Factors underlying flicker-induced time dilation: Temporal frequency, semantic content, and subjective saliency of visual stimuli Authors: Takeya Oda, Amirmahmoud Houshmand Chatroudi, Yuko Yotsumoto Published: 2026 ABSTRACT Flicker-induced time dilation refers to the robust illusion whereby temporally modulated visual stimuli are perceived as lasting longer than static stimuli. Although flicker-induced time dilation has been consistently demonstrated, it remains unclear whether its magnitude is determined solely by low-level temporal dynamics or whether higher-level semantic processing also contributes. The present study investigated the respective roles of flicker frequency, semantic content, and subjective saliency in shaping perceived duration. Using the semantic wavelet-induced frequency tagging image transformation method, we generated semantic (faces and houses) and scrambled flickers presented at 2, 4, and 6 Hz, with and without luminance modulation. In experiment 1, participants performed a duration comparison task to estimate perceived duration. In experiment 2, subjective saliency of the same stimuli was rated. Results showed that increasing flicker frequency reliably enhanced perceived duration. Crucially, semantic content did not modulate perceived duration at any frequency, regardless of luminance modulation. In contrast, subjective saliency increased both with flicker frequency and with semantic information. This dissociation indicates that, although semantic content enhances subjective saliency, it does not contribute to temporal dilation. These findings suggest that distortions in perceived time induced by visual stimuli are primarily shaped by low-level temporal dynamics and increased activity in early visual areas, whereas higher-order visual processing contribute little to the modulation of perceived duration. Introduction Time does not constitute a primary sensory modality, but is widely considered an emergent property of neural processing. In contrast with vision and audition, which are supported by specialized sensory receptors and primary cortical areas, there is no evidence for a dedicated sensory organ or primary cortical region exclusively responsible for time. Rather than being generated by a centralized neural clock, temporal perception is thought to arise from distributed computations within and across sensory and higher-order cortical areas (e.g., Johnston, Arnold, & Nishida, 2006; Mioni et al., 2016; Ortega, Guzman-Martinez, Grabowecky, & Suzuki, 2012; Salvioni Murray, Kalmbach, & Bueti, 2013; for reviews, see Buhusi & Meck, 2005; Grondin, 2010). Although all perceptual experiences, including vision and audition, are modulated by contextual and attentional factors, time differs in that it is not directly tied to a specific sensory receptor or stimulus dimension. Instead, perceived duration emerges from the integration and interaction of multiple sensory and cognitive processing. In this context, previous studies have shown that perceived duration can vary substantially depending on contextual and attentional influences (Ciria, López, & Lara, 2019; Mattes & Ulrich, 1998). Classic psychophysical findings have demonstrated that intervals containing a greater number of discrete or regularly patterned events are typically experienced as longer (Buffardi, 1971), highlighting the role of sensory and cognitive processes in shaping how humans experience the passage of time. Within the visual modality, many experiments have demonstrated that dynamic stimuli, particularly flickers, often appear to last longer than physically equivalent static stimuli (Droit-Volet & Wearden, 2002; Herbst, Javadi, van der Meer, & Busch, 2013; Johnston et al., 2006; Kanai, Paffen, Hogendoorn, & Verstraten, 2006; Okajima & Yotsumoto, 2016; Ortega et al., 2012; Ortega & López, 2008; Schiffman & Bobko, 1974; Treisman & Brogan, 1992). This perceptual expansion in time has been documented across a wide temporal range, from sub-second intervals (Li et al., 2024; Ortega & López, 2008; Plomp, van Leeuwen, & Gepshtein, 2012) to supra-second durations (Binetti, Lecce, & Doricchi, 2012). Collectively, these findings indicate that the visual system plays an important role in constructing subjective time and that the temporal dynamics of sensory input can strongly bias duration judgments. The phenomenon whereby flickering stimuli are perceived as lasting longer than static ones is commonly termed flicker-induced time dilation (FITD). This illusion provides a robust and reproducible means to investigate the mechanisms underlying time perception. Despite the consistent behavioral evidence for FITD, its mechanisms remain a matter of ongoing debate. Two principal theoretical accounts have been proposed to explain this effect. A prominent interpretation of FITD is the subjective saliency account, which suggests that temporal expansion