What makes a face recognizable?

大多数人可以毫不费力地识别成千上万的独特的面孔后一眼[1，2]。这是一个令人印象深刻的壮举，考虑到所有的脸都有相同的特征（即，两个眼睛，一个鼻子和一个嘴）以相似的结构布置（即，眼睛在鼻子上方，嘴巴上方）。我的长期目标是更好地了解人类如何识别面孔。这是一个具有挑战性的问题，尽管有大量关于这个主题的文献，但我们才刚刚开始了解这种非凡的能力。一些人认为，这一领域的研究进展缓慢，因为缺乏对熟悉面孔的现实世界识别的关注[15]。事实上，大多数研究都集中在识别不熟悉的面孔，这些面孔在实验之前对参与者来说是新的，并且在参与者被要求在“新”面孔中识别这些“老”面孔之前只显示一次。这与认识熟人或名人形成鲜明对比，这些人的面孔通过反复曝光逐渐熟悉。熟悉和不熟悉的人脸识别之间的一个关键区别是，熟悉的人脸识别对外观的变化是鲁棒的;我们可以识别熟悉的面孔，尽管观察姿势或面部表情发生了变化[13，14]。在接下来的五年里，我的目标是推进我们对如何建立强大的熟悉面孔表征的理解。视觉信息处理是由刺激的属性（自下而上的变量）和刺激出现的上下文（自上而下的变量）驱动的。我将研究刺激驱动的信息（目标1）和背景信息（目标2）如何支持稳健的熟悉面孔表征的形成。参与者将学习一组面孔，直到他们熟悉为止。不同的实验将进行的刺激驱动的信息的类型所传达的面孔（目标1）或所示的上下文信息是不同的（目标2）。当观看视觉场景时，我们的眼睛不断运动，有证据表明眼睛运动与记忆的形成功能相关[26，27]。我的第三个目标是探索眼动在熟悉面孔表征的形成和提取中所起的作用（目标3）。熟悉的面孔表征不仅强大，而且耐用，大多数人都能够识别他们已经超过35年没有见过的人[28，29]。我的第四个目标是通过测量短期和长期保持延迟后的面孔再认来研究面孔长期记忆的机制（目标4）。 冲击理解面孔识别对情景记忆模型和专门的视觉识别系统具有重要的理论意义。我的研究计划也有实际意义，以改善现实世界的问题，如目击者识别的准确性，在安全环境中的人脸识别（例如，护照或警察），以及自动面部识别技术。

Most humans can effortlessly recognize thousands of unique faces after a single glance [1, 2]. This is an impressive feat considering that all faces share the same features (i.e., two eyes, a nose and a mouth) arranged in a similar configuration (i.e., eyes above the nose above the mouth). My long-term objective is to better understand how humans recognize faces. This is a challenging question and despite a large body of literature on the topic, we are only beginning to understand this remarkable ability. Some have suggested that research in this area is progressing slowly because of a lack of focus on real-world recognition of familiar faces [15]. Indeed, most studies have focused on recognition of unfamiliar faces that are new to participants prior to the experiment and only shown once before participants are asked to recognize these `old' faces among `new' faces. This contrasts with recognition of acquaintances or famous people whose faces gradually become familiar with repeated exposure. A key difference between familiar and unfamiliar face recognition is that familiar face recognition is robust to transformations in appearance; we can recognize familiar faces despite changes in viewing pose or facial expression [13, 14]. My goal over the next five years is to advance our understanding of how robust familiar face representations are built. Visual information processing is driven by the properties of stimuli (bottom-up variables) and by the context within which stimuli appear (top-down variables). I will investigate how stimulus-driven information (Obj. 1) and contextual information (Obj. 2) support the formation of robust familiar face representations. Participants will learn a set of faces until they are familiar. Different experiments will be conducted where the type of stimulus-driven information conveyed by faces (Obj. 1) or the contextual information shown is varied (Obj. 2). When viewing a visual scene, our eyes are constantly in motion and there is evidence that eye movements are functionally relevant for the formation of memories [26, 27]. My third objective is to explore the role that eye movements play in the formation and retrieval of familiar face representations (Obj. 3). Familiar face representations are not only robust they are also durable, and most humans are able to recognize individuals they have not seen for more than 35 years [28, 29]. My fourth objective is to investigate the mechanisms that underlie long-term memory for faces (Obj. 4) by measuring face recognition after short- and long-retention delays. Impact. Understanding face recognition has important theoretical implications for models of episodic memory and specialized visual recognition systems. My research program also has practical implications for improving real-world problems such as the accuracy of eye-witness identification, face identification in security contexts (e.g., passport or police officers), and automatic facial recognition technology.