CASIA OpenIR  > 毕业生  > 博士学位论文
快速个性化人脸建模和动画的研究
其他题名Research on Fast 3D Individual Facial Modeling and Animation
张满囤
学位类型工学博士
导师王阳生
2005-06-01
学位授予单位中国科学院研究生院
学位授予地点中国科学院自动化研究所
学位专业模式识别与智能系统
关键词个性化人脸建模 标准模型检索 深度信息估计 表情动画 Individual Facial Modeling 3d Head Model Retrieval Depth Information Estimation Expression Animation
摘要个性化人脸造型和动画一直是当前研究的热点和难点。本文以“快速个性化人脸建模和动画”为主题,采用普通摄像头获取图像,围绕着其中的人脸特征点获取、特征点深度信息估计、标准模型检索和个性化人脸动画生成展开了研究。主要研究工作及贡献可以归纳如下: 1) 提出了基于两张垂直照片的个性化人脸建模方法。通过建立人脸特征点模板,来快速编辑特征点,从正面和侧面图像上获取了人脸的关键特征点的三维位置,然后对标有对应特征点的标准模型进行径向基函数变形,最后进行纹理拟合,获得了真实的个性化人脸模型。 2) 提出了利用一种改进的主动形状模型(ASM)方法,RealBoost-Gabor ASM(RG-ASM)方法,从视频中快速提取人脸特征点的XY信息;然后对事先建立的三维人脸数据库,利用人脸模型特征点分布的先验知识,采用最小平方误差(MMSE)算法,建立起人脸正面参数和侧面参数的对应关系,在计算出了演员的人脸正面参数后,可以得到演员人脸的侧面参数,从而估计出特征点的深度信息;最后利用人脸特征点的三维信息去变形从模型库中检索出的标准模型、进行纹理拟合,生成个性化人脸。 3) 提出了一种新的基于几何度量的三维人脸模型检索方法,对事先建立的男女人脸库中的标准人脸模型,计算它们的水平和垂直距离度量参数,并建立相应的水平比例参数和垂直比例参数来描述人脸几何形状,然后根据我们提出的相似性度量准则,可以从标准模型库中检索出和演员的人脸形状最接近的标准模型,从而减少了人脸变形的误差。 4) 在个性化人脸模型的基础上,我们提出了我们的表情克隆算法。它是通过对面部表情的特征点相对位置的变化得到特征点的变形系数,其他顶点的位移由径向基函数(RBF)插值得到的。把一个标准模型的表情动画,快速克隆到了个性化人脸上,实现了个性化人脸的表情动画,快速实现了不同性别的人脸真实再现和表情动画的过程。
其他摘要Fast individual facial modeling and animation is one of the most challenging and interesting research topics. In this paper, we present a fast 3D individual facial modeling and animation algorithm. With the ordinary camera, our research focus on extracting facial feature points, estimating depth information for the facial feature points, 3D head model retrieval and individual facial animation. Our main work and contribution are as follows: 1) Firstly, we present an image-based 3D facial modeling algorithm based on two orthogonal images. The proposed algorithm mainly has two steps. By establishing the feature point template for 3D face structure, we can quickly obtain facial key feature points from the frontal facial image and the lateral facial image. Then corresponding to the generic model we deform the feature points of the input images by Radial Basis functions (RBF). Texture mapping is based on the different directional projections to photo-realistically rendering 3D facial model. 2) Secondly, we propose the improved Active Shape Model (ASM) approach, namely the RealBoost Gabor ASM (RG-ASM) approach to quickly extract the frontal facial feature points from video stream. Then we construct the corresponding relation between the facial frontal parameters and the lateral parameters according to the prior knowledge of facial feature points of 3D head model by the Minimum Mean Square Error (MMSE) criteria. After obtaining the frontal parameters, we can calculate the lateral parameters and estimate the depth information of facial feature points. At last we deform the retrieving generic model by 3D facial feature points and do the texture mapping to generate the individual facial model. 3) Thirdly, we present the 3D head models retrieval algorithm based on geometrical measurement. This proposed method mainly has two steps. Firstly we separately create a male and a female dataset of generic models and gain their feature points which are used to identify horizontal and vertical measurements. Then we may construct horizontal proportions and vertical proportions to represent the facial geometrical shape. After extracting the individual frontal facial feature points, the shape which is mostly similar to the performer’s shape in dataset can be retrieved according to the similarity criterion. Secondly these feature points are used to deform the chosen generic model with corresponding feature points by Radial Basis functions (RBF) to reduce deformation error. Also the texture mapping can make the facial model more realistic.
馆藏号XWLW0
其他标识符200218014603235
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/5872
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
张满囤. 快速个性化人脸建模和动画的研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2005.
条目包含的文件
条目无相关文件。
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[张满囤]的文章
百度学术
百度学术中相似的文章
[张满囤]的文章
必应学术
必应学术中相似的文章
[张满囤]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。