영상 기반 3차원 얼굴 모델링을 위한 측면 얼굴 특징 검출

초록

In this paper we present a novel algorithm to detect profile facial features. We concentrate on the extraction of profile curve in a color image, which are important but receive little attention so far. The task is solved in several steps. First, the profile curve is extracted. Then we perform fiducial points localization on the detected profile curve. And two fiducial points are used to normalize the scale and orientation of the input image. By exploring the different image information such as skin color, edge and area segmentation, the algorithm is fully automatic, and achieves robust and accurate results for a variety of illumination, hair shape, gender and race.