Cortical dynamics of contextually cued attentive visual learning and search: spatial and object evidence accumulation

How do humans use target-predictive contextual information to facilitate visual search? How are consistently paired scenic objects and positions learned and used to more efficiently guide search in familiar scenes? For example, humans can learn that a certain combination of objects may define a context for a kitchen and trigger a more efficient search for a typical object, such as a sink, in that context. The ARTSCENE Search model is developed to illustrate the neural mechanisms of such memory-based context learning and guidance and to explain challenging behavioral data on positive-negative, spatial-object, and local-distant cueing effects during visual search, as well as related neuroanatomical, neurophysiological, and neuroimaging data. The model proposes how global scene layout at a first glance rapidly forms a hypothesis about the target location. This hypothesis is then incrementally refined as a scene is scanned with saccadic eye movements. The model simulates the interactive dynamics of object and spatial contextual cueing and attention in the cortical What and Where streams starting from early visual areas through medial temporal lobe to prefrontal cortex. After learning, model dorsolateral prefrontal cortex (area 46) primes possible target locations in posterior parietal cortex based on goal-modulated percepts of spatial scene gist that are represented in parahippocampal cortex. Model ventral prefrontal cortex (area 47/12) primes possible target identities in inferior temporal cortex based on the history of viewed objects represented in perirhinal cortex.     

Object-position binding in visual memory for natural scenes and object arrays

Nine experiments examined the means by which visual memory for individual objects is structured into a larger representation of a scene. Participants viewed images of natural scenes or object arrays in a change detection task requiring memory for the visual form of a single target object. In the test image, 2 properties of the stimulus were independently manipulated: the position of the target object and the spatial properties of the larger scene or array context. Memory performance was higher when the target object position remained the same from study to test. This same-position advantage was reduced or eliminated following contextual changes that disrupted the relative spatial relationships among contextual objects (context deletion, scrambling, and binding change) but was preserved following contextual change that did not disrupt relative spatial relationships (translation). Thus, episodic scene representations are formed through the binding of objects to scene locations, and object position is defined relative to a larger spatial representation coding the relative locations of contextual objects.     

The role of context on boundary extension

Boundary extension (BE) is a memory error in which observers remember more of a scene than they actually viewed. This error reflects one’s prediction that a scene naturally continues and is driven by scene schema and contextual knowledge. In two separate experiments we investigated the necessity of context and scene schema in BE. In Experiment 1, observers viewed scenes that either contained semantically consistent or inconsistent objects as well as objects on white backgrounds. In both types of scenes and in the no-background condition there was a BE effect; critically, semantic inconsistency in scenes reduced the magnitude of BE. In Experiment 2 when we used abstract shapes instead of meaningful objects, there was no BE effect. We suggest that although scene schema is necessary to elicit BE, contextual consistency is not required.     

Anticipation in Real‐World Scenes: The Role of Visual Context and Visual Memory

The human sentence processor is able to make rapid predictions about upcoming linguistic input. For example, upon hearing the verb eat, anticipatory eye‐movements are launched toward edible objects in a visual scene (Altmann & Kamide, 1999). However, the cognitive mechanisms that underlie anticipation remain to be elucidated in ecologically valid contexts. Previous research has, in fact, mainly used clip‐art scenes and object arrays, raising the possibility that anticipatory eye‐movements are limited to displays containing a small number of objects in a visually impoverished context. In Experiment 1, we confirm that anticipation effects occur in real‐world scenes and investigate the mechanisms that underlie such anticipation. In particular, we demonstrate that real‐world scenes provide contextual information that anticipation can draw on: When the target object is not present in the scene, participants infer and fixate regions that are contextually appropriate (e.g., a table upon hearing eat). Experiment 2 investigates whether such contextual inference requires the co‐presence of the scene, or whether memory representations can be utilized instead. The same real‐world scenes as in Experiment 1 are presented to participants, but the scene disappears before the sentence is heard. We find that anticipation occurs even when the screen is blank, including when contextual inference is required. We conclude that anticipatory language processing is able to draw upon global scene representations (such as scene type) to make contextual inferences. These findings are compatible with theories assuming contextual guidance, but posit a challenge for theories assuming object‐based visual indices.     