arises from the perceived prominence or distinctiveness of temporal changes within a stimulus sequence. Herbst et al. (2013) reported that FITD failed to emerge at frequencies that exceeded the critical fusion flicker threshold but remained below the steady-state visual evoked potential threshold, implying that perceptual awareness of flicker is essential for the illusion to occur. Another line of explanation attributes FITD to neural entrainment, as Hashimoto and Yotsumoto (2018) demonstrated that the magnitude of FITD covaried with the strength of neural entrainment measured by electroencephalography (EEG), consistent with oscillator-based models of temporal processing. These accounts are not mutually exclusive; rather, the saliency of temporal changes appears to play a primary role in determining both the occurrence and magnitude of time dilation, while neural entrainment exerts a secondary, modulatory influence (Li, Ito, & Yotsumoto, 2020). Understanding the FITD, therefore, provides a unique opportunity to bridge neural, perceptual, and cognitive perspectives on time perception. Investigating how temporal dynamics and subjective saliency interact may help to elucidate general principles of how the brain constructs temporal experience from sensory input. Although the FITD has been consistently replicated, the specific factors that determine its magnitude remain unclear. One central question concerns whether temporal distortion arises solely from low-level visual dynamics, such as luminance modulation and temporal frequency, or whether higher-level factors, including semantic processing, also contribute to the perceived expansion of time. Previous studies have mainly focused on luminance-based flicker stimuli, which mainly engage early visual areas (Bruno, Ng, & Johnston, 2013; Burr, Tozzi, & Morrone, 2007; Kanai et al., 2006; Morrone, Cicchini, & Burr, 2010). However, these studies have not fully considered the potential influence of semantic content and higher-order visual processing. Neuroimaging studies have demonstrated that the generation of steady-state visual evoked potentials elicited by flickering stimuli engages, not only early visual areas, but also higher-order regions along the ventral visual pathway (Di Russo et al., 2007). Moreover, recent evidence suggests that activity within this pathway can also influence subjective time perception. Ma, Cameron, and Wiener (2024) reported that more memorable images—those associated with stronger activation in ventral visual regions—tend to be perceived as lasting longer than less memorable ones. In a similar vein, studies have shown that stimuli conveying implied action or motion, even in the absence of physical movement, are perceived as lasting longer (e.g., Mioni et al., 2015; Yamamoto & Miura, 2012). Collectively, this evidence provides a strong theoretical basis for the hypothesis that higher-level representational strength and conceptual meaning in the ventral visual pathway can bias subjective duration. The established capacity for semantic processing to modulate duration in these contexts raises an intriguing possibility: if semantic processing in higher visual areas can modulate perceived duration, then similar high-level mechanisms may also contribute to the FITD. Accordingly, if the FITD reflects the overall strength of neural activation across the visual hierarchy, stimuli with semantic content might be expected to produce stronger temporal dilation than scrambled flickers that primarily activate early visual areas. In contrast, if the FITD is driven exclusively by low-level temporal dynamics, both meaningful and meaningless flickers should induce comparable levels of perceived time expansion. Houshmand Chatroudi and Yotsumoto (2025) investigated the role of semantic information in the FITD by comparing semantic and scrambled flickers generated using the semantic wavelet-induced frequency tagging (SWIFT) method (Koenig-Robert & VanRullen, 2013). The primary motivation of that study was methodological and neurophysiological; they introduced semantic flickers to demonstrate that such stimuli generated using the SWIFT technique reliably engage higher-order visual cortex. Their EEG results showed that semantic flickers elicited right-lateralized phase clustering in occipitotemporal regions, suggesting the involvement of higher-order visual areas. However, despite this neural difference, no significant difference in FITD magnitude was observed between semantic and scrambled conditions when using smoothly modulated 2-Hz flickers. Consequently, it is yet to be determined whether the absence of a semantic effect reflected a genuine lack of ventral contribution to FITD or was attributable to the specific properties of low-frequency, smoothly modulated stimulation. Another key problem was that their stimuli involved gradual transitions, which rendered