针对表情的人脸网格划分 - 数据采集与处理.DOC

基于单张人脸图像的表情生成方法(? 马禹骞, 钟宝江+ (苏州大学 计算机科学与技术学院,江苏 苏州 215006) An Expression Synthesis Method Based on Single Facial Image MA Yu-Qian, ZHONG Bao-Jiang+ (School of Computer Science and Technology, Soochow University, Suzhou 215006, China) + Corresponding author: Phn +86-512 E-mail: bjzhong@suda.edu.cn Abstract: Facial expression synthesis is an important approach of affective computing. In this paper, we study various facial and expressions features. By abstracting specific organ changes, several principles for the mesh partition of a human face are provided. A new mesh partition technique is then introduced for the synthesis of human expression. By simply positioning, the new technique divides organs of the human faces with polynomial curves. In consequence, every facial organ can be adjusted separately. The existing definition of six basic facial expressions are simplified. With this mesh generation technique, a new method to generate facial expression is obtained. At the same time we try a step to add skin texture so that the generated expression can be enhanced. Compared with the existing facial expression synthesis methods, the proposed method is simpler in mesh partition, more flexible in parameter setting, easier to be coded and needs less computational cost. Subjective evaluation show the generated facial expressions have a good visual effect. Key words: Affective Computing; Face Morphing; Expression Synthesis; Mesh 摘 要: 人脸表情生成是人机情感交互的重要实现方式之一。通过研究人脸及表情特点，抽象具体的器官变化，归纳出针对人脸图像的网格划分法法则。在此基础上提出了一种针对人脸表情生成的网格划分技术。该技术通过简单的定位，使用二次多项式曲线合理划分了人脸主要器官。继而将其参数化，实现对人脸各器官的独立调整变动。同时简化定义了六种基本表情的面部变化。使用新网格划分技术实现不同的人脸器官变化，生成各种特定的表情。同时尝试了皱纹等皮肤纹理的添加步骤。与已有的人脸表情生成方法相比，新方法具有网格划分方式简单、变动参数多样、易于编码实现和计算量小等优点。最终生成的人脸表情在主观评测中评价良好。 关键词: 情感计算；人脸变形；表情生成；网格 中图法分类号: TP301　　　文献标识码: A 引言 人脸表情生成是人工智能领域的一项有趣挑战，要求计算机能理解、分析和模拟人类情感，最终与人类进行情感交互。相关研究工作的一般关注点主要在人脸识别上，对人脸一部分特征的计算机识别率甚至已经高过人类本身。而随着计算机及相关学科不断的发展，人们对人工智能提出了更高的要求。其中，情感计算是近年来越来越热门的研究方向[1] ，它要求计算机不仅能识别人类情感，还能理解并做出反应，从而像人与人一样进行情感交流。以现