Abstract: This paper proposes a method for extracting facial feature points automatically using deformable templates whose shape and brightness are changeable. Also, we show adjustment of 3D shape model on facial images using the extracted feature points. The deformable templates are constructed with a mean facial image. The mean facial image is obtained by adjusting 3D shape model on 50 facial images and meaning those. There are the templates for whole face and 6 kinds of facial parts (right eyebrow, left eyebrow, right eye, left eye, nose and mouth). The whole face templates have 7 sizes and 7 levels of brightness for each size. The templates for each facial part have 7 sizes. This algorithm begins from searching face region with the whole face templates. The second step determines candidate region for each facial part by referring the searched face region. The third step searches each facial parts in each candidates region in the order of eyes, nose, eyebrows and mouth. The 3D shape model is adjusted on the facial images using the searched feature points. This method is more stable in extracting feature points than the traditional method with only shape because the shape and brightness of our templates are changeable.!17
展开▼
