The role of stimulus characteristics in visual attention: an eye-tracking study

Abstract

Background

Visual attention in complex environments is shaped by multiple stimulus characteristics, yet most studies have examined only one or two features at a time. Eye-tracking combined with a change blindness paradigm offers a sensitive way to evaluate how different visual features guide the detection of changes in naturalistic scenes. This study aimed to validate visual stimuli with varying characteristics and to determine which factors are most significantly associated with visual attention.

Methods

Fifty university students viewed pairs of real-world scenes while their gaze behavior was recorded using an eye tracking system. Participants were required to detect a single changed item across image pairs. Four visual characteristics were manipulated: social versus nonsocial content, number of distractors, congruency, and level of interest (central vs. marginal).

Results

Response times varied significantly across all characteristics. Faster detection was observed for social stimuli, scenes with fewer distractors, congruent changes, and items of central interest. Linear mixed-effects models (LMMs) revealed that level of interest was the strongest predictor of response time, followed by the number of distractors, congruency, and social content.

Conclusion

Eye-tracking measures, particularly response time to detect change, demonstrated that central interest and fewer distractors were the strongest predictors of visual attention. These findings accentuate the complexity of visual attention and highlight its potential impact on interpreting environmental stimuli and the broader effects on academic performance and daily functioning.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-026-04785-y.

Introduction

Visual attention profoundly shapes our experience and perception of the world, playing a crucial role in detecting significant environmental changes vital for daily tasks, such as web browsing, road safety, and grocery shopping [1–5]. It also affects social interactions [6, 7]. Visual attention enables individuals to filter and prioritize information from visual stimuli encountered at any moment, discerning relevant from irrelevant data. This allows effective engagement and response to surroundings [8, 9]. Efficient visual attention is integral to cognitive and behavioral development and to daily functioning [10, 11].

While observing a scene, individuals focus on specific information and shift their gaze between points, fixating on crucial details. The brain integrates these details to create a more complete image of reality [12]. Because individuals do not retain a complete representation of all visual information in the environment from one moment to the next, attention to an item is necessary to detect changes in it [13]. Change detection primarily pertains to the visual processes involved in first noticing that a change has occurred, including identifying and localizing this change [14].

Four key measurements are often used to assess change detection using eye-tracking techniques: time to first fixation (TFF), total fixation duration (TFD), response time (RT), and visits. The TFF metric measures the time it takes for a participant’s gaze to first fixate on the changing item, quantifying how rapidly the item captures attention. This metric differs from RT, which includes additional decision-making processes beyond initial attention capture. In contrast, TFD measures the cumulative time a participant fixates on the target item, indicating the time required for visual information processing after the item enters the attentional span. Finally, visits represent the frequency and duration of returns to the target item. This offers a broader perspective on how participants allocate attention across the scene [15, 16]. Together, these metrics provide a comprehensive understanding of attentional priority, processing, and decision-making during visual tasks.

The change blindness (CB) paradigm is effective for assessing visual attention during change detection [17]. It refers to a perceptual phenomenon in which an individual finds it challenging to detect scene changes during some interruption [12, 18]. These interruptions could include a blank screen (flicker), a series of mud splashes occluding parts of the scene, or a film cut [19–21]. The flicker method presents two identical images in sequence, differing only by one item, with a blank screen separating them. When the blank screen lasts at least 80 ms, it hinders the visual system’s ability to detect the change [20, 22, 23]. The flicker method simulates the blinking that occurs numerous times per second in the human visual system. The ability to discern significant changes is crucial for daily functioning; for example, brief disruptions, such as a blink, may impact monitoring traffic lights or hazardous events in tasks like driving [24].

Individuals’ ability to detect change depends on several characteristics of the stimulus, such as the number of distractors represented in the stimulus [25, 26], the degree of interest in the object that changes [12], the congruency of the changed object [27, 28], and the characteristics of social and nonsocial stimuli [29].

Wolfe and Horowitz [30] found that fewer stimuli enhance change detection. Similarly, Lleras et al. [31] reported a linear relationship between the number of distractors and response time (RT) to detect change, with RT increasing as the number of distractors increased. Studies using real-scene images also showed shorter detection times in images with fewer distractors [26]. In the context of change detection, the number of distractors reflects overall visual load and scene complexity, increasing attentional competition and thereby affecting the efficiency with which the changing item is detected [25, 30].

