Category differences in similarity between fixation density maps and psychological potential fields for sketches

arXiv:2611.06930 · q-bio.NC, cs.CV, cs.HC · Submitted 2026-09-25 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.

Ines: We're looking at the paper: "Category differences in similarity between fixation density maps and psychological potential fields for sketches".

Marcus: Here's the gist of the paper. When an object comes into view, humans can instantaneously recognize it if known to them This study reports how the similarity behaves across three categories—animal, non-animal, and meaningless shape—and under rotated presentation The baseline-adjusted AUC yielded significant differences among all three categories with large effect sizes, whereas the raw AUC did not distinguish between the animal and non-animal categories due to the influence of sketch representation This study characterizes the behavior of the FDM–PPF similarity under free viewing and indicates that baseline adjustment is necessary to separate category-related variation from stimulus-dependent baseline levels.

Ines: Pruning redundant reasoning tokens The research investigates the similarity between fixation density maps (FDM) and psychological potential fields (PPF) for sketches across animal, non-animal, and meaningless shape categories under free viewing conditions.

Marcus: The study examines how these similarity scores behave when stimuli are presented at rotated orientations, which are known to increase cognitive load during object recognition.

Ines: How the training signal is built The similarity was quantified using area under the curve (AUC) measures, including a baseline-adjusted AUC that isolates the early rise in similarity from stimulus-dependent baseline levels.

Title and authors: Ines: So we're looking at how the similarity between fixation density maps and psychological potential fields for sketches behaves across animal, non-animal, and meaningless shape categories when the stimuli are rotated.

Marcus: Right. The authors are examining this comparison under conditions that increase cognitive load during object recognition. They’re trying to isolate what drives that similarity change.

Yuki: It’s interesting because we often assume a universal pattern for how we categorize things, but this study tests if those patterns hold up when the physical presentation gets more complex, like rotation.

Ines: The main thing they report is that the baseline-adjusted AUC showed significant differences across all three categories with large effect sizes.

Marcus: That’s important because it means that even when we control for things that happen just because of the stimulus itself, there are still strong category effects in how similar these two maps are.

Yuki: So, what does this suggest about our underlying visual systems? Does it imply a more robust way to distinguish categories than just looking at the raw similarity score?

The paper's summary: Ines: To summarize, the paper is using area under the curve measures, specifically a baseline-adjusted AUC, to separate the early rise in similarity from any general stimulus-dependent baseline levels.

Marcus: That adjustment is key because they argue that without it, you can't really tell what’s category-related variation versus just how much the visual system is engaged by whatever is on the screen at that exact moment.

Yuki: It sounds like they are trying to find a way to clean up the data so we can see if the biological differences between animals and shapes are actually present in how our brain processes them.

Ines: Exactly. They also introduced a variability score, which is used to assess individual differences in how people react to these stimuli.

Marcus: And they found that when you look at low-level visual properties, like the average similarity across all orientations, those properties don't really predict the final similarity score as much as they might seem.

Yuki: That leads into what they call a robust pattern: meaningless shape stimuli consistently cluster in a region of low similarity and low variability.

The paper's improvements: Ines: The authors suggest that using this baseline-adjusted AUC is the way to effectively reduce the influence of those low-level sketch properties on the final similarity scores.

Marcus: So, if we want to build a better recognition system, we should focus on that adjustment because it seems to strip away a lot of noise related to how messy or simple the sketch itself is.

Yuki: That makes sense for understanding perception; if you can isolate the category effect from the physical sketch properties, you get a clearer picture of the actual cognitive mechanism at play.

Ines: They also found that rotating the presentation didn't create any significant interaction with categories, meaning rotation itself didn't specifically affect how much we categorize an animal versus a shape.

Marcus: That’s actually a relief for model building, since it means we don't have to worry about the orientation of the object confusing our category judgment.

Yuki: So, they suggest that for reliable recognition, focusing on that baseline adjustment and ignoring those orientation effects might be a good starting point when designing cognitive models.

