Top-down and bottom-up processes of scene categorization

自上而下和自下而上的场景分类过程

• 批准号: 8786558
• 负责人: Jiye G Kim
• 金额: $ 5.6万
• 依托单位: PRINCETON UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2013
• 资助国家: 美国
• 起止时间: 2013-01-01 至 2015-12-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/8786558
• 关键词: Address; Agnosia; Area; Attention; Attention deficit hyperactivity disorder; Behavioral; Brain; Brain Injuries; Categories; Cities; Classification; Clinical; Cognitive; Color; Complex; Cues; Diagnostic; Disease; Exhibits; Functional Magnetic Resonance Imaging; Glean; Goals; Health; Human; Image; Image Analysis; Individual; Knowledge; Lesion; Lighting; Modeling; Nature; Parietal; Parietal Lobe; Patients; Pattern; Perception; Performance; Phase; Photic Stimulation; Population; Process; Property; Prosopagnosia; Psychophysics; Research; Residual state; Semantics; Signal Transduction; Stimulus; Stream; Structure; Techniques; Testing; Texture; Vision; Visual; Visual Agnosias; Visual Cortex; Visual attention; Visual system structure; attentional bias; base; computer studies; expectation; experience; forest; innovation; insight; neuromechanism; novel; object perception; operation; paired stimuli; relating to nervous system; research study; response; sensory input; statistics; visual neuroscience

DESCRIPTION (provided by applicant): One of the great challenges of visual neuroscience is to reveal the neural mechanisms defining our perceptual experience. The human visual system is extremely efficient in categorizing complex scenes, even for novel ones never encountered before and often under poor viewing conditions such as brief exposure duration and dim lighting. What are the underlying neural operations that subserve such proficient scene processing? Our visual experience is not only determined by the sensory input, but also influenced by top- down cognitive mechanisms such as attentional biases and expectations. For example, in the midst of continual overflow of information, prior expectation about the visual world can facilitate recognition by biasing the relevant bottom-up input thereby allowing us to quickly deduce plausible interpretations. When anticipating a particular category of a scene (e.g., a city scene), the prior knowledge about the possible configuration or the objects typically found in that scene may be used to form attentional templates, or the internal representation of scenes, allowing for efficient categorization. However, the neural basis of this interactive relationship between top-down and bottom-up processing in scene perception is largely unknown. The main objective of the proposed research is in bridging the current gap between bottom-up and top-down neural processes of scene categorization. The central hypothesis is that the statistical regularities of scenes--defined in image statistics-- that are consistent acros scenes of a particular category (e.g., different city scenes) serve as effective top- down biasing signals (or category templates) in guiding efficient categorization. We will examine the neural mechanisms of scene perception in normal and object agnostic individuals, those who have severe difficulty recognizing objects due to brain damage in the ventral visual system. By using innovative functional magnetic resonance imaging (fMRI) and behavioral experiments and applying sophisticated fMRI analysis techniques, the proposed research will address the following specific aims: (1) characterize the neural representation of image statistics of scenes within and across categories, (2) investigate the neural basis of scene category template formation and (3) investigate scene processing in patients with visual agnosia. By examining scene categorization in the framework of attentional influences, the proposed research aims to elucidate the complex interaction between bottom-up and top-down processes on scene recognition, an ecologically valid but currently underexplored area of research. The innovative strategy of simultaneously studying normal and lesioned brain functions has the potential to significantly advance our knowledge about normal scene perception as well as potentially reveal alternative mechanisms for scene perception when object information is not available. Overall, a better understanding of the neural mechanisms of scene processing and its relation to attentional processes will provide direct translational implications for clinical population with a wide range of visual and attention disorders.

描述（由申请人提供）：视觉神经科学的最大挑战之一是揭示定义我们感知体验的神经机制。人类视觉系统在对复杂场景进行分类时非常有效，即使是以前从未遇到过的新场景，也经常在短时间曝光和昏暗灯光等不良观看条件下进行分类。是什么潜在的神经操作，使这种熟练的场景处理？我们的视觉体验不仅取决于感官输入，而且还受到自上而下的认知机制的影响，如注意偏差和期望。例如，在信息不断溢出的过程中，对视觉世界的先验预期可以通过偏置相关的自下而上的输入来促进识别，从而使我们能够快速推断出合理的解释。当预期场景的特定类别时（例如，城市场景），关于可能的配置或对象的先验知识通常 在该场景中发现的信息可以用于形成注意模板或场景的内部表示，从而允许有效的分类。然而，在场景感知中自上而下和自下而上处理之间的这种交互关系的神经基础在很大程度上是未知的。建议的研究的主要目标是在弥合自下而上和自上而下的场景分类的神经过程之间的差距。中心假设是场景的统计特征-在图像统计中定义-在特定类别的场景中是一致的（例如，不同的城市场景）在指导有效分类中充当有效的自上而下的偏置信号（或类别模板）。我们将研究正常和物体不可知个体的场景感知的神经机制，这些个体由于腹侧视觉系统的脑损伤而难以识别物体。通过使用创新的功能磁共振成像（fMRI）和行为实验，并应用先进的fMRI分析技术，拟议的研究将解决以下具体目标：（1）表征类别内和跨类别场景图像统计的神经表征，（2）调查场景类别模板形成的神经基础，（3）调查视觉失认症患者的场景加工。通过在注意力影响的框架下研究场景分类，该研究旨在阐明场景识别的自下而上和自上而下过程之间的复杂相互作用，这是一个生态有效但目前尚未探索的研究领域。同时研究正常和病变的大脑功能的创新策略有可能显着推进我们对正常场景感知的知识，并可能揭示当对象信息不可用时的场景感知的替代机制。总之，更好地理解场景加工的神经机制及其与注意过程的关系将为临床人群提供直接的翻译意义， 广泛的视觉和注意力障碍。