Looking and seeing: visual information processing, visual attention and eye movements.

自然场景包含了大量的视觉信息。为了解决这个问题，我们至少使用了两种策略，涉及视觉处理和眼球运动之间的积极相互作用。首先，视网膜的中心部分（即“中央凹”）和大部分负责视觉的大脑一起，对视觉场景中一小部分的视觉信息进行非常详细的处理。外周视网膜处理场景的其余部分的细节较少。大约每隔几百毫秒，一个快速的眼球运动（被称为扫视，持续时间少于50毫秒）就会重新定位中央凹，以接收来自场景中感兴趣的新区域的光；因此，不同的区域被依次处理得非常详细。第二，一个被称为视觉注意的过程增强了对当前场景中重要刺激的中央凹和外周视网膜信息的处理：例如，夜空中明亮的物体或寻找小丑鼻子时的红色圆圈。这种增强增强了我们对参与刺激的意识，并使它们更有可能被中央处理，被选为扫视目标。我们对注意力和眼球运动如何在简单场景中通过单个眼球运动相互作用已经有了相当多的了解。例如，在许多情况下，空间注意力的一个组成部分在扫视之前向扫视目标转移。然而，我们对这种从简单场景中获得的知识如何延续到人类在自然环境中的行为知之甚少。此外，即使对于简单的场景，我们也刚刚开始了解视觉注意力是如何作用于近中央凹刺激的，尽管我们的视觉体验是由中央凹刺激主导的。在另一个例子中，由于视觉大脑主要在以眼睛为中心的参照系中代表视觉场景，因此大脑迅速转移其注意力表征并与眼球运动同步，以尽量减少干扰。我最近与同事们的合作才刚刚开始阐明这一过程。这个研究项目的长期目标是建立对主动视觉的完整理解：当我们在看东西时，我们是如何环顾四周的。在短期内，这一提议将旨在弥合现实生活中的视觉和眼动功能与更简单任务中确定的基本机制之间的差距。我们将结合人类的感知和眼球追踪实验，对他们的注意力和眼球运动行为进行详细的计算建模，并对执行自然视觉搜索任务的猴子的已有神经元数据集进行分析。这项研究对我们理解人类在现实生活环境中的视觉至关重要，它在我们能想到的几乎所有现实生活视觉的例子中都有大量应用，从驾驶、阅读到网页设计，从放射科医生检查x射线，甚至是试图捕捉/复制人类视觉功能的机器学习算法。

Natural scenes contain a large amount of visual information. To deal with this, we use at least two strategies involving an active interplay between visual processing and eye-movements. First, a central portion of the retina (the "fovea") along with the large portion of the visual brain devoted to it, processes visual information from only a small part of the visual scene in great detail. The peripheral retina processes the rest of the scene in less detail. Approximately every few hundred milliseconds, a fast eye-movement (known as a saccade and lasting less than about fifty milliseconds) repositions the fovea to receive light from a new region of interest in the scene; different regions are thus processed sequentially in high detail. Second, a process known as visual attention enhances the processing of both foveal and peripheral retinal information from currently important stimuli in the scene: for example, bright shining objects in a night sky or red circles when searching for a clown's nose. This enhancement privileges our awareness of attended stimuli and makes them more likely to be processed foveally by being chosen as saccade targets. We know a fair amount about how attention and eye-movements interact in simple scenes with a single eye-movement. For example, a component of spatial attention shifts towards the saccade target prior to the saccade in many conditions. However, we know very little about how this knowledge gained from simple scenes carries over to humans functioning in naturalistic environments. Further, even for simple scenes, we are just starting to learn about how visual attention acts on near-foveal stimuli, even though our visual experience is dominated by foveal stimuli. In another example, since the visual brain represents the visual scene mostly in a eye-centered reference frame, the brain shifts its attentional representation rapidly and synchronized with the eye-movement in order to minimize disruptions. My recent work with my colleagues has just begun to shed light on this process. The long-term goal of this research program is to build a complete understanding of active vision: how do we look around when we are seeing. In the short-term, this proposal will aim to bridge the gap between real-life visual and eye-movement function and the basic mechanisms identified in much simpler tasks. We will use a combination of perceptual and eye-tracking experiments in humans, detailed computational modeling of their attention and eye-movement behavior, and the analysis of pre-existing neuronal datasets from monkeys performing naturalistic visual search tasks. This research is critical to our understanding of human vision in real-life environments, with numerous applications in almost any example of real-life vision that we can think of, from driving and reading to web-design, radiologists examining X-rays and even machine-learning algorithms trying to capture/replicate human visual function.