Conclusion: Ines: So, wrapping up this look at "Category differences in similarity between fixation density maps and psychological potential fields for sketches," the main implication is that controlling for stimulus-dependent baselines lets us clearly see that animal, non-animal, and shape categories are statistically distinguishable.

Marcus: And because of the large effect sizes they found with the adjusted AUC, we can be confident that this method effectively filters out those low-level sketch properties to reveal the underlying category differences.

Yuki: I think for us in population genetics, this suggests that if we look at how organisms process visual input across different contexts, adjusting for presentation noise could give us a clearer signal about species-specific visual learning strategies.

Ines: Yeah, and they also point out that the consistency of meaningless shape stimuli clustering in low similarity and low variability regions is a very robust pattern found under free viewing conditions without predefined category labels.

Marcus: So the paper suggests that for building systems, we need to prioritize that baseline adjustment to separate category variation from stimulus-dependent levels.

Yuki: It’s a solid piece of work because it shows how visualization theory connects directly to actual human recognition tasks across different object types.

Naoyuki Awano, Yuki Hayashi

Faculty of Business, Kokushikan University · Graduate School of Informatics, Osaka Metropolitan University

q-bio.NC, cs.CV, cs.HC

Submitted: 2026-09-25

Updated: 2026-09-25

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 78/100

The gist: When an object comes into view, humans can instantaneously recognize it if known to them This study reports how the similarity behaves across three categories—animal, non-animal…

Key concepts

Fixation Density Maps (FDM)
FDM measures where a person's eyes tend to fix their gaze on a visual stimulus, creating a map of attention intensity. It is used here as one input signal in comparing how well the brain processes different types of shapes.
Psychological Potential Fields (PPF)
PPF represents the internal cognitive landscape or 'potential' associated with a stimulus. This field reflects how cognitively relevant or salient a sketch is to the viewer, serving as the second input signal for similarity comparison.
Baseline-Adjusted AUC
This metric calculates similarity by subtracting the initial, stimulus-dependent baseline score from subsequent scores. This adjustment is vital because it isolates true category differences in processing ability from general variations in how easily any specific sketch is initially perceived.
Category × Orientation Interaction
This statistical term checks if the effect of presentation rotation (e.g., 0°, 90°, 180°) on similarity differs depending on the object category (animal vs. shape). Finding no significant interaction suggests that rotating a sketch does not specifically affect how well we recognize animals versus non-animals.

Terminology

Summary

When an object comes into view, humans can instantaneously recognize it if known to them This study reports how the similarity behaves across three categories—animal, non-animal, and meaningless shape—and under rotated presentation The baseline-adjusted AUC yielded significant differences among all three categories with large effect sizes, whereas the raw AUC did not distinguish between the animal and non-animal categories due to the influence of sketch representation This study characterizes the behavior of the FDM–PPF similarity under free viewing and indicates that baseline adjustment is necessary to separate category-related variation from stimulus-dependent baseline levels. p. 0

Where the paper says it

Results indicated that the baseline-adjusted AUC yielded significant differences among all three categories with large effect sizes, whereas the raw AUC did not distinguish between the animal and non-animal categories due to the influence of sketch representation. No significant Category × Orientation interaction was observed, indicating that rotated presentation did not produce a category-specific effect.

Pruning redundant reasoning tokens

The research investigates the similarity between fixation density maps (FDM) and psychological potential fields (PPF) for sketches across animal, non-animal, and meaningless shape categories under free viewing conditions. The study examines how these similarity scores behave when stimuli are presented at rotated orientations, which are known to increase cognitive load during object recognition. p. 1

Where the paper says it

This study reports the behavior of the FDM–PPF similarity on a larger and more structured set of stimuli. Using simple sketches as stimulus images— chosen both to simplify the problem and to satisfy the binary-image constraint of the PPF—we examined: (1) how the similarity scores differ among the primary categories of animal, non-animal, and meaningless shape; (2) how the scores change when stimuli are presented at rotated orientations, which are known to increase cognitive load during object recognition [13–16]; and (3) how the scores vary across individuals.