the perceptual saliency of the flicker relatively weak. To address this problem by enhancing the subjective saliency of the temporal modulation, we used more abrupt flickers, such as those generated by square-wave modulation or presented at higher frequencies. This design is particularly pertinent given that perceptual saliency itself is widely recognized as a critical factor for the FITD. Herbst et al. (2013) reported that time dilation reflects the perceived prominence or vividness of a stimulus. Although their study primarily characterized perceptual saliency in temporal terms, focusing on the perceived intensity and regularity of flicker-induced changes, saliency can also originate from semantic information, because semantically meaningful regions are known to attract attention independently of low-level visual features (Henderson & Hayes, 2017). It, therefore, remains unclear whether these two dimensions of saliency operate through a shared mechanism or reflect distinct processes. To address these questions, the present study undertook a systematic psychophysical investigation by manipulating both the temporal and semantic properties of flickering stimuli. Building on the prior neurophysiological validation of SWIFT as a method for engaging the ventral cortex, we aimed to provide a frequency-general test of whether ventral stream processing contributes to FITD. Using the SWIFT image transformation method (Houshmand Chatroudi & Yotsumoto, 2025; Koenig-Robert & VanRullen, 2013), we created semantic (faces and houses) and scrambled flickers, each presented at multiple frequencies (2, 4, and 6 Hz). Crucially, we shifted to square-wave modulation to produce sharper temporal transients and stronger perceptual flicker, thereby maximizing the likelihood of detecting behavioral consequences of semantic processing that may have been undetectable under the subtle conditions of prior research. By measuring both perceived duration and subjective saliency within a unified framework, we sought to clarify whether content-driven saliency and temporal dynamics exert dissociable influences on temporal perception. In Experiment 1, we examined whether the perceived duration of the temporally modulated stimuli varied as a function of the modulation frequency and semantic information. In Experiment 2, we measured the subjective saliency of the same stimuli to examine the relationship between the subjective saliency and the duration perception. By jointly assessing temporal distortion and perceived saliency, we aimed to test whether the magnitude of the FITD is primarily determined by low-level temporal dynamics or whether higher-level semantic factors and subjective impressions modulate the illusion. Based on previous findings, we hypothesized that higher flicker frequencies would lead to a stronger FITD and greater subjective saliency. However, if the FITD depends on neural activation strength across the visual hierarchy, semantic flickers, particularly those engaging category-selective regions, may also yield enhanced temporal dilation compared with scrambled stimuli. Experiment 1 Methods Participants Eleven participants took part in Experiment 1a (mean age, 21.45 ± 1.63 years) and 14 in Experiment 1b (mean age, 18.93 ± 1.00 years). There were no subjects who participated in multiple experiments. Experiments 1a and 1b were conducted in separate groups for practical reasons. Each experiment involved a substantial number of trials administered across 2 days. Combining both experiments within the same participants would have resulted in an excessively long and demanding procedure, increasing fatigue and potentially reducing the reliability of temporal judgments in this fine-grained discrimination task. To maintain data quality, the two experiments were therefore conducted in independent samples. Participants whose just noticeable difference (JND) exceeded 1.00 were excluded from further analysis. As a result, eight participants remained in analysis in each experiment. The sample size was determined based on a priori power analysis conducted in G*Power (version 3.1.9.6; Faul, Erdfelder, Buchner, & Lang, 2009) for a matched-pairs t-test. Effect sizes were taken from Houshmand Chatroudi and Yotsumoto (2025), who reported Cohen's d was 1.299 (static vs. scrambled flicker) and d was 1.250 (static vs. semantic flicker). Assuming an α of 0.05 and desired power of 0.80, the required sample size was estimated accordingly. All procedures were approved by the Ethics Committee of the University of Tokyo (protocol #987) and conducted in accordance with its ethical standards. Participants provided written informed consent and received monetary compensation for their participation. Apparatus To create stimuli and conduct experiments, we used MATLAB R2024b with Psychtoolbox-3 extension (Brainard, 1997; Kleiner et al., 2007; Pelli, 1997). The