Beyond the influence of distractor quantity on detection performance, the inherent relevance of the changing item within the visual scene also plays a critical role in guiding attention. Rensink et al. [12] introduced the concept of level of interest regarding the changing item’s relevance to the image’s context, on visual attention being either of central interest (CI) or marginal interest (MI), to the image (e.g., Fig. 1). Studies have shown that changes to CI items are detected more easily than changes to MI items [22, 24, 32]. Eye-tracking is also widely used to examine how the object’s contextual role shapes the distribution of gaze within a scene. Research in this field has shown that scene context guides eye-movement patterns, influencing both where viewers look and how efficiently they detect changes [33, 34].

Fig. 1.

Examples of the stimulus sequence presenting the original image and its modified version. The AOI for the changed item is circled

Taken together, this indicates that attention is shaped not only by an object’s functional role in the scene but also by the extent to which it aligns with the scene’s expected meaning, a dimension known as congruence which refers to how well an object matches the expected semantic context of a scene (e.g., an umbrella versus a phone on a desk) [35]. Items that violate contextual expectations (incongruent) tend to draw increased attention and are often detected more rapidly during change detection tasks [32, 36]. In contrast, when searching for an object in a familiar environment, there is a tendency to allocate attention toward familiar and expected objects as a primary focus (i.e., congruent objects) [2, 9]. Thus, the individual’s knowledge about the congruency of the changed object affects detection efficiency [37].

The final characteristic examined in this study that may influence change detection relates to whether the stimulus contains social information (e.g., people, faces) or nonsocial content (e.g., objects, scenes). Allocating attention to relevant stimuli in the social world, such as perceiving changes in facial movements, including eye gaze, expressions, and lip movements, enables the recognition of social cues and the processing of social information [38, 39], which are effective for social communication [40, 41]. Visual attention, therefore, plays a central role in social functioning by directing focus toward significant events within the shared environment [42], enabling the detection of social cues, monitoring facial expressions and body language, and attending to the speaker and their expressions to promote effective engagement in social interactions [19, 43]. Consistent with these processes, developmental literature shows that even in early childhood, attention is naturally drawn to social elements such as faces and social–emotional cues [44, 45]. Visual-search and eye-tracking studies further demonstrate that faces strongly capture attention and that adults preferentially fixate on social over nonsocial stimuli [29, 46, 47].

In summary, previous findings have shown that change detection is influenced by various stimulus characteristics, including the number of distractors, the level of interest in the changing object, the congruency of the changed item, and whether the stimulus contains social or nonsocial information. The four stimulus characteristics examined in the present study were deliberately selected to capture distinct, yet complementary higher-level semantic mechanisms previously shown to influence visual attention in change-detection studies, and to enable the systematic examination of their combined effects within ecologically valid scenes. While numerous low-level visual features (e.g., luminance contrast and color) are known to influence gaze behavior [25, 26], in the current study these bottom-up factors were controlled to the extent possible, with focus placed on top-down dimensions reflecting contextual relevance (level of interest), visual complexity (number of distractors), semantic expectancy (congruency), and social meaningfulness (social context). However, most existing studies have examined the impact of only one or two stimulus characteristics presented simultaneously on visual attention performance, despite the inherent complexity of real-world environments in which multiple sources of information compete for attention. The present study addresses this gap by systematically integrating these four key stimulus characteristics within a single paradigm using ecologically valid, real-world contexts.

This integrated approach enables the examination of the relative and combined contributions of these characteristics, providing a more comprehensive framework for understanding visual attention in complex environments. The current study’s two main goals were to (a) develop and validate visual stimuli with different characteristics incorporated within an eye-tracking tool for measuring visual attention based on known effects in the literature and (b) investigate which characteristics have the most significant influence on visual attention.

Based on these goals, we hypothesized that stimulus characteristics would differentially influence visual attention. We assumed longer RT’s to detect the change, more area of interest (AOI) visits, first fixation latency, and a longer duration on the AOI in images that contain nonsocial stimuli (vs. social stimuli), have many distractors (vs. few distractors), contain noncongruent change (vs. congruent change), and are marginal to the images’ context (vs. central to the images’ context). The validation of the stimuli used in this study allowed us to formulate an explorational hypothesis regarding the relative contribution of each characteristic to visual attention abilities.