How the training signal is built

The similarity was quantified using area under the curve (AUC) measures, including a baseline-adjusted AUC that isolates the early rise in similarity from stimulus-dependent baseline levels. A variability score was introduced to assess individual differences. p. 0

Where the paper says it

The similarity was quantified through area under the curve (AUC) measures, including a baseline-adjusted AUC that isolates the early rise in similarity from stimulus-dependent baseline levels. A variability score was introduced to assess individual differences, and linear mixed models were used to evaluate category and orientation effects.

The similarity score S is redefined as S(If dm, Ippf) = σf dm−ppf + σf dmσppf / 2σf dmσppf, where If dm represents an FDM image, Ippf a PPF image. p. 3
Where the paper says it

Since this method utilizes the correlation coefficient as an indicator of image similarity rather than under a general interpretation, the value range may hinder an intuitive understanding of the similarity. Therefore, we redefine the similarity score S for enhanced clarity and normalize it as follows: S(If dm, Ippf ) = σf dm−ppf + σf dmσppf 2σf dmσppf , (3) 4

The baseline-adjusted AUC is defined as Adj. p. 4
Where the paper says it

1 i s, (4) summed over the similarity scores sampled every 100 ms. In addition, we define the baseline-adjusted AUC (Adj.

AUC = X 10 i=0 S(ti) − S(t0), where S(t0) denotes the similarity score of the first sample (0 − 100 ms). p. 4
Where the paper says it

AUC) as Adj. AUC = X 10 i=0 S(ti) − S(t0) , (5) where S(t0) denotes the similarity score of the first sample (0 − 100 ms).

Where the gains come from

The linear mixed models (LMMs) were used to evaluate category and orientation effects, including a Category × Orientation interaction term. Results indicated that the baseline-adjusted AUC yielded significant differences among all three categories with large effect sizes, whereas the raw AUC did not distinguish between the animal and non-animal categories due to the influence of sketch representation. p. 0

Where the paper says it

A variability score was introduced to assess individual differences, and linear mixed models were used to evaluate category and orientation effects. Results indicated that the baseline-adjusted AUC yielded significant differences among all three categories with large effect sizes, whereas the raw AUC did not distinguish between the animal and non-animal categories due to the influence of sketch representation.

No significant Category × Orientation interaction was observed, indicating that rotated presentation did not produce a category-specific effect. p. 0
Where the paper says it

Results indicated that the baseline-adjusted AUC yielded significant differences among all three categories with large effect sizes, whereas the raw AUC did not distinguish between the animal and non-animal categories due to the influence of sketch representation. No significant Category × Orientation interaction was observed, indicating that rotated presentation did not produce a category-specific effect.

The statistical analysis involved fitting LMMs using restricted maximum likelihood estimation, including Category (animal, non-animal, meaningless shape) and Orientation (0◦, 90◦, 180◦) as fixed effects. Participant and Stimulus were included as crossed random effects to account for both individual differences and stimulus-specific variability. p. 4

Where the paper says it

Category (animal, non-animal, meaningless shape) and Orientation (0 ◦ , 90◦ , 180◦ ) were included as fixed effects, with their interaction term. Participant and Stimulus were included as crossed random effects, with Participant as the grouping factor and Stimulus as a variance component, thereby accounting for both individual differences and stimulus-specific variability.

The analysis of low-level visual properties showed that Spearman rank correlations between each feature and the stimulus-level AUC and Adj. AUC averaged across orientations were computed. p. 5

Where the paper says it

As these features are invariant to rotated presentation, Spearman rank correlations were computed between each feature and the stimulus-level AUC and Adj. AUC averaged across orientations.

