Research On Image Foreground Extraction

Foreground extraction and image compositing are fundamental operations in image processing, and they have become crucial and frequently used operations in visual effects production. In addition, matting and compositing are used in much of today's media, including magazines, 2D arts and graphics, television and advertising etc.According to assumption of whether there are semi-transparent pixels; foreground extraction can be divided into two categories: Forground Segmentaion and Foreground Matting. In this paper, after a brief introduction to the research background of Foreground Extraction, we first summarize the state of the arts of Foreground Extraction and Image Compositing; then we analyze and compare the pros and cons of several algorithms. Our main contributioins are as follows:We study and implement Lazy Snapping algorithm, an interactive image foreground cutout tool. Instant feedback is made possible with a novel image segmentation algorithm which combines graph cut algorithm with watershed algorithm. A set of user interface is implemented to provide flexible control and editing for the users.We study and implement Robust Matting algorithm, which combines sampling-based method and propagation-based method, whereas avoid their limitations as much as possible. We use optimized color sampling to sample foreground and background samples for unknown pixels, but most importantly, we compute confidence for these samples. Only high confidence samples are chosen to estimate matte. Then we use a Random Walker to optimize the matte. Experimental results have proved superiority of our approach over previously proposed algorithms.We also develop a foreground extraction application system, which integrates several foreground segmentation and foreground matting tools. Our system requires users to specify an original image, an optional trimap image and an optional new background image. Users can also create a trimap manually. Our system integrates Lazy Snapping, Bayes Matting, and Robust Matting currently. We expected our system to be a very useful tool.