A fresh look at visual sampling: How are fixational eye-movements optimised? [PhysFEM]

感官系统，如视觉或听觉，应该如何最佳地感知世界？太多的细节需要花费太长时间，需要获取和处理太多的资源；太少的风险无法捕捉到我们周围正在发生的重要信息。人类的视觉就是一个引人入胜的例子——即使当我们的目光“固定”在一个目标物体上，或者在执行一项特定的任务，比如决定两个物体中哪一个更高的时候，眼睛也会处于持续的、明显随机的、我们无法理解的运动中。有人可能会认为，这种眼睛不自觉的运动只会“模糊”视力，但有理由相信，它们实际上可能会增强视力。怀疑这一点的一个原因是，我们知道视觉的许多方面都朝着最好的表现发展。这个项目——眼球固定运动的物理学（PhysFEM）——通过结合随机运动的理论物理学、视觉神经科学和心理学的生命科学的思想、方法和人员，解决了我们眼睛一直存在的不自主运动的挑战。这种结合是全新的，并且具有潜在的强大功能：几乎没有物理领域不考虑寻找优化某些东西的轨迹路径。例子出现在复杂的经典力学、量子力学和随机过程的热物理中。这些来自物理学的想法提供了一种自然而新鲜的方式来思考注视的眼球运动优化视觉某些方面的可能性。不仅如此，他们还带来了新的计算方法来寻找这些路径，从而得出了经验上可测试的预测，即在特定任务要求下，眼睛如何移动以最大限度地获取可用信息。在受控条件下对人类视觉表现的客观测量——“心理物理学”领域——完成了模型预测和测试的迭代循环。物理学也进入了这个项目的实验模式，因为牛津感知实验室使用了设备来测量眼球运动。该团队使用“自适应光学”技术——最初是为大型天文望远镜开发的，用于校正大气的光学畸变——对眼睛后部的视网膜成像，同时校正眼睛充满液体的光学元件的畸变。这种校正允许在没有任何侵入性程序的情况下，对人类视网膜中的单个细胞进行高分辨率成像。自适应光学扫描激光检眼镜（AO-SLO）将在PhysFEM中同时执行三项任务：(i)将受控图像投射到视网膜上；（ii）以单个光探测细胞（“光感受器”）的分辨率对视网膜成像；（iii）在一系列特定的视觉任务下，以前所未有的精度测量眼球运动。该项目的研究结果将通过与美国宾夕法尼亚大学的发起人和主任合作，建立在一个不断发展的、开放获取的视觉科学和技术计算工具ISETBio中。ISETBio是一套开源软件工具，它描述了早期视觉的感觉过程，并为神经处理如何结合fem的现实计算实现和模型评估提供了一个平台，并将其转移到生物、计算机视觉和医学的国际研究人员社区。

How should a sensory system, such as vision or hearing, optimally sample the world? Too much detail takes too long, and too much resource to acquire and process; too little risks failing to capture the vital information on what is going on around us. Human vision is a fascinating example - for even when the gaze is 'fixed' onto a target object or performing a specific task such as deciding which of two objects is higher, the eyes are in constant, apparently-random, motion that we do not understand. One might assume that such involuntary movements of the eye could only 'blur' vision, but there are reasons to believe that they might actually enhance it. One reason for suspecting this is that we know that many aspects of vision have evolved towards the best performance possible.This project - the Physics of Fixational Eye Movements (PhysFEM) - addresses the challenge of these ever-present involuntary movements of our eyes by combining ideas, methods and people from theoretical physics of random motion, and from the life sciences of visual neuroscience and psychology. The combination is new and potentially powerful: there is almost no realm of physics that does not think about finding paths of trajectories that optimise something. Examples arise in complex classical mechanics, quantum mechanics, and the thermal physics of random processes. These ideas from physics provide a natural but fresh way of thinking about the possibility that fixational eye movements optimise some aspects of vision. More than that, they bring new calculation methods to find such paths, leading to empirically testable predictions about how the eyes might move to maximise the information available given particular task-demands. The objective measurement of human visual performance under controlled conditions - the domain of 'psychophysics' - completes the iterative cycles of model predictions and testing.Physics enters the experimental mode of this project also in terms of the equipment used in in the Oxford Perception Lab to measure eye movements. The team use 'adaptive optics' - first developed for large astronomical telescopes to correct the optical distortions of the atmosphere - to image the retina at the back of the eye whilst correcting for distortions in the eye's fluid-filled optical components. Such correction permits high-resolution imaging of individual cells in the human retina without any invasive procedure. The Adaptive Optics Scanning Laser Ophthalmoscope (AO-SLO) will perform three simultaneous tasks in PhysFEM: (i) projecting controlled images onto the retina; (ii) imaging the retina at the resolution of individual light-detecting ('photoreceptor') cells; (iii) measuring eye movements with unprecedented accuracy, under a series of specific visual tasks.The findings of the project will be built into a growing, open-access computational tool for vision science and technology, ISETBio, through a collaboration with its originator and director at the University of Pennsylvania, USA. ISETBio is an open-source set of software tools that characterise the sensory processes of early vision, and provides a platform for realistic computational implementation and evaluation of models for how neural processing incorporates FEMs, and transferring this to the international community of researchers in biological- and computer-vision and medicine.