the foreground object from each frame of a video and composite (or transfer) it to a different background or a different video. An automatic scribbling approach based on the motion of the foreground object in a video is introduced. Probability maps (PMs) are computed by analysing the motion of object in the video. In the proposed technique the corrupted PMs are refined and reconstructed using morphological techniques. The frames are scribbled with reference to the reconstructed PMs. Closed form solution based matting technique uses the scribbled frame to yield the matte and the foreground.

Video matting by automatic scribbling using Quadra directional filling of segmented frames

“…After that probability maps (PM) are constructed that distinguishes foreground and background. Automatic scribbling [1] is done on each frame with reference to its probability map. This scribbled frame is then fed into the closed form solution [2], to get the alpha matte.…”

“…We propose a matting technique that is based on the matting method proposed by Abhishek et al [1] which scribbles the frames automatically by referring to their constructed probability maps and then closed form solution which is proposed by Levin et al [2] is applied to get the frame with new background. The disadvantage of their method [1] is that they use motion analysis for determining the probability maps which is time consuming and it also requires extensive refinement and reconstruction of the probability maps which increases its complexity and chances of error are increased.…”

“…The disadvantage of their method [1] is that they use motion analysis for determining the probability maps which is time consuming and it also requires extensive refinement and reconstruction of the probability maps which increases its complexity and chances of error are increased. In contrast, our technique uses Image segmentation technique to construct more accurate probability maps.…”

Automated segmentation of gallstones in ultrasound images

“…Automatic Scribbling, proposed by Abhishek et al [5] and based on checking of the neighbouring pixel, is now put to use. Probability Maps have white colour in foreground and black in the background.…”

“…They derived cost function for grayscale images (ultrasound images in our case) with the assumption that the foreground and background remain steady over a small area around each pixel. Here, automatic scribbling method, proposed by Abhishek et al [5], is used in place of manual inputs, to provide scribbles. Fig.…”