The Research For Human Face Detection And Modeling

As the development and maturation of computer technology and the progress ofdata measurement technology, the research on detecting and modeling human face hasbeen a typical, fresh and up-spring domain in reverse engineering due to its complexgeometric figuration and varying curvature. The methods of detecting and modelinghuman face can be classified into some categories according to different dataacquiring modes. So far there have been many reports about this issue all over theworld, but the product, which can satisfy the most demand of applications, still do notcome into being. In this paper we will discuss not only the face detection methods butalso the face modeling technology based on 3-D measurements. Firstly, we have donesome research about the face detection methods. As the foundation of facereconstruction methods which are based on images, it is one of difficulties facing usthat how to obtain the characteristic information that we need from a pictureautomatically. And completely automatic face detection and recognition technique,that can be applied widely, is still an open scientific research subject all over theworld. Since many factors, such as the different background, complex head geometricfigure and diverse decorations, are all possible to affect the effectiveness of facedetection and recognition, in this paper we design a fully automatic face detectionsystem to solve our own practice demand, which is based on the idea of complexionGauss model, image segmentation and region merger. This system can detect the faceregion in a middle-complex picture automatically and validate the detect result bysome face characters, which not only makes the system more robust but also improvesthe efficiency of our detection algorithm. Secondly, we will introduce a threedimensional imaging and modeling system (3DIMS) which is designed based onfringe pattern analysis technique. This system takes an acquisition of encoded fringepattern from different angles, and then follows a series of automatic fringe patternanalysis to extract the spatial phase distribution that is proportional to the range imageof tested object surface. Furthermore, the system performs the registration of partialrange images acquired from different views and the integration of those partial rangeimages in order to establish a complete, non-redundant geometrical representation.However, the image obtained by this method still can not satisfy the demand ofcomputer animation, etc. because of the big volume of data and bad image quality. Sowe introduce the technique of implicit surface, mesh simplification and subdivisionsurface to the post-processing of 3DIMS's output, which can not only reduce the datavolume but also improve the image quality.