These results support the claim that baseline adjustment reduces the influence of low-level stimulus properties. p. 0
Where the paper says it

The meaningless shape stimuli yielded consistently low similarity and variability scores, clearly separating them from the two meaningful object categories. These results characterize the behavior of the FDM–PPF similarity under free viewing and indicate that baseline adjustment is necessary to separate category-related variation from stimulus-dependent baseline levels.

The consistent clustering of meaningless shape stimuli in the low similarity and low variability region is the most robust pattern in the data. p. 8
Where the paper says it

Blue markers represent animal stimuli, green for non-animal stimuli, and red for meaningless shape stimuli. In both panels, the meaningless shape stimuli cluster in the lower-left region, indicating consistently low similarity and variability.

The Adj. AUC yielded significant differences for all three pairwise comparisons, with large effect sizes (all d > 1. p. 5

Where the paper says it

AUC, all three pairwise comparisons were significant (all p < . 001, Bonferroni-corrected), with large effect sizes (all |d| > 1.

0), indicating that baseline adjustment effectively reduces the influence of low-level sketch properties. p. 0
Where the paper says it

Results indicated that the baseline-adjusted AUC yielded significant differences among all three categories with large effect sizes, whereas the raw AUC did not distinguish between the animal and non-animal categories due to the influence of sketch representation. No significant Category × Orientation interaction was observed, indicating that rotated presentation did not produce a category-specific effect.

The similarity and variability scores did not capture the cognitive load associated with rotated presentation. p. 0
Where the paper says it

A few studies using simple sketches have reported varying degrees of image similarity between the fixation density map (FDM), which visualizes eye movements and durations, and the psychological potential field (PPF), which visualizes part of the visual cognition process, depending on the object category. This study reports how the similarity behaves across three categories—animal, non-animal, and meaningless shape—and under rotated presentation, which is known to increase cognitive load during object recognition.

p. 0
Where the paper says it

A few studies using simple sketches have reported varying degrees of image similarity between the fixation density map (FDM), which visualizes eye movements and durations, and the psychological potential field (PPF), which visualizes part of the visual cognition process, depending on the object category. This study reports how the similarity behaves across three categories—animal, non-animal, and meaningless shape—and under rotated presentation, which is known to increase cognitive load during object recognition.

p. 0
Where the paper says it

A few studies using simple sketches have reported varying degrees of image similarity between the fixation density map (FDM), which visualizes eye movements and durations, and the psychological potential field (PPF), which visualizes part of the visual cognition process, depending on the object category. This study reports how the similarity behaves across three categories—animal, non-animal, and meaningless shape—and under rotated presentation, which is known to increase cognitive load during object recognition.

The consistent clustering of meaningless shape stimuli in the low similarity and low variability region is the most robust pattern in the data, and it was obtained under free-viewing conditions without predefined category labels. p. 0
Where the paper says it

The meaningless shape stimuli yielded consistently low similarity and variability scores, clearly separating them from the two meaningful object categories. These results characterize the behavior of the FDM–PPF similarity under free viewing and indicate that baseline adjustment is necessary to separate category-related variation from stimulus-dependent baseline levels.

p. 0
Where the paper says it

The meaningless shape stimuli yielded consistently low similarity and variability scores, clearly separating them from the two meaningful object categories. These results characterize the behavior of the FDM–PPF similarity under free viewing and indicate that baseline adjustment is necessary to separate category-related variation from stimulus-dependent baseline levels.

p. 0
Where the paper says it

The meaningless shape stimuli yielded consistently low similarity and variability scores, clearly separating them from the two meaningful object categories. These results characterize the behavior of the FDM–PPF similarity under free viewing and indicate that baseline adjustment is necessary to separate category-related variation from stimulus-dependent baseline levels.

p. 0
Where the paper says it

The meaningless shape stimuli yielded consistently low similarity and variability scores, clearly separating them from the two meaningful object categories. These results characterize the behavior of the FDM–PPF similarity under free viewing and indicate that baseline adjustment is necessary to separate category-related variation from stimulus-dependent baseline levels.