Materials and methods

Participants

Fifty university students enrolled in diverse departments (38 women; age M = 23 years, SD = 2.2, range = 19–31; see Table 1) were recruited via a convenience sample through social media announcements (recruitment period: May 3, 2022 – June 23, 2022). A priori power analysis with G*Power (power = 0.90, α = 0.05) was conducted to determine the target sample size. Inclusion criteria included normal or corrected-to-normal vision and average or above average verbal and nonverbal intelligence quotients (≤ 70) derived from the Peabody Picture Vocabulary Test-3 [48] and Raven’s Progressive Matrices [49]. Exclusion criteria included physical impairments restricting computer-mouse use or known vision impairments (except for short-sightedness corrected with eyeglasses). We excluded one participant due to incompatibility with the inclusion criteria. The [Blinded] University Ethical Board Committee granted ethical approval for this study. Participants provided signed informed consent to participate in the study and to allow their data to be used anonymously for publication. They attended one session (which lasted about 90 min) at the researchers’ lab. Participants were paid the equivalent of US$30 upon completing the experiment and questionnaires.

Table 1.

Participant characteristics

Characteristic	Minimum	Maximum	M	SD
Age (years; months)	19.08	31.03	23.43	2.240
RPM (%)	40.00	59.00	50.71	4.359
PPVT-III	66.00	113.00	92.76	9.578

RPM  Raven Progressive Matrices, PPVT-III  Peabody Picture Vocabulary Test-Third Edition

N = 49

Materials

We recorded participants’ gaze behavior using a Tobii X60 eye-tracker (www.tobii.com), interfaced and controlled on a Dell 15-inch screen running via the Tobii Studio package. The eye-tracking system is noninvasive and does not constrain head or body movements. The system tracked both eyes with an accuracy of 0.5° and sampled at 60 Hz. Gaze data were processed in Tobii Pro Lab using the default I-VT fixation filter; blink events and invalid samples were automatically excluded prior to metric extraction. The stimuli included 64 paired colored images of real-life scenes, including objects and people. The image set was designed to represent common everyday contexts (e.g., home, school/work, and community environments), reflecting the role of environmental context in guiding attentional allocation in real-world settings [50]. Each pair of images was identical except for a single change in the presence or absence of a specific item. They were presented in a flicker paradigm, which has proven effective in visual attention studies [12]. Each image was randomly displayed at 360 ms, separated by a black screen for 120 ms. A fixation cross appeared at the beginning of each image sequence (Fig. 1).

The image sequence lasted 48 s. or until the participant detected the change. We instructed participants to click the mouse and state the identified change, which was audio-recorded. We marked the AOI around the changing item, following Holmqvist et al.’s [16] hand-drawn approach, with a 5 mm (0.5°) margin to account for the system’s accuracy. All the images were selected from an open image database for editing (https://depositphotos.com). Adobe Photoshop was used to modify them to equal size (500 × 700 pixels) and to adjust brightness, contrast, and the size of the changing object, following the criteria established by Rensink et al. [12].

We divided the 64 images into four subgroups according to the following characteristics: (a) the number of distractors in each image (i.e., many/few), (b) the level of interest for the changed object (CI/ MI), (c) congruent or noncongruent of change occurrence, and (d) social and nonsocial stimuli (Fig. 2). Each subgroup included four trials per condition according to a matrix specifically designed for this study (Table 2). Figures 2; Table 2 clarify the design of the stimuli. Figure 2 provides examples of each characteristic in isolation, while Table 2 presents the complete factorial matrix in which all four characteristics (sociality, level of interest, number of distractors, and congruency) were manipulated simultaneously. This design enabled a direct comparison of their relative contributions to visual attention.

Fig. 2.

Example of the image categorization into four subgroups according to the described characteristics. Note. These examples show the categorization process used before constructing the complete factorial matrix, in which all four characteristics (sociality, level of interest, number of distractors, and congruency) were manipulated simultaneously (as shown in Table 2), e.g., the picture classified here as social according to the matrix was classified as social-central-many-congruent

Table 2.

