SGER: Developing a high-definition face modeling system for recognition and generation of face and face expressions

人脸分析和合成在人机交互、安全和娱乐等领域的应用，已经成为科学界和工业界一个取之不尽的研究课题。用于识别或动画的最新技术的3D面部模型不能表示和识别面部外观或特征配置中的细微差别，这是在个体之间唯一区分所必需的。在这个项目中，PI将开发一种新的技术，以比目前可能实现的更详细的水平分析和合成人脸。提出了一种用于人脸表面标注和建模的三维地形表示方法，其中人脸由一个基于曲面曲率和地形标签的可变分辨率的活动网格来表示。PI期望在多个尺度上跟踪某些特征的行为将揭示关于潜在物理过程的性质的宝贵信息，从而导致在面部特征和表面属性(即，峰、坑、脊、沟、鞍座、山、平等)之间建立内在关系。这反过来将以直观和详细的方式提供对面部外观和表示如何随表情变化而变化的科学理解，并将为3D面部表面如何包含2D平面图像中未展示的嵌入特征提供深刻的新见解。PI预计，这项试点研究最终将使他能够开发出一种识别人脸及其表情(甚至是情感)的“人性化”系统，以及一个自动生成逼真面部表情的系统。广泛影响：这项研究将提高我们对人类视觉认知过程的理解，并影响机器可以识别面部表情的下一代人机界面的发展。这项工作还将在执法(面部照片搜索、安全监控和测谎)、访问控制(生物识别身份验证)、娱乐(电影、媒体和游戏)、视频会议(包括支持唇读)和医学(颅面整形手术规划和化妆品)中有重要的应用。

Face analysis and synthesis has become an inexhaustible research topic in both the scientific community and industry, driven by applications in human-computer interaction, security, and entertainment. State of the art 3D face models used for recognition or animation cannot represent and identify subtle differences in facial appearance or feature configuration necessary to uniquely discriminate among individuals. In this project, the PI will develop a novel technique for analysis and synthesis of the human face at a higher level of detail than it is currently possible to achieve. A so-called 3D topographic representation is proposed for labeling and modeling the facial surface, in which the face is represented by an active mesh with a variable resolution based on the surface curvatures and the topographic labels. The PI expects that tracing the behavior of some features across multiple scales will reveal precious information about the nature of the underlying physical process, and thus lead to establishing an intrinsic relationship between facial features and surface properties (i.e., peak, pit, ridge, ravine, saddle, hill, flat, etc). This will, in turn, provide a scientific understanding of how facial appearance and representation vary along with expression changes, in an intuitive and detailed manner, and will offer profound new insights into how 3D facial surfaces contain embedded features that are not exhibited in 2D planar images. The PI anticipates that this pilot research will eventually enable him to develop a "humanized" system for recognizing human faces and their expressions (even emotions), as well as an automatic system for generating life-like facial expressions.Broader Impacts: This research will improve our understanding of the cognitive processes involved in human vision, and impact the development of next generation human-computer interfaces in which machines can recognize facial expressions. The work will also have important applications in law enforcement (mug shot searches, security surveillance, and lie detection), access control (biometric authentication), entertainment (films, media, and games), videoconferencing (including support for lip-reading), and medicine (craniofacial plastic surgery planning, and cosmetics).