CASIA OpenIR  > 毕业生  > 硕士学位论文
真实感人脸建模和动画研究
其他题名Lifelike Facial Modeling and Animation
李永林
学位类型工学硕士
导师陶建华
2006-06-04
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
关键词个性化真实感人脸建模 真实感人脸动画 可视语音合成 Photo Realistic Facial Modeling Behavior Realistic Facial Animation Talkinghead
摘要三维几何建模和动画是当前活跃的研究方向,在计算机图形学和计算机视觉领域都受到了广泛的关注,它在人机交互、计算机动画、计算机游戏、虚拟现实、身份识别、多媒体内容传输编解码中扮演着非常重要的角色。本文以真实感人脸建模和动画为主题,探讨了该主题内的两个关键问题,个性化真实感建模和真实感人脸动画。围绕普通二维人脸图像的三维建模、网格自适应变换的人脸建模、MPEG-4兼容的真实感人脸动画等问题进行了初步探讨。 本文的主要工作包括: 提出了一种新的基于正交图像的个性化真实感人脸建模方法,通过对三维通用模型进行投影匹配正交人脸图像的方法,得到和正交图像精确匹配的投影网格,对之进行三维重建得到个性化的三维人脸模型。采用多尺度纹理映射技术得到真实感三维人脸模型。建模方法有效地降低了计算复杂度,增加了静态真实感。 提出了一种基于网格自适应变换策略的人脸建模和动画的方法,根据不同的平台要求,对于需要进行网格密集化的场合,采用三维扫描仪得到的点云输入数据,利用特征点匹配,采用多精度拟和策略得到密集化后的个性化人脸模型;对于需要进行网格简化的场合,采用改进的边折叠删除算法得到简化后的个性化人脸模型。利用DFFT插值技术、信息保留的拓扑变换策略对变换模型进行人脸动画。 在上述个性化真实感人脸建模的基础上,基于MPEG-4,对真实感人脸动画进行了研究,结合语音合成的基础,对真实感可视语音合成Talkinghead进行了探讨,利用语音合成得到的结果,基于MPEG-4的动画规定,实现了一个具有表现力的可视语音合成系统。 在以上三点研究的基础上,针对个性化真实感人脸建模和动画的整个过程开发出了人脸建模和动画的实用平台,并在嵌入式平台上实现了多个人脸建模和动画的工具。结合语音合成实现了一整套的人脸动画、可视语音合成的应用系统。
其他摘要Three-dimensional human face modeling is an active research topic in both computer vision and computer graphics area. It plays important, even vital role in a lot of applications such as natural human-computer interaction, computer animation, computer games, model-based coding, virtual reality, face recognition. The key issues mainly focus on the two main topics in lifelike facial modeling and animation, photo realistic facial modeling and behavior realistic facial animation. The content of this paper include photo realistic facial modeling from two orthogonal face images, self adaption of the mesh deformation for facial modeling and animation, MPEG-4 compatible behavior realistic facial animation. The main contributions of the paper are: First, we proposed a new way in photo realistic facial modeling technique from two orthogonal face images, based on the projection of the three-dimension generic face model, we have made the match between the face images and projective mesh exactly. Construction of the three-dimension face model comes directly after precise matching. After the multi-dimension texture mapping, the face model has shown a photo realistic look. The algorithm has reduced the computation cost and enhanced the static realism effectively. Second, we proposed a new way for facial modeling and animation from self adaption of the mesh deformation. According to the different demand of the application and platform, for those where need more detailed mesh, take the data from 3d scanner as input, we use feather points matching and multi precision matching method to get a personalized face model. For those where need predigested mesh, take the original generic model as input, we use edge contraction method to get the personalized face model. DFFT and information saving topology transform to generate facial animation. Third, based on the photo realistic facial modeling and voice synthesization, we proposed a new way for one kind of facial animation----talkinghead, because the animation is MPEG-4 compatible, we have made a way for getting the facial animation parameters from motion capture data. Fourth, based on all the research above, we have applied some new software platform related to facial modeling and animation. Some toolkit and SDK are also presented according to our research. At the meanwhile, we also present some system like facial animation in embedded platform.
馆藏号XWLW1016
其他标识符200328014604136
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/7388
专题毕业生_硕士学位论文
推荐引用方式
GB/T 7714
李永林. 真实感人脸建模和动画研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2006.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20032801460413(2198KB) 暂不开放CC BY-NC-SA请求全文
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
[李永林]的文章
百度学术
百度学术中相似的文章
[李永林]的文章
必应学术
必应学术中相似的文章
[李永林]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。