Neural mechanisms underlying categorical perception in the processing of faces

人脸感知对于正常的社会互动至关重要，而人类观察者擅长快速可靠地将视觉模式归类为人脸或非人脸。即使这种能力的轻微损害也可能造成毁灭性的后果，自闭症和其他发育障碍就是明证。尽管面孔感知的神经机制一直是灵长类电生理学和脑成像研究的主要焦点，但大脑如何处理面孔的计算机制仍然远不清楚。在国家科学基金会的支持下，达特茅斯学院的孟明博士正在通过综合心理物理学、人脑成像、统计数据挖掘和计算机视觉的技术来解决这个问题。首先，这个项目正在测量大脑对使用计算机视觉汇编的图像集的反应。这些图像在图像级别的面部相似性方面各不相同，从非人脸到真正的人脸。计算机视觉系统可能会错误地将许多类似人脸的非人脸分类为人脸。相比之下，直截了当的感知使人类观察者能够对这些图像是人脸还是非人脸做出明确的感知判断。其次，要了解人脸处理神经网络，比了解每个单独大脑区域的平均反应激活更重要的是了解几个大脑区域之间的因果关系。这个项目应用最先进的数据挖掘技术，为各种因素之间的直接和间接关系的动态因果建模提供基础，这些因素包括低级视觉特征、人脸相似度和人脸/非人脸分类，将神经处理阶段从初级视觉分析连接到感知决策。最后，为了确定自下而上和自上而下对分类面孔知觉的调节，复杂的心理物理范式被用来研究视觉知觉和分类面孔分析的神经关联之间的潜在交互作用。通过这种方法，可以将被认为独立于视觉感知的刺激驱动和前馈人脸加工模型与涉及认知反馈调节和视觉感知的模型进行比较。这个项目的结果有望更好地理解人类大脑如何在脸部感知的背景下处理视觉信息，脸部感知是感官组织最引人注目的例子之一。了解人类视觉系统分析人脸的过程可能最终导致人工人脸识别系统的成功设计。此外，在人脸感知中，连续的图像级面部相似性分析与二进制分类判断是快速识别人脸的关键社交能力的基础。将这两种分析分开，可能有助于描述涉及面部感知缺陷的发育障碍的神经病理特征。根据这些结果，可以有策略地设计一种认知疗法来精确定位这些缺陷，从而帮助患有这些障碍的儿童和成年人。

Face perception is crucial for normal social interactions, and human observers excel at rapidly and reliably categorizing visual patterns as faces or non-faces. Even subtle impairment in this ability can have devastating consequences, as evidenced by autism and other developmental disorders. Although the neural mechanisms underlying face perception have been a major focus of primate electrophysiology and brain-imaging research, the computational mechanisms underlying how the brain processes faces are still far from clear. With support from the National Science Foundation, Dr. Ming Meng of Dartmouth College is addressing this question by synthesizing techniques drawn from psychophysics, human brain imaging, statistical data mining, and computer vision. First of all, this project is measuring brain activation in response to image sets that have been compiled by using computer vision. These images vary in their image-level facial similarity, ranging from non-faces to genuine faces. Computer vision systems may falsely categorize many of the face-like non-faces as faces. By contrast, categorical perception enables human observers to make unambiguous perceptual judgments on whether these images are faces or non-faces. Second, to understand the face processing neural network, it is more important to understand causal relationships among several brain regions than average response activation of each separate brain region. This project is applying state-of-the-art data mining techniques to provide the basis for dynamic causal modeling of the direct and indirect relationships among various factors, such as, low-level visual features, face semblance and the face/non-face categorization, linking neural processing stages from primary visual analysis to perceptual decision. Finally, to determine bottom-up versus top-down modulation on categorical face perception, sophisticated psychophysical paradigms are being used to investigate potential interactions between visual awareness and the neural correlates of categorical facial analysis. Through this approach, stimulus-driven and feed-forward models of face processing that are assumed to be independent of visual awareness can be compared to models that involve cognitive feedback modulations and visual awareness. The results of this project are expected to lead to a better understanding of how the human brain processes visual information in the context of face perception, a domain that provides one of the most compelling examples of sensory organization. Understanding the process of how the human visual system analyzes faces may ultimately lead to the successful design of artificial face recognition systems. Moreover, continuous image-level facial similarity analysis versus binary categorical judgment in face perception underlies the crucial social ability to rapidly recognize a face. Teasing the two analyses apart may help characterize the neural pathology in the developmental disorders that involve face perception deficits. Based on the results, a cognitive therapy can be strategically designed to pinpoint such deficits, and to therefore help children and adults with these disorders.