基于样例的表情生成方法研究

【作者】 赵新；

【导师】 陆耀；

【作者基本信息】 北京理工大学 ， 生物医学工程， 2010， 硕士

【摘要】 基于样例的表情生成是指利用一对样例表情人脸,通过表情计算生成一致的目标表情人脸。通过基于样例人脸的表情生成方法研究,探讨人脸建模、表情映射和纹理贴图等关键技术,对深入研究逼真的表情合成方法具有重要的意义。此外,表情生成方法在智能人机交互、计算机动漫、数字娱乐等领域具有广泛的应用价值。?论文的主要工作和创新如下:?首先,本文采用CANDIDE‐3模型作为标准的人脸网格模型。通过基于OpenGL的特征点拾取技术的人机交互方式对CANDIDE‐3三角网格模型上顶点的进行调整后,实现了CANDIDE‐3标准模型与特定人脸的匹配,最后得到了一个符合特定人脸特征的三角网格模型。?其次,对样例人脸的三角网格模型进行倾斜度校正等预处理后,本文实现的基于样例的表情映射算法把样例人脸的表情运动数据,映射到目标人脸网格模型上,计算出目标人脸相应表情的三角网格模型。由于表情映射后网格模型上存在一些奇异点,本文对这个网格模型添加了两个约束,一是对模型上网格点的运动范围进行约束,二是依据人脸的对称性对网格模型上的网格点添加对称性约束。?第三,基于OpenGL纹理映射技术的图像变形算法,对表情映射生成的三角网格模型进行纹理贴图。本算法依据目标人脸的中性表情图像、中性表情对应的人脸三角网格模型和表情映射后的三角网格模型数据,得到了目标人脸的逼真的不走样的表情图像。?最后,本文对所提出的表情生成算法的相关步骤逐步进行了一系列实验,最后对整体算法给出了综合实验和交叉验证实验,验证了算法的有效性。?更多还原

【Abstract】 Example-Based Expression Analogy can be defined as: given an neutral expression image and its another expression image, expression blending retargets the sample’s expression motion data to another neutral expression image and gets its another expression image.My work, example-based expression analogy, mainly including human face modeling, expression transformation, texture mapping, is of significance to purse further study on expression morph. Moreover, expression blending plays an important part in Intelligent Human-Computer Interaction, Computer Animation and Comics, entertainment and so on.The work presented in this paper is as follows.Firstly, I choose CANDIDE-3 triangle mesh as standard face model. With the aid of graphic toolkit OpenGL’s picking method, I move every vertex of CANDIDE-3 triangle mesh combining with the procedure of 2D facial modeling through Human-Computer Interaction. I get a triangle mesh of the target face model matching his face picture finally.Secondly, after preprocessing the example face triangle mesh, I suggest a new expression blending algorithm based on the example model’s facial expression. It calculates the expression motion data of sample model and retargets it to the target model’s facial triangle mesh. And then, I adjust the triangle mesh of target face model based on two hypothesizes—every mesh vertex has its own motion zone and the human face is symmetric.Thirdly, so as to apply textures to target model’s facial triangle mesh, I put forth my image morphing algorithm based on OpenGL texture mapping technology. On the basis of target model’s neutral facial image, its triangle mesh of its neutral facial image, and its another expression triangle mesh, I can get target model’s anther facial expression which is vivid and anti-aliasing.Finally, reference to every step of my expression blending method I do some experiments step by step to verify the validation of my algorithm.更多还原