【作者】 孙凯；

【导师】 徐勇；

【作者基本信息】 哈尔滨工业大学 ， 计算机科学与技术， 2010， 硕士

【摘要】 人类对自身之美的研究已有几千年的历史了,人类对美的不断追求,促进了人的审美意识不断提高,同时也促进了社会的发展,而面部美学又是其中的研究重点。国内外很多学者对于美及其规律也得到出了一些初步结论,但是传统的人脸美学研究由于其主观性、手动化等缺点严重阻碍了其理论发展和实际应用。本文基于传统的医学美学与理论美学提出了计算人脸医学美学概念,将面部美学研究量化与自动化,并希望其在临床美容,整形与修复当中具有指导意义。首先,本文基于传统的人体测量学和解剖学对人脸几何形状进行了标准化定义,即对描述人脸几何特征的特征点给出标准化定义,并通过稳定性分析对其数量选择给出了一系列准则,从而将底层误差限制在最小范围内。其次,本文使用主动轮廓模型和改进后的主动形状模型对面部特征进行提取,前者虽然基于可变模板,但是以牺牲模板的鲁棒性来换取可变性;后者训练时间过长,而且搜索过程中易受目标结构周围物体影响。因此本文使用改进后的主动形状模型进行面部特征点的提取,将提取出的特征点构造成一个更具有视觉感知意义的面部美学向量。然后,使用支持向量回归技术建立面部美学向量与人脸美丽指数之间的回归模型,对于新的图像使用该模型求与其面部美学向量相接近但具有更高美丽指数的目标向量。最后,使用移动最小二乘图像变换技术,并以原始面部美学向量和其目标向量为控制条件,对面部特征进行变换,从而达到美容,整形和修复等目的。更多还原

【Abstract】 Research on human’s esthetics has a history of several thousand years, in which facial esthetics is its essential focus. But traditional facial esthetics research has some inherent shortcomings, such as subjectivity during the research process, time consuming and so on, all of this prevent the theoretical progress and practical application of this field. So in this paper, based on Medical esthetics and traditional theoretical we propose the concept of Computed Facial Medical Esthetics, aiming at quantifying and automating this research, we also hope that the result can be used to guide clinical application, such as cosmetology, plastics surgery and so on.First, we give a comprehensive definition for feature points which are used to depict facial features.As for the number of these points we use an optimal selection algorithm by analyzing their stability. So we can reduce the low-level inaccuracy.Secondly, we aim at extracting these feature points automatically, there are some algorithms based on flexible models, of which the active contour models and active shape models has the best performance. We analyze them and make some improvements, using these algorithms we can enlarge our research on large amount of samples.Thirdly, after obtaining these feature points, we do not use them directly; instead we construct a geometrical facial beauty vector of visual sensation significance, which consists of outline information and the facial parameter information. Then we obtain a model between this vector and beauty indices by support vector regression technology.Then for a given facial image, we can extract its feature points, construct GFBV and use SVR and optimization to get a target vector, which is near to this GFBV but has higher beauty index andt map to facial feature points.Finally, using these target feature points and original feature points we can implement image deformation, which can produce a target face with higher beauty index. The result can be used to guide cosmetology, plastics surgery and so on.更多还原