stimuli were presented on a 23.6-inch VIEWPixx 3D CRT monitor (VPixx Technologies, Saint-Bruno, QC, Canada) with a refresh rate of 120 Hz and a resolution of 1,920 × 1,200 pixels. Participants were seated in a dimly lit, sound-attenuated room at a viewing distance of 70 cm, with their head stabilized on a chin rest. Stimuli The same set of images were used as in Houshmand Chatroudi and Yotsumoto (2025) to generate stimuli, including eight faces and eight houses, all of which were provided as free-to-use materials. The images were converted to grayscale, resized to 300 × 300 pixels (6.7°), and contrast-equalized using the histogram equalization function in MATLAB. Subsequently, the average luminance of the images was equalized. SWIFT cycles were then generated using the SWIFT function (Koenig-Robert & VanRullen, 2013). SWIFT is an image transformation technique in which an image is decomposed using wavelets and its phase-defined edge structures (contours) are rotated along an isoenergetic sphere. This procedure generates scrambled versions of the original image while preserving local energy and spatial frequency distributions. As a result, low-level visual properties are maintained, whereas higher-level semantic structure is disrupted (Koenig-Robert, VanRullen, & Tsuchiya, 2015; Koenig-Robert & VanRullen, 2013). Importantly, previous EEG studies have shown that SWIFT selectively tags high-level object representations: significant neural phase-locking to the SWIFT tagging frequency was observed between approximately 1.5 and 4.3 Hz (Koenig-Robert & VanRullen, 2013), indicating that the method effectively modulates higher-level object processing within this temporal range. Using this method, we created eight SWIFT cycles (480 frames) for each image and calculated the relative image similarity index for each frame. This index is obtained by averaging the absolute differences between the wavelet coefficients of each frame and those of the original image (Houshmand Chatroudi & Yotsumoto, 2025; Koenig-Robert et al., 2015). A normalization process was applied to each SWIFT sequence. Initially, frame-wise contrast was standardized using the contrast-limited adaptive histogram equalization function implemented in MATLAB. The pixel values of all frames were then transformed into the frequency domain through a Fourier transform. After this, the merged pixel frequency spectra underwent amplitude normalization and were low pass filtered by cutting off frequencies beyond the midpoint between harmonics succeeding the last harmonic of f0 (refer to Houshmand Chatroudi & Yotsumoto [2025] for a detailed explanation). Finally, the overall luminance of each frame was adjusted to mid-gray (0.5 on a normalized 0–1 grayscale scale), which corresponded to 21.44 cd/m² on the display used in the experiment (with 0 = 0.09 cd/m² and 1 = 90.27 cd/m²). By applying this normalization, variability in low-level visual features such as contrast, pixel modulation, and global luminance was eliminated, ensuring that only the spatial frequency differences caused by isoenergetic wavelet rotation remained between frames (Houshmand Chatroudi & Yotsumoto, 2025). The original SWIFT sequences used in Koenig-Robert and VanRullen (2013) and Koenig-Robert et al. (2015), as well as the semantic flickers used by Houshmand Chatroudi and Yotsumoto (2025), were smoothly modulating throughout the sequence. In the present study, we opted for square-wave semantic flickers instead to produce a more distinct and noticeable flickering percept, as well as to make the temporal modulation comparable with the previous studies that examined the effects of luminance flickers on FITD (Figure 1). For this purpose, frames corresponding with the highest and lowest relative image similarity indices, representing the most (distinct) and least (vague) recognizable images, were obtained from the previously generated SWIFT sequences, and these two images were presented in an alternating manner for 300 frames, which corresponds with 2.5 seconds. Importantly, the presentation duration of the distinct image was kept constant across flicker frequencies to ensure comparable perceptual visibility and recognizability of the image content across conditions. Flicker frequency was manipulated by varying the overall cycle duration. As a consequence, the duration of the vague image necessarily differed across frequency conditions to achieve the intended temporal rates (2, 4, and 6 Hz). This design allowed us to examine frequency-dependent effects while holding the exposure time of the distinct image constant. The distinct image served as the stimulus for the semantic static condition. The scrambled flicker and scrambled static stimuli were derived from the vague frame of the semantic flicker sequence, which was reprocessed using the SWIFT method. From the newly generated sequences, two different vague