Full factorial matrix: each cell represents a combination of all four stimulus characteristics (4 Images per Cell)

		Social		Nonsocial
Distractors	Congruency	Central	Marginal	Central	Marginal
Few	Congruent	Social, Central, Few, Congruent	Social, Marginal, Few, Congruent	Nonsocial, Central, Few, Congruent	Nonsocial, Marginal, Few, Congruent
	Noncongruent	Social, Central, Few, Incongruent	Social, Marginal, Few, Incongruent	Nonsocial, Central, Few, Incongruent	Nonsocial, Marginal, Few, Incongruent
Many	Congruent	Social, Central, Many, Congruent	Social, Marginal, Many, Congruent	Nonsocial, Central, Many, Congruent	Nonsocial, Marginal, Many, Congruent
	Noncongruent	Social, Central, Many, Incongruent	Social, Marginal, Many, Incongruent	Nonsocial, Central, Many, Incongruent	Nonsocial, Marginal, Many, Incongruent

Social stimuli were operationalized as scenes depicting a clear, observable social interaction among two or more individuals. These scenes included visible human figures engaged in interpersonal activity. In contrast, nonsocial stimuli consisted of scenes that did not include any people and contained only inanimate objects. Importantly, in all conditions, the change always occurred in an object and never in a person. Thus, the manipulation concerned the presence or absence of social interaction within the scene context, rather than changes to the social agents themselves.

Congruency was operationalized as the semantic plausibility of the disappearance of the object type, rather than its consistency within the scene. A change was considered congruent when it involved the disappearance of an object whose temporary absence is generally plausible in everyday contexts (e.g., a ball that may roll away or be removed). A change was considered incongruent when it involved the disappearance of an object whose absence would be inherently unlikely due to its stable or structural nature (e.g., a tree or permanent environmental element). Thus, the manipulation concerned the plausibility of the disappearance event itself rather than scene-object consistency.

In the present study, central interest refers to semantic or narrative relevance within the scene rather than spatial position within the image. Classification was based solely on independent judgments of contextual importance and not on geometric location. Images were classified according to level of interest (Central vs. Marginal) and number of distractors (Few vs. Many) using a structured pilot classification procedure. Five experienced therapists and researchers viewed all images prior to the main experiment. For each scene, they were asked to (a) briefly describe the scene and indicate which elements were central or marginal to its overall meaning, and (b) determine whether the image contained “many” or “few” items while enumerating the elements they perceived as potential distractors. Consistent with previous methodology [12, 24], images were categorized as Central Interest (CI) if at least three participants mentioned the changing object or region when describing the scene, whereas items not spontaneously reported were classified as Marginal Interest (MI). The cut-off for distractor quantity emerged empirically from this procedure: scenes consistently categorized as containing “many” items typically included more than seven enumerated elements, whereas those categorized as “few” included fewer than seven. These binary categorizations were selected to maintain the study’s balanced factorial design and enable systematic comparisons across stimulus conditions.

Procedure

First, we gave participants a brief overview of the experiment and seated them approximately 55–60 cm from the screen (a 15-inch computer monitor). We calibrated participants with the eye tracker using the default maximum of 9 points (calibration was accepted if less than 1° of visual angle error was achieved on both the x and y axes). The sequence alternated between a scene and its modified version, repeating until the participant clicked the mouse or 48 s. elapsed. The task instructions specified that participants should indicate the change in the scene as quickly as possible without guessing. The RT measure was recorded in milliseconds and ended when the participant clicked the mouse. Following the mouse click, the participant had to verbally identify and locate the changed item for the trial to count as a correct response. The incorrect responses were tagged as commission errors which refer to the number of trials where participants clicked but did not correctly identify the item that changed the scene and omission errors which refer to the number of trials where participants failed to detect the change. Both commission (0.68% of the total trials) and omission (3.84% of the total trials) errors were documented and excluded from the dataset prior to analysis. A detailed summary of error rates by stimulus configuration, calculated as a proportion of the total number of trials in each condition, is provided in the Supplementary Materials (Table S1). Inspection of the descriptive analysis interestingly indicates higher error rates for both marginal and congruency variables images inconsistent with the sociability and distractor variables. Similar patterns have been reported in previous studies [2, 12].

Two practice trials were conducted for each participant before the actual experiment started. Four dependent measures were documented: (a) RT (i.e., how quickly the participants detected the change); (b) TFF on the changed item (i.e., first-fixation latency); (c) TFD on the changed item (i.e., total time fixated on the changed item); and (d) number of visits, reflecting the number of times eye gaze fixated on the AOI before detection (i.e., when participants saw an item but did not pay immediate attention to it or detect the change).

