Chinese Name:蔡依婷
English Name:I-Ting Tsai
Job:software engineer
Grades: Master Students



Group in CVLAB: Learning group
Thesis:Automatic Human Segmentation for Background Substitution
In this thesis, we propose an automatic human image segmentation system for background substitution. Since human has translucent edges in hair region in reality, our framework is based on automatic segmentation of human image into the corresponding trimap, which is composed of the foreground, background, and unknown regions. Then an image matting technique is applied to achieve a naturally-looking image synthesis. In offline training, we collect a training dataset of human images to establish a probabilistic body mask. For a test image, we first segment the skin and hair regions based on the color distortion to the corresponding training regions, which are determined by the face and eye detection. To segment the body and background regions, we propose a modified EM parameters estimation method for Gaussian Mixture Models with weighted data. A mixture reweighting scheme is also used to refine the result of body/background segmentation. We demonstrate the effectiveness of the proposed background substitution system through experiments on some real images.