Initial stages of visual information processing

我计划在三个不同但相关的领域开展工作： 迭代再入模型。神经科学的进步已经引起人们对折返信号的关注，因为它是通过双向通路连接的大脑区域之间的主要通信形式。然而，知觉理论继续以前馈序列的形式出现。我们已经开发了另一种解释，认为感知是从大脑区域之间的迭代交换中产生的。这一解释源于对象替换掩蔽(OSM)的发现，在OSM中，目标和掩码的初始组合显示继续仅与掩码一起进行。当所有显示元素一起终止时，很容易识别目标。仅当遮罩保持在可视状态时，就会发生遮罩，其强度会随着遮罩持续时间的延长而增加。OSM牵涉到重入处理，因为它不能仅基于前馈处理来解释。该研究方案的主要目标是系统地探索反复折返过程调节视觉知觉的知觉领域，以及这种调节如何受到注意在空间和时间上的分布的调节。这些领域包括基于刺激的形状感知，这些刺激由亮度轮廓、相对运动、分离纹理和双目视差定义。 AB期间的视觉搜索效率。由搜索功能的斜率(集合大小上的反应时间)索引的“弹出”显示中的视觉搜索效率是否受注意力的调节？早期的工作试图通过在AB期间注意力耗尽时提出搜索阵列来回答这个问题。由于各种方法论的原因，这项工作未能产生一个明确的答案。例如，在一种情况下，在AB序列中充当第二目标(T2)的搜索阵列保持在屏幕上直到响应，从而允许在T1已经被处理之后全神贯注地搜索该阵列。在另一种情况下，由于仅使用单一集合大小，因此无法评估搜索效率。在一项避免了早期工作中缺陷的试点研究中，我发现，在AB过程中注意力耗尽时，以搜索斜率为索引的弹出搜索的效率会受到损害。我将用不同类型的搜索(特征搜索、复合搜索、联合搜索)来研究这一发现，并将继续探讨其含义，即建立特征梯度(例如，纹理分割)需要注意。 视觉中的时间整合。很久以前，我当时的研究生约翰·霍本和我设计了一种在视觉中研究时间整合的方法。连续显示点阵图案的两半，由空白刺激间间隔(ISI)隔开。当ISI增加到100ms时，两个半部分的知觉整合会恶化，并最终崩溃。在一项试点研究中，我改变了模式的两个部分的持续时间，从20毫秒到100毫秒，并使用了阶梯阈值跟踪程序(PEST)，该程序动态调整ISI，以便收敛到一个关键的ISI，产生80%的准确率。我发现，在超过约60ms的曝光时间内，需要高达100ms的负ISI才能达到80%的准确率。也就是说，一种模式，当它作为一个整体短暂地展示时，很容易被察觉到，当它之前和之后是它自身的部分时，它就被摧毁了。我打算继续这个有趣而又违反直觉的发现。 前两个项目还将涉及记录事件相关电位，以研究由各种刺激序列触发的大脑电活动。最终目标是将所有三个项目的结果包含在一个基于迭代重入处理的综合理论中。

I plan to work in three distinct but related areas: ITERATIVE-REENTRY MODEL. Advances in neuroscience have called attention to reentrant signalling as the main form of communication between brain regions connected by two-way pathways. Yet, theories of perception continue to be couched in feed-forward sequences. We have developed an alternative account in which perceptions are held to emerge from iterative exchanges between brain regions. That account emerged from the discovery of Object-Substitution Masking (OSM) in which an initial combined display of target and mask continues with the mask alone. The target is identified easily when all display elements terminate together. When the mask alone remains on view, masking occurs, with its strength increasing with the duration of the mask. OSM implicates reentrant processing because it cannot be explained on the basis of feed-forward processes alone. The main objective of the research programme is to explore systematically the perceptual domains in which iterative reentrant processes mediate visual perception, and how that mediation is modulated by the distribution of attention in space and time. These domains include the perception of form based on stimuli that are defined by luminance contours, by relative motion, by segregated texture, and by binocular disparity. EFFICIENCY OF VISUAL SEARCH DURING THE AB. Is the efficiency of visual search in “pop-out” displays, as indexed by the slope of the search function (reaction time over set size), modulated by attention? Earlier work sought to answer this question by presenting the search array while attention was depleted during the period of the AB. For various methodological reasons, that work failed to produce an unequivocal answer. For example, in one case, the search array that acted as the second target (T2) in the AB sequence remained on the screen until response, thus allowing the array to be searched with full attention after T1 had been processed. In another case, search efficiency could not be assessed because only a single set size was used. In a pilot study that avoided the flaws in the earlier work, I found that the efficiency of pop-out search, as indexed by the search slope, is impaired while attention is depleted during the AB. I will examine this finding with different types of search (feature, compound, conjunction search), and will pursue its implication that the establishment of a feature gradient (e.g., texture segmentation) requires attention. TEMPORAL INTEGRATION IN VISION. Quite some time ago, my then graduate student John Hogben and I devised a method of studying temporal integration in vision. Two halves of a dot-matrix pattern are displayed successively, separated by a blank inter-stimulus interval (ISI). Perceptual integration of the two halves deteriorates, and eventually breaks down, as the ISI is increased to 100 ms. In a pilot study, I varied the duration of the two halves of the pattern from 20 to 100 ms, and used a staircase threshold-tracking procedure (PEST) that adjusted the ISI dynamically so as to converge on a critical ISI that yielded an accuracy of 80%. I found that, at exposure durations beyond about 60 ms, a NEGATIVE ISI of up to 100 ms was required to achieve an accuracy of 80%. Namely, a pattern that was easily perceived when displayed briefly as a whole, was destroyed when it was preceded and followed by parts of itself. I plan to pursue this fascinating and counterintuitive finding. The first two projects will also involve the recording of event-related potentials to study the brain’s electrical activity triggered by various stimulus sequences. The ultimate objective is to encompass the results of all three projects within a comprehensive theory based on iterative reentrant processing.