COMPUTER VISION ANALYSIS OF DYNAMIC FACIAL BEHAVIOR

描述(申请人摘要)：面部表情为研究情绪、认知过程和社会互动提供了一种重要的行为衡量标准，并包含神经疾病和抑郁症的诊断信息。与EEG、EMO、ANS或大脑成像测量相比，通过视频测量面部表情作为情绪活动的指标的侵入性较小。目前，测量是由人类专家进行的。本研究的目标是开发一种利用计算机视觉技术对视频中的面部表情进行自动识别、测量和编码的系统。这个项目将利用动力系统的概率模型来模拟观察到的图像序列背后的面部行为。这些技术包括隐马尔可夫模型，以及Movellan和他的同事开发的一种新的随机建模技术，称为扩散网络[38]。与传统的动力学模型相比，扩散网络提供了允许连续时间动态和连续状态的优势。自动化系统将使面部表情测量作为行为科学和情绪神经基础调查的研究工具得到更广泛的应用。

DESCRIPTION (Applicant's abstract): Facial expressions provide an important behavioral measure for the study of emotion, cognitive processes, and social interaction, and contain diagnostic information for neurological disorders and depression. Facial expression measurement from video is less intrusive than EEG, EMO, ANS or brain imaging measurements as an indicator of emotional activity. The measurement is presently performed by human experts. The goal of this research is to develop an automatic system for recognition, measurement, and coding of facial expressions from video using computer vision technology. This project will utilize probabilistic models of dynamical systems to mod& the facial behavior underlying the observed image sequences. These techniques include hidden Markov models, and a new stochastic modeling technique developed by Movellan and colleagues called diffusion networks [38]. Diffusion networks offer the advantage over traditional dynamical models of allowing continuous time dynamics and continuous states. An automated system would make facial expression measurement more widely accessible as a research tool in behavioral science and investigations of the neural substrates of emotion.