Consistency effects between objects in scenes

How does context influence the perception of objects in scenes? Objects appear in a given setting with surrounding objects. Do objects in scenes exert contextual influences on each other? Do these influences interact with background consistency? In three experiments, we investigated the role of object-to-object context on object and scene perception. Objects (Experiments 1 and 3) and backgrounds (Experiment 2) were reported more accurately when the objects and their settings were consistent than when they were inconsistent, regardless of the number of foreground objects. In Experiment 3, related objects (from the same setting) were reported more accurately than were unrelated objects (from different settings), independently of consistency with the background. Consistent with an interactive model of scene processing, both object-to-object context and object-background context affect object perception.     

Perceptual effects of scene context on object identification

Summary In a number of studies the context provided by a real-world scene has been claimed to have a mandatory, perceptual effect on the identification of individual objects in such a scene. This claim has provided a basis for challenging widely accepted data-driven models of visual perception in order to advocate alternative models with an outspoken top-down character. The present paper offers a review of the evidence to demonstrate that the observed scene-context effects may be the product of post-perceptual and task-dependent guessing strategies. A new research paradigm providing an on-line measure of genuine perceptual effects of context on object identification is proposed. First-fixation durations for objects incidentally fixated during the free exploration of real-world scenes are shown to increase when the objects are improbable in the scene or violate certain aspects of their typical spatial appearance in it. These effects of contextual violations are shown to emerge only at later stages of scene exploration, contrary to the notion of schema-driven scene perception effective from the very first scene fixation. In addition, evidence is reported in support of the existence of a facilitatory component in scene-context effects. This is taken to indicate that the context directly affects the ease of perceptual object processing and does not merely serve as a framework for checking the plausibility of the output of perceptual processes. Finally, our findings are situated against other contrasting results. Some future research questions are highlighted.     

Object–scene inconsistencies do not capture gaze: evidence from the flash-preview moving-window paradigm

In the present study, we investigated the influence of object-scene relationships on eye movement control during scene viewing. We specifically tested whether an object that is inconsistent with its scene context is able to capture gaze from the visual periphery. In four experiments, we presented rendered images of naturalistic scenes and compared baseline consistent objects with semantically, syntactically, or both semantically and syntactically inconsistent objects within those scenes. To disentangle the effects of extrafoveal and foveal object-scene processing on eye movement control, we used the flash-preview moving-window paradigm: A short scene preview was followed by an object search or free viewing of the scene, during which visual input was available only via a small gaze-contingent window. This method maximized extrafoveal processing during the preview but limited scene analysis to near-foveal regions during later stages of scene viewing. Across all experiments, there was no indication of an attraction of gaze toward object-scene inconsistencies. Rather than capturing gaze, the semantic inconsistency of an object weakened contextual guidance, resulting in impeded search performance and inefficient eye movement control. We conclude that inconsistent objects do not capture gaze from an initial glimpse of a scene.     

真实场景视觉搜索中的背景线索效应特点与机制

背景线索效应揭示了个体在视觉搜索过程中对刺激之间具有的稳定空间关系（刺激间不变的相对空间位置）的学习能够提高搜索效率。本文基于经典背景线索效应在内隐习得空间布局的机制下结合真实场景视觉搜索的相关理论，对真实场景背景线索效应的实验范式、学习性质与内容进行归纳梳理，将真实场景视觉搜索中影响背景线索效应的视觉信息分为低水平物理特征及高水平语义信息两个维度进行论述。虽然当前研究涉及真实场景背景线索效应在不同场景维度信息的加工机制，但对于发挥作用的场景信息类别以及作用阶段还较少涉及，未来研究仍需进一步的探讨。     

The effect of spatial and nonspatial contextual information on visual object memory

Recent research has found visual object memory can be stored as part of a larger scene representation rather than independently of scene context. The present study examined how spatial and nonspatial contextual information modulate visual object memory. Two experiments tested participants’ visual memory by using a change detection task in which a target object's orientation was either the same as it appeared during initial viewing or changed. In addition, we examined the effect of spatial and nonspatial contextual manipulations on change detection performance. The results revealed that visual object representations can be maintained reliably after viewing arrays of objects. Moreover, change detection performance was significantly higher when either spatial or nonspatial contextual information remained the same in the test image. We concluded that while processing complex visual stimuli such as object arrays, visual object memory can be stored as part of a comprehensive scene representation, and both spatial and nonspatial contextual changes modulate visual memory retrieval and comparison.     

