Perceptual Organization of Two Dimensional Patterns

描述（由申请人提供）：人类视觉系统的力量在很大程度上源于其感知分组机制的能力，即在视网膜编码的图像中发现结构和组织。我们建议继续并扩展我们的一般定量方法来理解知觉分组的机制。我们的方法包括：(1)测量与感知分组相关的自然图像的统计特性；(2)根据自然图像的统计特性、低水平视觉的生理和心理物理学以及计算原理开发感知分组的模型；(3)在人类心理物理实验中测试这些模型和竞争模型的预测。我们提出用两种方法测量图像轮廓和区域的自然场景统计量。一种方法是测量图像中的局部特征，然后计算简单的共现统计量，即不同可能特征值的联合概率。另一种方法是手动分割图像，测量图像中的局部特征，然后计算以分割为条件的共现统计量。这些统计量，特别是条件统计量，将用于生成轮廓和区域分组的贝叶斯模型。在初步研究中，我们发现了自然图像的一些强大的统计特性，并开始开发感知分组的模型，其中这些特性以最优（理性）的方式被利用。提出了一系列的轮廓分组和区域分组实验来检验这些模型。在之前的项目期间，我们已经取得了良好的进展。这项工作很重要，因为理解自然刺激的物理学对于理解自然（真实）感知任务和视觉系统的几乎每一个设计方面都是必不可少的，因为感知分组机制在所有高阶视觉功能（包括形式、形状、物体和场景感知）中起着核心作用。值得注意的是，这些研究将首次提供近距离树叶的轮廓和区域统计的详细测量，这对研究猴子的视觉机制特别相关。同样值得注意的是周边视觉和眼球运动在自然任务中的作用的研究。了解这些能力对于理解低视力对视觉表现的影响以及如何最好地减轻这些影响至关重要。

DESCRIPTION (provided by applicant): The power of the human visual system derives in large part from the ability of its perceptual grouping mechanisms to find structure and organization in the images encoded by the retinas. We propose to continue and extend our general quantitative approach to understanding the mechanisms of perceptual grouping. Our approach consists of: (1) measuring the statistical properties of natural images that are relevant for perceptual grouping, (2) developing models of perceptual grouping that are informed by the statistical properties of natural images, by the physiology and psychophysics of low-level vision, and by computational principles, and (3) testing the predictions of these models and competing models in human psychophysical experiments. We propose to measure natural scene statistics for image contours and regions using two methods. One involves measuring local features in images and then computing simple co- occurrence statistics, i.e., the joint probabilities of different possible feature values. The other involves hand segmenting images, measuring local features in the images, and then computing co-occurrence statistics that are conditional on the segmentation. These statistics, especially the conditional statistics, will be used to generate Bayesian models for contour and region grouping. In preliminary studies, we discovered a number of robust statistical properties of natural images and have begun developing models of perceptual grouping where those properties are exploited in an optimal (rational) fashion. A number of contour grouping and region grouping experiments are proposed to test these models. We have made good progress with this general approach during previous project period. This work is important because understanding the physics of natural stimuli is essential for understanding natural (real) perceptual tasks and just about every design aspect of the visual system, and because perceptual grouping mechanisms play a central role in all higher order visual functions including form, shape, object and scene perception. Of note are studies that will provide the first detailed measurements of contour and region statistics of close-up foliage, which is of particular relevance for studies of visual mechanisms in the monkey. Also of note are studies of the role of peripheral vision and eye movements in natural tasks. Understanding these capabilities is crucial for understanding the effects of low vision on visual performance and how best to mitigate those effects.