Automatic human-head and shoulder segmentation of frontal-view face images

Object segmentation is one of the basic issues in image processing and computer vision. However, especially human-head and shoulder segmentation is a topic which was introduced only recently, gaining in importance for a wide area of computer vision applications, such as testing compliance for ID document issuing, improving images for facial recognition or even used in the upcoming e-governmental self services and commercial sector. In this thesis we address the problem of automatic human-head and shoulder segmentation of frontal-view face images from non-uniform complex backgrounds and propose an approach composed of different subtask. These subtasks can be viewed individually and consist of a novel face skin silhouette detection approach based on supervised classification learners, a study of two state-of-the-art superpixel algorithms in relation to the specified problem statement and a novel hair and shoulder segmentation approach. We discuss and evaluate our methods and present competitive results for each subtask.