The overall trend observed in the similarity scores aligns with previous research, particularly the rapid increase within 1 second for animal stimuli, which mirrors the speed of animal cognition. The LMM analysis confirmed that the three categories are statistically distinguishable, particularly when the baseline-adjusted measure is used to control for the influence of sketch representation. p. 0

Where the paper says it

A variability score was introduced to assess individual differences, and linear mixed models were used to evaluate category and orientation effects. Results indicated that the baseline-adjusted AUC yielded significant differences among all three categories with large effect sizes, whereas the raw AUC did not distinguish between the animal and non-animal categories due to the influence of sketch representation.

p. 0
Where the paper says it

A variability score was introduced to assess individual differences, and linear mixed models were used to evaluate category and orientation effects. Results indicated that the baseline-adjusted AUC yielded significant differences among all three categories with large effect sizes, whereas the raw AUC did not distinguish between the animal and non-animal categories due to the influence of sketch representation.

The similarity and variability scores did not capture the cognitive load associated with rotated presentation. The consistent clustering of meaningless shape stimuli in the low similarity and low variability region is the most robust pattern in the data, and it was obtained under free-viewing conditions without predefined category labels.

The study reported that baseline adjustment effectively reduces the influence of low-level sketch properties. The consistent clustering of meaningless shape stimuli in the low similarity and low variability region is the most robust pattern in the data, and it was obtained under free-viewing conditions without predefined category labels. The LMM analysis confirmed that the three categories are statistically distinguishable, particularly when the baseline-adjusted measure is used to control for the influence of sketch representation. The similarity and variability scores did not capture the cognitive load associated with rotated presentation. The consistent clustering of meaningless shape stimuli in the low similarity and low variability region is the most robust pattern in the data, and it was obtained under free-viewing conditions without predefined category labels.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved object recognition systems can utilize a baseline-adjusted similarity score to effectively separate category-related variation from stimulus-dependent baseline levels, as indicated by baseline adjustment is necessary to separate category-related variation from stimulus-dependent baseline levels.

  2. Enhanced systems can better discriminate between meaningful object categories (animal, non-animal) and meaningless shapes by using the Adj. AUC because "the Adj. AUC yielded significant differences for all three pairwise comparisons with large effect sizes, indicating that baseline adjustment effectively reduces the influence of low-level sketch properties on the similarity scores."

  3. The system can be designed to identify cognitive load effects from presentation changes by observing that no significant Category × Orientation interactions were observed, indicating that rotated presentation did not produce a category-specific effect.

Abstract

When an object comes into view, humans can instantaneously recognize it if known to them. However, assessing how this recognition takes place remains challenging. A few studies using simple sketches have reported varying degrees of image similarity between the fixation density map (FDM), which visualizes eye movements and durations, and the psychological potential field (PPF), which visualizes part of the visual cognition process, depending on the object category. This study reports how the similarity behaves across three categories---animal, non-animal, and meaningless shape---and under rotated presentation, which is known to increase cognitive load during object recognition. The similarity was quantified through area under the curve (AUC) measures, including a baseline-adjusted AUC that isolates the early rise in similarity from stimulus-dependent baseline levels. A variability score was introduced to assess individual differences, and linear mixed models were used to evaluate category and orientation effects. Results indicated that the baseline-adjusted AUC yielded significant differences among all three categories with large effect sizes, whereas the raw AUC did not distinguish between the animal and non-animal categories due to the influence of sketch representation. No significant Category times Orientation interaction was observed, indicating that rotated presentation did not produce a category-specific effect. The meaningless shape stimuli yielded consistently low similarity and variability scores, clearly separating them from the two meaningful object categories. These results characterize the behavior of the FDM--PPF similarity under free viewing and indicate that baseline adjustment is necessary to separate category-related variation from stimulus-dependent baseline levels.

Related papers