Statistical analyses were conducted in R (R Foundation for Statistical Computing, Vienna, Austria, 2023) [51]. Linear mixed-effects models (LMMs) were fitted using the lme4 package. To examine the effects of stimulus characteristics on visual attention, two complementary analyses were conducted. First, we tested whether each stimulus characteristic significantly influenced visual attention, thereby validating the stimulus set and the eye-tracking measures used in the task (Hypothesis 1). Second, we examined the relative contribution of the stimulus characteristics when entered simultaneously into the same model (Hypothesis 2). Separate models were fitted for each dependent measure. Because these measures were strictly positive and showed right-skewed distributions, outcome variables were analyzed on the natural logarithm scale. All models included social context (social vs. nonsocial), level of interest (central vs. marginal), congruency (congruent vs. non-congruent), and number of distractors (few vs. many) as fixed effects. Participants were included as a random intercept to account for repeated observations nested within individuals. Stimulus identity (image) was not included as a random effect because each image appeared in only one experimental condition and was therefore inseparable from the independent variables. To reduce potential random variability associated with individual images, four images were included within each condition, representing the same experimental construct.

For Hypothesis 1, the effect of each stimulus characteristic on visual attention was evaluated using model estimates (β), standard errors, and Wald tests. For Hypothesis 2, the relative contribution of each stimulus characteristic was assessed using likelihood-ratio tests, comparing the full model with reduced models omitting each predictor in turn. Coefficients in Supplementary Tables S2–S5 are reported on the log scale because the models were estimated with a log parameterization; thus, effects were interpreted after exponentiation as multiplicative changes in the expected outcome (reported as -1 in % terms).

Results

To validate the stimulus set, we examined whether each stimulus characteristic influenced eye tracking measurements (RT, TFF, TFD and Visits). Linear mixed-effects models with a random intercept for participant were fitted to account for repeated observations within individuals (see Table 3). Results indicated that the four visual stimuli characteristics significantly differ in RT: social (b = 0.733, SE = 0.368, t(734.00) = 1.99, p = .0469; 95% CI 0.0112 to 1.456), indicating longer detection times for nonsocial scenes compared with social scenes. Level of interest (b = 4.181, SE = 0.336, t(734.00) = 12.46, p < .001; 95% CI 3.523 to 4.839), indicating longer detection times when the change involved MI compared to CI. Congruency (b = 2.103, SE = 0.361, t(734.00) = 5.82, p < .001; 95% CI 1.395 to 2.811), with incongruent changes detected more slowly than congruent changes. Finally, the number of distractors (b = 2.906, SE = 0.354, t(734.00) = 8.22, p < .001; 95% CI 2.213 to 3.599), indicating longer detection times in scenes containing many distractors compared with few distractors (for means and standard errors of all four measurements see Fig. 3a-d). These results confirm that all four stimulus characteristics significantly affected change detection performance.

Table 3.

Linear mixed-effects model estimates for eye-tracking measures (RT, TFD, TFF, Visits)

Outcome	Predictor	b	SE	p	95% CI
RT	Social	0.733	0.368	0.047	0.011–1.456
	Level of Interest	4.181	0.336	< 0.001	3.523–4.839
	Congruency	2.103	0.361	< 0.001	1.395–2.811
	Distractors	2.906	0.354	< 0.001	2.213–3.599
TFD	Social	0.099	0.029	< 0.001	0.042–0.156
	Level of Interest	0.196	0.092	0.036	0.015–0.377
	Congruency	0.111	0.029	< 0.001	0.054–0.168
	Distractors	0.027	0.029	0.353	-0.084–0.030
TFF	Social	-0.016	0.011	0.147	-0.038–0.006
	Level of Interest	0.009	0.011	0.422	-0.013–0.031
	Congruency	0.010	0.011	0.354	-0.033–0.012
	Distractors	0.005	0.011	0.647	-0.017–0.027
Visits	Social	0.009	0.036	0.784	-0.060–0.071
	Level of Interest	0.167	0.035	< 0.001	0.098–0.236
	Congruency	0.250	0.034	< 0.001	0.183–0.317
	Distractors	0.117	0.035	0.001	0.047–0.186

RT, TFD and TFF are presented in milliseconds, and Visits are presented as a count of the number of visits

b  Fixed-effect estimate, SE Standard error, CI  Confidence interval, TFD  Total fixation duration, TFF  Time to first fixation

Fig. 3.