Guidance of attention to objects and locations by long-term memory of natural scenes

Four flicker change-detection experiments demonstrate that scene-specific long-term memory guides attention to both behaviorally relevant locations and objects within a familiar scene. Participants performed an initial block of change-detection trials, detecting the addition of an object to a natural scene. After a 30-min delay, participants performed an unanticipated 2nd block of trials. When the same scene occurred in the 2nd block, the change within the scene was (a) identical to the original change, (b) a new object appearing in the original change location, (c) the same object appearing in a new location, or (d) a new object appearing in a new location. Results suggest that attention is rapidly allocated to previously relevant locations and then to previously relevant objects. This pattern of locations dominating objects remained when object identity information was made more salient. Eye tracking verified that scene memory results in more direct scan paths to previously relevant locations and objects. This contextual guidance suggests that a high-capacity long-term memory for scenes is used to insure that limited attentional capacity is allocated efficiently rather than being squandered.     

Testing a conceptual locus

Changes to objects that are inconsistent with the scene in which they appear are detected more accurately than changes to consistent objects. In three experiments, we tested whether this inconsistent object advantage derives from the differential retention of conceptual codes generated from a brief view of a real-world scene in accordance with a conceptual short-term memory (CSTM) hypothesis. A scene was presented for 250 msec, followed by a brief mask and a test scene in which a target object was either changed or not changed. In Experiment 1, changes that altered conceptual content (object deletion) were contrasted with visual changes (left-right orientation changes). In Experiment 2, the duration of the mask was manipulated to vary the amount of time available for conceptual consolidation of the initial scene. In Experiment 3, the type of mask was manipulated: Either a meaningless pattern mask or a meaningful, and thus conceptually disruptive, scene was shown. The inconsistent object advantage was obtained in each experiment, yet in none was it modulated in the direction predicted by the CSTM hypothesis. Instead, the inconsistent object advantage is likely to be caused by contextual influence on memory for visual object representations.     

How do targets,nontargets, and scene context influence real-world object detection?

Humans excel at finding objects in complex natural scenes, but the features that guide this behaviour have proved elusive. We used computational modeling to measure the contributions of target, nontarget, and coarse scene features towards object detection in humans. In separate experiments, participants detected cars or people in a large set of natural scenes. For each scene, we extracted target-associated features, annotated the presence of nontarget objects (e.g., parking meter, traffic light), and extracted coarse scene structure from the blurred image. These scene-specific values were then used to model human reaction times for each novel scene. As expected, target features were the strongest predictor of detection times in both tasks. Interestingly, target detection time was additionally facilitated by coarse scene features but not by nontarget objects. In contrast, nontarget objects predicted target-absent responses in both person and car tasks, with contributions from target features in the person task. In most cases, features that speeded up detection tended to slow down rejection. Taken together, these findings demonstrate that humans show systematic variations in object detection that can be understood using computational modeling.     

Local and global contextual constraints on the identification of objects in scenes.

Objects likely to appear in a given real-world scene are frequently found to be easier to recognize. Two different sources of contextual information have been proposed as the basis for this effect: global scene background and individual companion objects. The present paper examines the relative importance of these two elements in explaining the context-sensitivity of object identification in full scenes. Specific sequences of object fixations were elicited during free scene exploration, while fixation times on designated target objects were recorded as a measure of ease of target identification. Episodic consistency between the target, the global scene background, and the object fixated just prior to the target (the prime), were manipulated orthogonally. Target fixation times were examined for effects of prime and background. Analyses show effects of both factors, which are modulated by the chronology and spatial extent of scene exploration. The results are discussed in terms of their implications for a model of visual object recognition in the context of real-world scenes.     

The influence of color on the perception of scene gist

In 3 experiments the authors used a new contextual bias paradigm to explore how quickly information is extracted from a scene to activate gist, whether color contributes to this activation, and how color contributes, if it does. Participants were shown a brief presentation of a scene followed by the name of a target object. The target object could be consistent or inconsistent with scene gist but was never actually present in the scene. Scene gist activation was operationalized as the degree to which participants respond "yes" to consistent versus inconsistent objects, reflecting a response bias produced by scene gist. Experiment 1 demonstrated that scene gist is activated after a 42-ms exposure and that the strength of the activation increases with longer presentation durations. Experiments 2 and 3 explored the contribution of color to the activation of scene gist. The results revealed that color has an influence across a wide variety of scenes and is directly associated with scene gist.     

