PERCEPTUAL CONSTRAINTS FOR IMAGE UNDERSTANDING IN HUMANS

这个项目的目标是了解人类视觉是如何 系统使用纹理和轮廓信息来感知空间 环境中的曲面布局。计算的一个结果 纹理和轮廓信息的分析，以及一些 演示的一个重要特点是，可以使用多个不同的约束 以帮助解释表面几何形状，反映 产生环境的基本范畴结构 到约束。拟议的研究将侧重于揭示 视觉系统使用的约束结构和 它决定应用哪些约束的策略。的 建议的战略将是研究如何纹理和轮廓 信息与其它通常可用的线索（如立体声）相互作用， 调节和模糊来确定表面几何形状的感知。 线索之间的相互作用模式，例如纹理和 曲面几何形状的轮廓信息偏差解释， 立体声，将被用来调查约束结构的基础 纹理和轮廓信息的解释-什么 视觉系统可以使用的约束，以及它如何确定哪些约束 对任何给定场景施加的约束;例如，是否 以及视觉系统如何协同使用立体信息 用纹理和轮廓线索来确定 适合用于线索解释。研究将涉及 这是对以往工作方法的重大改进。我们将 研究纹理和表面轮廓的感性解释 信息在半自然的刺激，我们将产生的 将纹理或轮廓图案投影到所观察的真实的表面上 双眼。刺激投射的物理设置将使 我们可以操纵所观察表面的3D方向 独立于投射在它们身上的图案。通过使用表面 以深度为导向的刺激，我们减少了内在的冲突， 在CRT上呈现的模拟3D显示器;包括 线索之间的冲突和不同知觉之间的冲突 刺激解释的模式;也就是说，作为2D模式投影 在正面平行显示器上或作为3D的图形表示， 场景结果将提供更深入的了解 3D表面纹理和轮廓线索的微观结构 几何

The goal of this project is to understand how the human visual system uses texture and contour information to perceive the spatial layout of surfaces in the environment. One result of computational analyses of texture and contour information, as well as some demonstrations, is that multiple, different constraints may be used to aid in the interpretation of surface geometry,reflecting the essential categorical structure of the environment which gives rise to the constraints. The proposed research will focus on uncovering the structure of the constraints which the visual system uses and the strategies by which it determines which constraints to apply. The proposed strategy will be to investigate how texture and contour information interacts with other normally available cues like stereo, accommodation and blur to determine percepts of surface geometry. Patterns of interaction between the cues, such as how texture and contour information bias interpretations of surface geometry from stereo, will be used to investigate the constraint structure underlying the interpretation of texture and contour information-what constraints the visual system can use and how it determines which constraints to impose for any given scene; for example, whether and how the visual system uses stereo information cooperatively with texture and contour cues to determine what constraints are appropriate to use for cue interpretation. The research will involve a significant methodological advance over previous work. We will study the perceptual interpretation of texture and surface contour information in semi-naturalistic stimuli, which we will generate by projecting texture or contour patterns onto real surfaces viewed binocularly. The physical set-up for stimulus projection will enable us to manipulate the 3D orientations of viewed surfaces independently of the patterns projected on them. By using surfaces oriented in depth for stimuli, we reduce the conflicts inherent in simulated 3D displays presented on a CRT; including both gross conflicts between cues and conflicts between different perceptual modes of stimulus interpretation; that is, as 2D patterns projected on a fronto-parallel display or as pictorial representations of 3D scenes. The results will provide a deeper understanding of the microstructure of texture and contour cues for 3D surface geometry.