Automatic Human Upper Body Cutout From Image And Video

Foreground segmentation is among the most challenging and researched fields in Computer Vision and Computer Graphics. So far, most algorithms have been proposed with an interactive working mode. In the present thesis, a novel algorithm of automatic foreground extraction for special object is presented and its effectiveness with human upper body images and videos is then verified. The main idea is to make use of the prior knowledge to guide the segmentation. No user interaction is required.For human upper body images, we first detect face and a few facial features, which helps to estimate a shape model of the body. The algorithm then creates a color model by sampling the colors near the shape model's outline. Based upon the information derived from the previous two models, the hard edge of foreground is then obtained by an iterative graphcut algorithm.For video sequences, we make use of the continuity among adjacent frames to further constrain the segmentation and get more accurate results. The process can be carried out in realtime and a maximum of 15 frames could be handled per second using the present algorithm.Our main contributions lie in the following parts:? We presented a novel algorithm for automatic foreground extraction of human upper body in images, which demonstrated the feasibility of automatic segmentation of special objects.? We presented a realtime algorithm for human upper body extraction in videos, which is more practical than existing algorithms.? We developed the GMM color model and proposed an upper body shape model to guide the automatic segmentation, which can be extend to other special objects of interest.Human face and body appear most frequently in images and videos thus making the accurate and automatic extraction of them very practical and mean-ingful. The results of our work can be further used for NPR rendering, background replacing and foreground composition. With the advent of increasing number of digital devices, the present work is of great significance for those with embedded chips and without convenient interactive tools, such as digital cameras, mobile phones as well as PC cameras.