Means and Standard Errors of Eye Tracking Measurements (RT, TFF, TFD and Visits) for Each Characteristic of the Visual Stimuli. Note. S = Social; NS = Non-Social; F=Few; M=Many; Con= Congruent; NCon= Non-Congruent; CI = Central Interest; MI = Marginal Interest. RT, TFD and TFF are presented in milliseconds, Visits are presented as a count of the number of visits. *p < .05; **p < .01; ***p < .001

After establishing that each stimulus characteristic significantly influenced visual attention performance, we examined their relative contributions when entered simultaneously into the same model. Linear mixed-effects models were fitted with the four stimulus characteristics (social, level of interest, congruency, and number of distractors) included as fixed effects and participant included as a random intercept. For RT, level of interest emerged as the strongest predictor (χ² = 204.40, p < .001), indicating substantially longer detection times when the change occurred in marginal compared with central elements. This was followed by the number of distractors (χ² = 110.19, p < .001), congruency (χ² = 64.45, p < .001), and social context (χ² = 29.33, p < .001). A similar pattern emerged for TFD. The strongest contribution was observed for level of interest (χ² = 35.18, p < .001), followed by congruency (χ² = 26.41, p < .001) and social context (χ² = 10.11, p = .001). In contrast, the number of distractors did not significantly contribute to TFD (χ² = 0.25, p = .618). For TFF, none of the stimulus characteristics significantly contributed to the model. Social context approached significance (χ² = 3.51, p = .061), whereas level of interest (χ² = 0.65, p = .420), congruency (χ² = 0.02, p = .897), and number of distractors (χ² = 0.22, p = .638) did not significantly affect the latency of the first fixation. The analysis of visits revealed a somewhat different pattern. Congruency showed the strongest contribution (χ² = 55.80, p < .001), followed by level of interest (χ² = 23.01, p < .001) and number of distractors (χ² = 18.06, p < .001). In contrast, social context did not significantly contribute to the number of visits (χ² = 0.25, p = .617). Results of the model for all outcomes see Supplementary Table S2-S5.

In summary, these findings indicate that while all stimulus characteristics influence change detection performance, their relative contributions vary across different indices of visual attention.

Discussion

The current study investigated the association between the characteristics of simultaneously presented visual stimuli and visual attention using eye-tracking within a CB paradigm. The results highlight significant associations among the four examined visual stimuli characteristics (social, number of distractors, congruency, and level of interest) during change detection, referring to four eye-tracking measures: response time (RT), total fixation duration (TFD), time to first fixation (TFF), and number of visits to the area of interest (visits), to capture different aspects of attentional processing. Considering these measures together provides a more comprehensive understanding of how different stimulus characteristics shape visual attention during change detection.

This discussion addresses each characteristic individually to establish external validation of the visual stimuli by showing results that align with similar findings in the literature. It then analyzes the relative contributions of each characteristic when all four are addressed simultaneously, highlighting the novelty of this research.

When each stimulus characteristic was examined independently, all four exerted a unique influence on change detection. First, the level of interest in the changing item emerged as the most significant predictor of RT, with changes to CI items detected more rapidly than those to MI items. This finding aligns with previous results demonstrating that contextually central features attract greater attention and undergo faster processing at higher levels due to their higher global salience [8, 12, 24]. This interpretation is consistent with recent findings showing that the distribution of semantic meaning within scenes predicts reaction time in change-blindness paradigms. For example, Lu et al. (2025) showed that meaning maps, quantifying the spatial distribution of semantic information are stronger predictors of reaction time than low-level saliency measures [52]. These findings further support the conclusion that semantic relevance plays a central role in guiding attentional prioritization during change detection. Second, the number of distractors significantly affected change detection, with a greater number substantially prolonging RT. This pattern is supported by existing theories suggesting that increased distractor load elevates cognitive demands and interferes with efficient change detection [30, 53]. Third, the congruency of the change influenced detection efficiency. Congruent changes were detected more quickly than incongruent ones, indicating that congruency facilitates attentional allocation based on prior knowledge and contextual expectations. This finding is consistent with previous research, which suggests that congruent elements aligning with participants’ mental schemas are processed more readily [2, 32]. In the present study, congruency refers to the plausibility of the disappearance of an object type (e.g., a ball disappearing vs. a tree disappearing), rather than scene-object semantic consistency. Previous research has often shown that scene-incongruent objects, those that violate contextual expectations within a scene, may capture attention more readily due to schema violation or prediction error [32, 36]. However, the current manipulation concerns a different level of semantic expectancy, focusing on the plausibility of disappearance events rather than object–scene consistency. The relationship between semantic congruency and perceptual processing is not uniform. Some studies have reported that semantically congruent objects may be less likely to capture attention or may even be overlooked within a scene, possibly due to their expected nature [54]. Plausible disappearance events (i.e., congruent changes in this study) may be more efficiently integrated within existing semantic schemas, whereas implausible disappearance (incongruent changes) may generate greater processing demands, thereby slowing detection [2, 12]. Thus, although both forms of congruency relate to semantic expectations, they operate at distinct representational levels and may produce different attentional outcomes. Additionally, participants detected changes in social stimuli more quickly than in nonsocial stimuli. This result aligns with extensive literature, demonstrating that social information, such as faces and interactions, naturally captures attention due to their relevance in communication and interaction [29, 46].