场景知觉及其研究范式

场景知觉关注的是人如何知觉和加工复杂的真实环境信息。场景包括物体和背景两个重要的组成部分,根据复杂性与真实性程度,场景刺激材料可以分为三种不同的类型。已有研究主要从自上而下和自下而上两种方式来解释场景知觉中信息的提取和加工,也有研究试图从二者交互的角度来进行解释。此外,基于不同的实验目的和技术,研究者分别采用了眼动、背景提示、物体觉察、变化觉察和点线索追随几种不同的研究范式来探讨场景信息的知觉问题。场景知觉研究在场景的定义、不同范式间的整合、研究内部效度和不同加工阶段的加工方式四个方面还需要进一步深入和探讨     

Categorical implicit learning in real-world scenes: evidence from contextual cueing

The present study examined the extent to which learning mechanisms are deployed on semantic-categorical regularities during a visual searching within real-world scenes. The contextual cueing paradigm was used with photographs of indoor scenes in which the semantic category did or did not predict the target position on the screen. No evidence of a facilitation effect was observed in the predictive condition compared to the nonpredictive condition when participants were merely instructed to search for a target T or L (Experiment 1). However, a rapid contextual cueing effect occurred when each display containing the search target was preceded by a preview of the scene on which participants had to make a decision regarding the scene's category (Experiment 2). A follow-up explicit memory task indicated that this benefit resulted from implicit learning. Similar implicit contextual cueing effects were also obtained when the scene to categorize was different from the subsequent search scene (Experiment 3) and when a mere preview of the search scene preceded the visual searching (Experiment 4). These results suggested that if enhancing the processing of the scene was required with the present material, such implicit semantic learning can nevertheless take place when the category is task irrelevant.     

Contextual guidance of eye movements and attention in real-world scenes: the role of global features in object search

Many experiments have shown that the human visual system makes extensive use of contextual information for facilitating object search in natural scenes. However, the question of how to formally model contextual influences is still open. On the basis of a Bayesian framework, the authors present an original approach of attentional guidance by global scene context. The model comprises 2 parallel pathways; one pathway computes local features (saliency) and the other computes global (scene-centered) features. The contextual guidance model of attention combines bottom-up saliency, scene context, and top-down mechanisms at an early stage of visual processing and predicts the image regions likely to be fixated by human observers performing natural search tasks in real-world scenes.     

Human heading judgments in the presence of moving objects

When moving toward a stationary scene, people judge their heading quite well from visual information alone. Much experimental and modeling work has been presented to analyze how people judge their heading for stationary scenes. However, in everyday life, we often move through scenes that contain moving objects. Most models have difficulty computing heading when moving objects are in the scene, and few studies have examined how well humans perform in the presence of moving objects. In this study, we tested how well people judge their heading in the presence of moving objects. We found that people perform remarkably well under a variety of conditions. The only condition that affects an observer’s ability to judge heading accurately consists of a large moving object crossing the observer’s path. In this case, the presence of the object causes a small bias in the heading judgments. For objects moving horizontally with respect to the observer, this bias is in the object’s direction of motion. These results present a challenge for computational models.     

Visual context modulates potentiation of grasp types during semantic object categorization

Substantial evidence suggests that conceptual processing of manipulable objects is associated with potentiation of action. Such data have been viewed as evidence that objects are recognized via access to action features. Many objects, however, are associated with multiple actions. For example, a kitchen timer may be clenched with a power grip to move it but pinched with a precision grip to use it. The present study tested the hypothesis that action evocation during conceptual object processing is responsive to the visual scene in which objects are presented. Twenty-five healthy adults were asked to categorize object pictures presented in different naturalistic visual contexts that evoke either move- or use-related actions. Categorization judgments (natural vs. artifact) were performed by executing a move- or use-related action (clench vs. pinch) on a response device, and response times were assessed as a function of contextual congruence. Although the actions performed were irrelevant to the categorization judgment, responses were significantly faster when actions were compatible with the visual context. This compatibility effect was largely driven by faster pinch responses when objects were presented in use-compatible, as compared with move-compatible, contexts. The present study is the first to highlight the influence of visual scene on stimulus–response compatibility effects during semantic object processing. These data support the hypothesis that action evocation during conceptual object processing is biased toward context-relevant actions.