Neural Basis of Shape from Texture

根据纹理确定形状的神经基础

• 批准号: 8658075
• 负责人: Qasim Zaidi
• 金额: $ 35.47万
• 依托单位: STATE COLLEGE OF OPTOMETRY
• 依托单位国家: 美国
• 财政年份: 2001
• 资助国家: 美国
• 起止时间: 2001-03-01 至 2016-04-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/8658075
• 关键词: 3-Dimensional; Amblyopia; Attention; Bayesian Analysis; Biological; Brain; Cognition; Complex; Computer Simulation; Computer Vision Systems; Conflict (Psychology); Contracts; Convergence Insufficiency; Cues; Data; Development; Discrimination; Eye; Eye Movements; Frequencies; Geometry; Head Movements; Image; Learning; Light; Measurement; Measures; Methods; Modeling; Motion; Movement; Neurologic; Neurons; Output; Patients; Pattern; Perception; Performance; Process; Property; Psychophysics; Relative (related person); Resolution; Retina; Retinal; Rotation; Shapes; Signal Transduction; Simulate; Slide; Space Perception; Staging; Stimulus; Strabismus; Stream; Surface; Swimming; System; Testing; Texture; Translating; Variant; Vision; Visual; Visual Perception; Walking; Work; analog; area MT; area V1; base; design; movie; nervous system disorder; neural model; novel; object motion; object shape; prototype; relating to nervous system; research study; response; sample fixation

DESCRIPTION (provided by applicant): People often need to judge the shapes and movements of 3-D objects from a distance. Images on the retinae are 2-D, but pattern and motion information contain cues about 3-D shapes. Building on our previous work we expect to make significant progress in understanding 3-D shape perception from these cues, and thus offer a prototype for how the brain extracts information from the world to infer environmental properties. When surfaces have texture patterns, deformations of these patterns in retinal images provide clues for the brain to judge 3-D shape. When signals from the eyes reach the first cortical area V1, they are processed by neurons that are selectively tuned to orientations and spatial frequencies. We parsed texture deformations into orientation flows and spatial frequency gradients, to show that particular orientation flows evoke percepts of specific 3-D shapes, whereas frequency gradients provide cues to relative depth. These results led to models of how later cortical neurons could extract texture patterns and signal 3-D shapes. We now propose to extend our approach beyond static objects. When objects change shape (e.g. by bending, coiling, or contracting) as they move (e.g. walk, tumble, crawl, hop, slide, or swim), changes in the retinal image create patterns of local velocities that provide additional cues to 3-D shape. Since any retinal image can result from projections of many different 3-D objects, the brain relies on prior assumptions to infer the correct shape. Previous studies have only looked at rigid objects and used shape inference models based either on the brain assuming that the object is rigid or that faster points are nearer. We will use novel stimuli that put these prior assumptions in conflict, and thus examine how the brain potentiates a prior. This section will culminate in a model for choosing between conflicting assumptions, something that is often required in perception and cognition. Next, we will use randomly deforming non-rigid 3-D waves to examine how global shape properties, e.g. symmetry, influence perceived object motions by selectively combining disparate outputs of motion sensitive neurons. These results will unveil interactions between the form and motion cortical systems. Finally, we will examine observers' percepts of dynamic shape changes that require more sophisticated analyses of retinal velocity patterns. Neurons in the motion sensitive cortical area MT respond to 1-D motion shear and compression/divergence, so we will extract these qualities and combine them into 2-D velocity patterns of divergence, rotation, and deformation. Neural filters formed by these patterns will be used to explain perceived changes in 3-D shapes. We will base our filters on responses of MT and later cortical neurons to our stimuli, measured in a parallel project. We will thus present the first neural model that can explain observers' percepts of both rigid and non-rigid textured objects. The performance of our model will be compared against the best computer-vision models on motion-capture data from real deforming objects. We expect this project to introduce new ideas, methods and results for understanding visual perception of 3-D shapes and its deficits in neurological patients.

描述（申请人提供）：人们经常需要从远处判断三维物体的形状和运动。视网膜上的图像是二维的，但图案和运动信息包含有关三维形状的线索。在我们以前的工作的基础上，我们希望在从这些线索中理解3D形状感知方面取得重大进展，从而为大脑如何从世界中提取信息来推断环境属性提供原型。当表面有纹理图案时，视网膜图像中这些图案的变形为大脑判断三维形状提供了线索。当来自眼睛的信号到达第一皮层区域V1时，它们被选择性地调谐到方向和空间频率的神经元处理。我们将纹理变形解析为方向流和空间频率梯度，以表明特定的方向流引起对特定3-D形状的感知，而频率梯度提供相对深度的线索。这些结果导致了后来皮质神经元如何提取纹理模式和信号3D形状的模型。我们现在建议将我们的方法扩展到静态对象之外。当物体在移动（例如行走、翻滚、爬行、跳跃、滑动或游泳）时改变形状（例如弯曲、盘绕或收缩）时，视网膜图像中的变化会产生局部速度的模式，为3D形状提供额外的线索。由于任何视网膜图像都可以由许多不同的3D物体的投影产生，因此大脑依赖于先前的假设来推断正确的形状。以前的研究只关注刚性物体，并使用基于大脑假设物体是刚性的或更快的点更近的形状推理模型。我们将使用新的刺激，使这些先验假设发生冲突，从而研究大脑如何增强先验。本节将以一个在相互冲突的假设之间进行选择的模型为结尾，这是感知和认知中经常需要的。接下来，我们将使用随机变形的非刚性3-D波来研究全局形状属性（例如对称性）如何通过选择性地组合运动敏感神经元的不同输出来影响感知到的对象运动。这些结果将揭示形式和运动皮质系统之间的相互作用。最后，我们将研究观察者对动态形状变化的感知，这需要对视网膜速度模式进行更复杂的分析。运动敏感皮层区MT中的神经元对一维运动剪切和压缩/发散做出反应，因此我们将提取这些特性并将它们联合收割机组合成发散、旋转和变形的二维速度模式。由这些模式形成的神经过滤器将用于解释3D形状的感知变化。我们将根据MT和后来的皮层神经元对我们的刺激的反应，在一个平行的项目中测量我们的过滤器。因此，我们将介绍 第一个神经模型，可以解释观察者对刚性和非刚性纹理物体的感知。我们的模型的性能进行比较，对最好的计算机视觉模型的运动捕捉数据从真实的变形对象。我们希望这个项目能为理解神经系统患者对三维形状的视觉感知及其缺陷引入新的想法、方法和结果。