The visual information subtending 3D shape perception and recognition in humans

人类 3D 形状感知和识别的视觉信息

• 批准号: RGPIN-2014-03789
• 负责人: Arguin, Martin
• 金额: $ 1.89万
• 依托单位: Université de Montréal
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2016
• 资助国家: 加拿大
• 起止时间: 2016-01-01 至 2017-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=610645
• 关键词: visual; information; subtending; 3D; shape

Visual shape perception is crucial for effective interactions with our environment and it is an essential basis for the recognition of visual objects. In the present application, we propose to use a relatively novel approach for the study of visual perception which aims to determine directly, without theoretical a priori or bias, the visual information subtending the recognition of visual shapes that are rotated in depth in normally functioning adult human observers. The new knowledge acquired through the approach proposed here should offer significant theoretical advances since it will inform on three major dimensions that distinguish among competing theories: 1- Type of image feature: The class of descriptors used by the visual system to represent shape determines the type of features that is most relevant for performance. 2- Feature stability: Whether the descriptors by which shapes are represented are stable or variable according to the exposure conditions and stimulus context. 3- Feature conjunctions: Whether feature conjunctions are used to code the spatial relations within an object or alternatively, that the spatial relations among features are irrelevant for shape recognition. We will first use the classification images techniques of Adaptive Bubbles and of Visual Spread to determine the image features that underlie the recognition of objects from different classes of synthetic 3D shapes shown from variable viewpoints. For each object, these effective locations will be cross-referenced across viewpoints to determine features that are either viewpoint invariant (i.e. which occur in more than one view of the object) and viewpoint specific (i.e. which occur in only one view). These features will be compared to those used by an ideal observer model and characterized along theoretically motivated principles to extract general rules that can be applied to predict the image regions that will be critical for the recognition of novel object sets. Tests will then be conducted to confirm the value of these predictions on already used object sets (verification tests) as well as to assess them in new sets of objects not studied before with classification image techniques (generalization tests). Eye movements will be recorded during the verification and generalization tests as an additional index of the features that are relevant for shape processing. The object classes studied will vary in terms of their complexity and their capacity to generalize to objects we encounter in daily life. They will be: 1- single geons (Biederman, 1987); 2- assemblies made of four interconnected geons; 3- the “surface shapes” used by Hung et al. (2012) and Yamane et al. (2008) in their study of single-neuron shape selectivity in the macaque inferotemporal area, which are relatively elementary objects that may nevertheless present a rich surface topography; and 4- “medial axial shapes” developed by Hung et al. (2012), which are rather complex objects, many of them reminiscent of animal or insect shapes. Elucidating the features underlying visual shape perception is of fundamental importance in the field. The current proposal will contribute to advance our knowledge on the issue through its use of sophisticated psychophysical techniques that have proven their capacity for our purposes, and the use of various object classes that will sample a subset of crucial constraints that our visual system must meet for everyday object recognition.

视觉形状感知对于与环境的有效交互至关重要，是视觉对象识别的重要基础。在本申请中，我们提出使用一种相对新颖的方法来研究视觉感知，其目的是直接确定，没有理论上的先验或偏见，视觉信息对着识别的视觉形状，旋转在正常运作的成年人观察者的深度。 通过这里提出的方法获得的新知识应该提供显着的理论进步，因为它将告知三个主要维度，区分竞争的理论：1-图像类型功能：视觉系统用来表示形状的描述符的类别确定了与性能最相关的特征类型。2-特性稳定性：根据曝光条件和刺激背景，表示形状的描述符是稳定的还是可变的。3-特征连词：特征合取是否用于编码对象内的空间关系，或者特征之间的空间关系与形状识别无关。 我们将首先使用自适应气泡和视觉传播的分类图像技术来确定图像特征，这些图像特征是从可变视点显示的不同类别的合成3D形状识别对象的基础。对于每个对象，这些有效位置将跨视点交叉参考，以确定视点不变（即，在对象的多于一个视图中出现）和视点特定（即，仅在一个视图中出现）的特征。这些功能将被比较到所使用的一个理想的观察者模型和特点沿着理论上的动机原则，以提取一般规则，可以应用于预测的图像区域，这将是至关重要的识别新的对象集。 然后将进行测试，以确认这些预测对已经使用的对象集的价值（验证测试），以及评估它们在新的对象集之前没有研究过的分类图像技术（泛化测试）。在验证和泛化测试期间将记录眼球运动，作为与形状处理相关的特征的附加索引。 所研究的对象类在其复杂性和推广到我们日常生活中遇到的对象的能力方面会有所不同。它们将是：1- single geons（Biederman，1987）; 2-由四个相互连接的geons组成的组件; 3-Hung等人（2012）和Yamane等人（2008）在猕猴颞下区的单神经元形状选择性的研究中使用的“表面形状”，其是相对基本的对象，但可能呈现丰富的表面形貌;和4-由Hung等人（2012）开发的“中轴形状”，其是相当复杂的对象，其中许多使人联想到动物或昆虫形状。 阐明视觉形状感知的基本特征在该领域具有根本的重要性。目前的建议将有助于推进我们对这个问题的认识，通过使用复杂的心理物理技术，已经证明了他们的能力，为我们的目的，并使用各种对象类，将采样的一个子集的关键约束，我们的视觉系统必须满足日常对象识别。