Developing an early vision model of lightness perception: experimentand theory

这个项目的目标是描述允许人类视觉系统从视网膜输入中提取有意义的感知类别的计算的特征。特别是，我们研究了表面亮度的感知属性，即感知的表面反射率，是如何从刺激视网膜中光感受器阵列的亮度模式中计算出来的。目前还没有关于明度感知的计算模型。亮度感知的问题在于亮度信号的模糊性，因为相同的亮度值可以由在不同照明下看到的不同表面反射率产生。传统上，亮度感知是在简化的刺激中研究的，其中不同亮度的目标被呈现在不同亮度的背景上。然而，这样的刺激一直没有关于目标的表面属性以及关于其中看到目标表面的场景的深度和照明结构的信息。最近的证据表明，这种不自然的观看情况严重误导了我们对视觉感知的理解，并阻碍了我们观察视觉机制，因为它们在更现实的条件下运行。这种不受约束的刺激可能带来了在更自然的条件下不会发生的知觉模糊。在这里，我们采用一种不同的方法，使用中等复杂性的现实刺激。我们使用的刺激是五颜六色的棋盘图案，这是计算机渲染的三维场景图像，其中我们可以指定棋盘中每个棋盘的反射率及其照明。它们已被证明是一个受控良好的实验试验台，用于研究涉及亮度感知的计算，因为它们允许我们捕捉与视觉刺激的外部世界属性相关的知觉表现。它们足够简单，可以将它们重建为真实世界的三维模型，同时它们允许复杂度的不断增加来研究场景分割机制对亮度感知的影响。我们将详细阐述基于亮度边缘对比度的早期视觉感知模型，该模型是从我们自己的团队和其他研究人员最近的一些发现中出现的。我们将使用该模型作为本提案中所描述的研究问题的起点和理论框架。

The goal of this project is to characterize the computations that allow the human visual system to extract meaningful perceptual categories from the retinal input. In particular, we study how the perceptual attribute of surface lightness, i.e. perceived surface reflectance, is computed from the luminance pattern that stimulates the photoreceptor array in the retina. At present there is no computational model of lightness perception. The problem of lightness perception lies in the ambiguity of the luminance signal, because the same luminance value can result from different surface reflectances that are seen in different illuminations. Traditionally lightness perception has been studied in simplified stimuli in which targets of varying luminance have been presented on backgrounds of varying luminance. However, such stimuli have been void of information about the surface properties of a target and about the depth and illumination structure of the scene in which the target surface was seen. Recent evidence suggests that this unnatural viewing situation has seriously misled our understanding of visual perception and has prevented us from observing visual mechanisms as they would operate under more realistic conditions. Such under-constrained stimuli might have introduced perceptual ambiguities that would not occur under more natural conditions. Here we pursue a different approach using realistic stimuli of moderate complexity. The stimuli that we use are variegated checkerboard patterns that are computer-rendered images of three-dimensional scenes in which we can specify the reflectances of each check in the checkerboard as well as their illumination. They have proved to be a well-controlled experimental test bed to study the computations involved in lightness perception, because they allow us to capture perceptual performance in relation to the external world properties of visual stimuli. They are simple enough to rebuild them as real-world three-dimensional models, and at the same time they allow the successive increase in complexity to study the influence of mechanisms of scene segmentation on lightness perception. We will elaborate an early vision model of lightness perception that is based on the contrast at luminance edges and that has emerged from a number of recent findings from our own group and from other researchers. We will use the model as the starting point and theoretical framework for the research questions described in this proposal.