EAGER: RI: Developing a Framework for Automated Measurement of the Intensity of Spontaneous Facial Expressions

自发的面部表情是日常生活中面部表情的典型代表。研究自发(非姿势)面部表情和面部动作编码系统(FACS-AU)需要自发的面部表情数据，其中动作单元是手动且可靠地编码的。目前，计算机视觉社会对这样的数据库有很高的需求。这项研究的第一个目标是获取一个3D面部表情数据库，并将其公开供其他对研究自发面部表情感兴趣的人使用。作为基本事实，我们还手动编码了此数据集中重要行动单位(即AU 1、2、4、6、9、10、12、14、15、20、25和26)的存在和强度。这些动作单位描述积极和消极的情绪表达。这项研究的第二个目标是开发一个框架，自动测量从成人捕获的视频中自发的FACS动作单位的强度。面孔。我们的目标是研究基于字典的表示(例如，小波和Shearlet的组合)来对人脸图像中的人脸外观进行建模。这是因为人脸图像包含边缘、皱纹和纹理。我们研究了这种类型的组合表示，以提供包含边缘和纹理的图像的稀疏表示，并将其用于测量AU强度。本研究的成果、人脸数据库、AUS的手动编码以及所开发的AUS测量框架，可供计算机视觉学会的其他研究人员公开使用。

Spontaneous facial expressions are representative of facial expressions in daily life. Studying spontaneous (non-posed) facial expressions and Facial Action Coding System-AUs (FACS-AUs) requires spontaneous face expression data where the action units are manually and reliably coded. Currently, there is a high demand in computer vision society for such a database. The first objective of this research is to capture a 3D facial expression database and make it publically available to others interested in studying spontaneous facial expressions. As ground-truth, we also manually code the existence and the intensity of important action units (i.e., AU 1, 2, 4, 6, 9, 10, 12, 14, 15, 20, 25, and 26) in this dataset. These action units describe positive and negative emotional expressions. The second objective of this research is to develop a framework to automatically measure the intensity of spontaneous FACS action units in videos captured from adults? faces. We aim to study dictionary-based representation (e.g., combination of wavelet and shearlet) to model face appearance in facial images. This is due to the fact that human facial images contain of edges, wrinkles and texture. We study this type of combined representations to provide a sparse representation of images containing edges and textures and utilize it in measuring the AU intensities. The results of this research, the face database, the manually codes of AUs, and the developed framework in the measuring the AUs, are made publically available to other researchers in computer vision society.