Automatic Segmentation Of Head-and-Shoulder Image And Video

As is well known, segmentation is a fundamental task in computer vision. As a popular tech-nique, it extracts objects of interest from images or video sequences. Two main approaches of segmentation can be summarized as:interactive and automatic. The first approach with signifi-cantly advanced research has already been published a lot in recent years. Nevertheless, it is still a challenging task for fully automatic segmentation due to the inherent difficulty and ambiguity in complex scenes.Automatic segmentation without any user interaction is very difficult due to potentially high complexity of the scene. No wonder, most existing segmentation algorithms are based on user interactions. However, automatic segmentation in some special situations has great significance. For example, network conference, video chatting, phone camera, and so on. In this thesis, the main research work is related with automatic segmentation for head-and-shoulder image and video, and includes the following contents:1) A brief survey for previous research works of segmentation, and giving some discussion and evaluation for recent segmentation algorithms on video and image. Advantages and limits of algorithms are presented in the first part.2) Introduces an automatic segmentation algorithm for frontal head-and-shoulder images. Our algorithm combines edge feature and shape prior to extract the foreground silhouette automatically. The novelty of our approach lies in two aspects, namely, the Cost Path Segmentation (CPS) algorithm to extract the initial foreground silhouette, and a general active prior shape model, to extract the final foreground segmentation. We demonstrate the high quality and performance of the proposed approach with a variety of head-and-shoulder images. Compared with previous methods, our approach is much more robust for images with complex color distributions in foreground and background.3) Automatic segmentation of images is more difficult than automatic segmentation of video. The main reason lies on more feature cues can be used in video segmentation. We presents an automatic segmentation of head-and-shoulder video. The proposed approach can initialize auto- matically, and detect error frames when scene mutation. Our video segmentation algorithm is based on motion estimation for key frames, combines color model to finish segmenting for all frames in video.4) Concludes our work in this thesis, and presents some limits which can be studied in the future.