| 题名 | 基于人脸特征和面部运动单元的卡通生成系统 |
| 作者 | 车建峰 |
| 学位类别 | 工学硕士 |
| 答辩日期 | 2011-05-30 |
| 授予单位 | 中国科学院研究生院 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 王欣刚 |
| 关键词 | 人脸特征 卡通人脸 面部运动单元 Delaunay三角 纹理映射 Facial feature Cartoon face Action Units Delaunay triangulation Texture mapping |
| 其他题名 | Facial Feature and Action Units Based Cartoon System |
| 学位专业 | 计算机应用技术 |
| 中文摘要 | 随着数字媒体和动漫产业的不断发展，在某些情况下人们已经不再满足于使用真实照片，而是追求真实照片的卡通化。如何利用计算机将已有的真实人脸图片转变为具有卡通效果的人脸图片日益受到广泛关注。在实际的人脸照片中，真实人物的表情可能表现为中性、高兴、悲伤、厌恶、惊讶、生气和害怕等基本表情，但更多的可能是表现出非上述基本表情的“亚表情”。为了能更好地抓住真实人脸所表现出来的情绪状态，Paul Ekman等人提出面部运动编码系统(FACS)，该系统将人脸的肌肉运动划分为41个运动单元，通过41个运动单元和其组合可以表达出多种复杂的表情。 本文通过对输入的真实人脸图像的分析、识别，提出一种基于人脸特征和面部运动单元的卡通生成系统，该系统将输入的正面真实人脸图像自动转化为保持真实人脸特征的卡通人脸图像；系统同时可以识别输入真实人脸的面部运动单元，并根据识别结果生成具有相应运动单元的卡通人脸。具体地，本文完成的工作和取得的研究成果主要包括以下4个方面： （1）分析不同质量的人脸图像，找到较为鲁棒的人脸特征点定位技术，从模型训练和算法性能的角度提高定位技术的准确性和适用性。 （2）分析、归纳得到人脸表现出各种表情、亚表情时频繁出现的面部运动单元(AUs)，找到可以较为鲁棒地识别出归纳得到的面部运动单元的模式分类方法。 （3）利用计算机图形学中Delaunay三角面片的方式做卡通纹理的渲染，同时分别利用图像贴图和曲线绘制的方式为初步得到的卡通人脸添加头发和耳朵。使用不同的模板，系统可以生成不同风格的卡通纹理以及发型。 （4）对于具有非中性表情的人脸图像，系统可以识别其表现出来的运动单元并最终利用卡通生成算法生成具有相应运动单元的卡通人脸。 |
| 英文摘要 | With the development of digital media and animation industry, in some cases, people are not satisfied with using real image, but pursue cartoon of the real one. How to turn real facial photos to cartoon ones is receiving common attentions. In the real facial photos, the person’s expression can be one of the prototypic emotional expressions, i.e., neutral, happy, grief, disgust, surprise, angry and fear, however, such prototypic expressions occur relatively infrequently, sub-expressions are occurred more frequently. In order to grasp the emotion state of real face better, Paul Ekman et al develop the Facial Action Coding System (FACS), which describes facial expressions by 41 action units. These 41 AUs and AU combinations can express a variety of complex expressions. In this paper, based on analysis and recognition of the input real facial image, we propose a facial feature and action units-based cartoon system, which can generate a realistic cartoon face form an input image; The system can also recognize action units of the input one, and generate a cartoon face that owns the specific action units. Specifically, in this paper, the work that has been finished includes following four aspects: (1) Analyze facial images of different quality, find the robust facial feature points locating technology, and improve the accuracy and applicability of locating technology in model training and algorithm performance. (2) Analyze action units that prototypic expressions and sub-expressions display, access the frequently occurring action units, and find pattern classification method that can recognize above action units robustly. (3) Utilize Delaunay triangulation in computer graphics to do cartoon texture mapping, and make use of image mapping and curve drawing to add ears and hair to cartoon face. The system can generate different cartoon texture and hair by different cartoon texture template and hair-style template. (4) For a non-neutral expression face image, the system can identify the action units of its performance，and generate cartoon face that owns the specific action units by cartoon generation algorithm. |
| 语种 | 中文 |
| 其他标识符 | 200828014628026 |
| 内容类型 | 学位论文 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/7561]  |
| 专题 | 毕业生_硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 车建峰. 基于人脸特征和面部运动单元的卡通生成系统[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2011. |
| 个性服务 |
| 查看访问统计 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
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。
修改评论