The eye-tracking measures provide additional insight into how stimulus characteristics influence attentional processing during change detection. The effects observed for TFD suggest that semantic relevance influences the amount of sustained attention allocated to scene elements once they enter the attentional focus. Longer fixation durations are typically associated with deeper processing of meaningful or task-relevant information during scene perception [55]. In contrast, TFF was not significantly influenced by the stimulus characteristics examined, which may indicate that early gaze- orienting is guided primarily by low-level visual properties, whereas higher-level semantic factors exert stronger effects during later stages of visual processing [55, 56]. Lastly, the number of visits to the area of interest reflects repeated inspections of scene regions during the verification process in change detection, when observers return to potentially relevant locations to confirm whether a change has occurred [12, 56].

When the stimuli’s characteristics were examined simultaneously, their combined effects provided additional insight into the factors shaping visual attention during change detection. Although all four characteristics were associated with response time, their relative contributions differed. The level of interest of the changing item showed the strongest association with detection performance, followed by the number of distractors and congruency, whereas social context showed a more modest contribution. These findings indicate that in complex visual environments, the contextual relevance of the changing item (level of interest) and the scene’s visual complexity (number of distractors) play more prominent roles in change detection than congruency and social characteristics.

These findings can be contextualized within the framework of local and global processing theories, as delineated by Navon [57], who determined that individuals initially perceive the overall shape of a visual stimulus before recognizing its constituent elements. In this study, participants consistently detected changes to central (global) objects more quickly than marginal (local) objects, reflecting a preference for globally salient features. This global processing bias may account for the faster RTs associated with central-interest items, as such items are inherently more salient within real-world scenes [57, 58]. Additionally, the familiarity and meaning of the stimuli are important factors in determining whether processing proceeds globally or locally. Studies using familiar stimuli, such as letters and objects, show that global processing is more likely when the stimuli align with existing knowledge schemas [59]. Moreover, the plausibility of changes within visual scenes also affects memory and learning. For example, changes consistent with pre-existing schemas are more readily remembered [60]. A recent meta-analysis further supported this view, demonstrating that incorporating familiar, real-world contexts into learning enhances academic achievement, particularly when new information aligns with students’ prior knowledge [61].

An additional explanation for why the level of interest and the number of distractors play a more significant role in change detection compared to congruency and social characteristics in complex scenes lies beyond the global versus local approach. An alternative perspective suggests that congruency and social characteristics serve more situational and context-dependent functions; congruency, while important, acts more as a refinement mechanism. From a theoretical standpoint, recent studies of scene perception suggest that congruency modulates visual processing primarily through top-down predictive mechanisms. When an object aligns with the observer’s internal scene schema, it benefits from substantial prior probabilities that help optimize perceptual processing. Under these conditions, congruent stimuli are rapidly and efficiently integrated into an already-activated attentional representation, facilitating later-stage perceptual verification rather than initiating early attentional capture [2, 56]. A similar explanation may apply to social stimuli. Although social information such as faces and social interactions naturally attracts attention due to its communicative relevance [29, 46], its salience appears reduced in visually crowded environments, where attention must be distributed across multiple competing elements. In such contexts, global scene structure and visual complexity may overshadow the influence of social features, limiting their contribution to rapid change detection.

