CHS: Small: Using Virtual Reality for the Dynamic, Real-Time Optimization of Human Visual Perception

计算视觉和视觉科学传统上依靠自然界的统计数据以及彼此对视觉处理的洞察。直到最近，这些方法主要是静态的和相关的：自然世界一直被视为图像的集合，处理应该进行优化，自然场景中的平均规律性已经被证明与感知偏差相关。对最近影响感知的体验进行的任何动态调整往往被最小化，这在很大程度上是因为还没有办法破坏环境并测试其影响。本研究将移动计算技术与沉浸式增强现实技术相结合，探索视觉感知如何动态适应环境中遇到的规律。PI将研究方向知觉，这是人类视觉处理的第一皮质层的一个特征，因此是一个逻辑起点。如果刺激是在使用最新环境统计数据来动态优化感知的框架下编码的，那么改变典型的环境规则应该会对人类的视觉表现产生可预见的影响。PI认为，现有的人类感知计算模型可以扩展到预测输入中的哪些变化将改善(或抑制)人类感知性能。反过来，这将开启训练人类感知的可能性，以优化在现实世界任务中的表现，这些任务以前需要广泛的专业培训或昂贵的定制软件。为了通过量化编码偏差可以被改变或消除的程度来创建更精确的人类视觉系统灵活性模型，该项目将包括三个相互关联的推力。首先，PI将开发一套软件工具来近乎实时地处理视觉环境，并将使用这些工具来系统地调查人类感知的变化，以响应对统计内容非典型环境的体验。他将测量人类在各种真实世界任务(例如物体检测)上感知性能的变化，以响应非典型环境输入的身临其境体验。他将开发和测试这种人类知觉学习的计算模型。初步研究表明，计算机图像过滤和虚拟现实硬件的结合可以用来改变后续的视觉处理，其方式基于过滤后的输入是可预测的。

Computational vision and vision science have traditionally looked to the statistics of the natural world and each other for insights into visual processing. Until recently, these approaches have been primarily static and correlational: the natural world has been treated as a collection of images for which processing should be optimized, and the averaged regularities in natural scenes have been shown to be correlated with perceptual biases. Any dynamic adjustment to recent experience influencing perception has often been minimized, in large part because there have not been ways to disrupt the environment and test the effects. But recent advances in computing and virtual reality hardware have made possible the manipulation of visual input in near-real time.This research combines mobile computing technology with immersive augmented reality to explore how visual perception dynamically adapts to encountered regularities in the environment. The PI will investigate perception of orientation, a feature of the first cortical layer of human visual processing, and thus a logical starting point. If stimuli are encoded under a framework that uses recent environmental statistics to dynamically optimize perception, then altering the typical environmental regularities should have predictable effects on human visual performance. The PI argues that existing computational models of human perception can be extended to predict which changes in the input will improve (or inhibit) human perceptual performance. This, in turn, will open up the possibility of training human perception to optimize performance on real world tasks that previously required extensive specialized training or costly, custom-built software. With the goal of creating a more precise model of the flexibility of the human visual system by quantifying the extent to which encoding biases can be altered or obliterated, this project will include three interrelated thrusts. First, the PI will develop a suite of software tools to process the visual environment in near real-time, and will use these tools to systematically investigate changes in human perception in response to experience with environments whose statistical content is atypical. He will measure changes in human perceptual performance on a variety of real-world tasks (e.g., object detection), in response to immersive experience with atypical environmental input. And he will develop and test a computational model of this human perceptual learning. Preliminary research suggests that the combination of computer image-filtering and virtual reality hardware can be used to change subsequent visual processing in ways that are predictable based on the filtered input.