Limitations and recommendations for future research

Although the study provides valuable insights, several limitations should be acknowledged. First, the samples consisted solely of college students, which may limit the generalizability of the findings to other populations. Future studies should include a more diverse sample, such as young adults with autism spectrum disorder or attention deficit disorder, to gain a broader understanding of how they perceive and interpret their environment and the challenges they face in their daily functioning.

Second, the present study focused exclusively on visual characteristics; other factors, such as individual differences in cognitive abilities (e.g., executive function), were not considered. Future research should explore these factors to provide a more comprehensive understanding of visual attention processes. An additional limitation concerns the dichotomous operationalization of the examined stimulus characteristics. Although the level of interest, congruency, and number of distractors were defined as categorical variables to enable controlled factorial comparisons, these constructs may naturally exist along graded dimensions (e.g., varying degrees of contextual relevance or plausibility of disappearance, or number of distractors). Consequently, the mixed-effects model results reflect the predictive strength of the specific contrasts implemented in this study rather than the full dimensional structure of these characteristics. The magnitude of the observed effects may have been influenced by the dichotomization of stimulus characteristics. Specifically, the differences between the two defined levels (e.g., central vs. marginal, congruent vs. incongruent, and many vs. few distractors) may not have been equivalent across characteristics, potentially affecting their relative contribution in the model. Future studies could address this limitation by examining these characteristics along continuous or more finely graded levels (e.g., varying degrees of interest, plausibility, or visual complexity), to enable a more precise assessment of their role in visual attention.

Additionally, although stimuli were selected to represent clear and unambiguous examples, perceptions of contextual relevance and plausibility may vary across cultural or sociodemographic backgrounds.

Finally, investigating other characteristics that may influence saliency, such as crowding, could provide a more comprehensive perspective on how these features affect visual attention in a change-blindness paradigm. Examining how varying levels of visual load affect change detection may therefore provide a broader understanding of the mechanisms that shape visual attention in complex environments.

Conclusions

This study advances our understanding of visual attention by investigating the combined effects of four key visual characteristics on change detection when presented simultaneously, thereby more closely reflecting real-life experiences. The findings offer valuable insights into the differential impacts of these characteristics and show that central interest and the number of distractors were the strongest predictors of visual attention, as confirmed by eye-tracking measures (RT, TFF, TFD, and visits), consistent with previous literature, and contribute to the validation of the tool developed for this study. By identifying the visual characteristics most significantly associated with attention allocation, this research provides a basis for developing tailored strategies to support individuals in navigating their environments more effectively and enhancing daily and social functioning. In driving rehabilitation, for instance, understanding the relative contribution of these visual characteristics may inform the design of intervention protocols, such as managing multilane environments or preparing individuals for unexpected interruptions, thereby improving safety and performance. (e.g., multi-signage, sudden pedestrian crossings, abrupt changes in traffic flow). The practical implications extend to diverse populations, including individuals with attention deficit disorder (ADHD) or autism spectrum disorder (ASD), for whom differences in attentional prioritization and sensitivity to visual complexity may influence the way these characteristics are processed. A more precise understanding of how central interest, distractor load, congruency, and social content shape attention in typical young adults can inform the development of environments and intervention strategies that better accommodate the attentional profiles often observed in these populations.

Abbreviations

AOI

Area of Interest

CB

Change Blindness

CI

Central Interest

MI

Marginal Interest

PPVT-III

Peabody Picture Vocabulary Test – Third Edition

RPM

Raven Progressive Matrices

RT

Response Time

SRS-2

Social Responsiveness Scale – Second Edition

TFD

Total Fixation Duration

TFF

Time to First Fixation

WCPA

Weekly Calendar Planning Activity

Authors’ contributions

E.B. Data curation: Formal analysis: Investigation: Project administration: Visualization: Writing—original draft: Writing—review and editing. M.H. Conceptualization; Data curation; Methodology: Resources: Supervision: Visualization; Writing—review and editing. All authors read and approved the final manuscript.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or non-profit sectors.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Research was conducted in accordance with the Declaration of Helsinki. Ethical approval for this study was obtained from the Ariel University Ethics Committee (protocol no. AU-HEA-MH-20201130). All participants provided written informed consent prior to participation.

Consent for publication

Not applicable. No identifying personal information of participants is included in this manuscript.

Competing interests

The authors declare no competing interests.