images were selected and alternated to form the scrambled flicker condition, while the image with the higher relative image similarity index served as the scrambled static stimulus. Finally, the semantic and scrambled luminance flickers were created by adding a constant luminance increment of 0.25 on a normalized 0 to 1 grayscale scale to the respective semantic and scrambled static images, corresponding to an increase in global luminance from 0.50 (21.44 cd/m²) to 0.75 (52.12 cd/m²) as measured on the experimental display. The modulation of flickering stimuli. (a) In the 2-Hz flicker condition (cycle duration = 500 ms), the distinct image was presented for 100 ms and the vague image for 400 ms within each cycle. (b) In the 4-Hz condition (cycle duration = 250 ms), the distinct image was presented for 100 ms and the vague image for 150 ms within each cycle. (c) In the 6-Hz condition (cycle duration = 167 ms), the distinct image was presented for 100 ms and the vague image for 67 ms within each cycle. Procedure A duration comparison task was conducted in Experiments 1a and 1b (Figure 2a). Each trial started with a fixation cross presented for a randomly determined duration between 1.0 and 1.5 seconds, followed by a blank screen lasting 0.3 to 0.6 seconds. Subsequently, a standard stimulus and a comparison stimulus were shown sequentially, separated by a randomly assigned inter-stimulus interval of 1.0 to 1.5 seconds. After an additional blank period of 0.8 to 1.2 seconds, participants indicated by pressing a key which of the two stimuli appeared to last longer in a two-interval forced-choice paradigm. They were asked not to count the duration of stimuli. The comparison stimulus durations were 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, and 4.0 seconds, and the order of presentation between standard and comparison stimuli was randomized across trials. The standard stimulus was fixed at 2.5 seconds in duration. In Experiment 1a, eight types of standard stimuli were tested: scrambled static stimuli, scrambled flickers at 2, 4, and 6 Hz, semantic static stimuli, and semantic flickers at 2, 4, and 6 Hz (Figure 2b). In contrast, Experiment 1b included six stimulus types, consisting of scrambled luminance flickers and semantic luminance flickers presented at 2, 4, and 6 Hz (Figure 2b). All comparison stimuli were static images, identical to those in the static standard condition. For Experiment 1a, there were 16 images × 7 durations × 8 conditions, resulting in a total of 896 trials. All trials were randomized and then divided into 7 blocks per day across 2 days (64 trials per block). For Experiment 1b, there were 16 images × 7 durations × 6 conditions, resulting in a total of 672 trials, which were also fully randomized and divided into 6 blocks per day across 2 days (56 trials per block). Experimental design and stimulus conditions in experiments 1a and 1b. (a) Duration comparison task used in Experiments 1a and 1b. In each trial, a standard stimulus and a comparison stimulus were presented sequentially, separated by fixation and blank intervals. Participants indicated which stimulus appeared to last longer in a two-interval forced-choice task and were instructed not to count stimulus duration. The presentation order of standard and comparison stimuli was randomized across trials. (b) Stimulus conditions used in Experiments 1a and 1b. In Experiment 1a, eight types of standard stimuli were tested, including scrambled static stimuli, scrambled flickers (2, 4, and 6 Hz), semantic static stimuli, and semantic flickers (2, 4, and 6 Hz). In Experiment 1b, six stimulus types were used, consisting of scrambled luminance flickers (2, 4, and 6 Hz) and semantic luminance flickers (2, 4, and 6 Hz). Analysis Psychometric functions were fitted for each participant and condition using a logistic function implemented in the Palamedes toolbox (Prins & Kingdom, 2018), where the vertical axis was defined as the proportion of trials in which the comparison stimulus was judged longer than the standard and the horizontal axis as the duration of the comparison stimulus. The lapse rate parameter was fixed at 0.02 to compensate for potential attentional lapses associated with the extended duration of the task (Houshmand Chatroudi & Yotsumoto, 2025; Kingdom & Prins, 2016), and the guess rate was fixed to zero. From the fitted functions, the point of subjective equality (PSE) and the JND were derived. The PSE was defined as the comparison duration that yielded 50% comparison longer responses on the psychometric function, and JND as one-half the difference between the durations corresponding to the 25% and 75% comparison longer points. In the first stage of the analysis, a Bayesian repeated-measures analysis of variance was conducted in JASP (version 0.19.3; JASP Team, 2025), and the inclusion Bayes factors (BFincl) were computed to evaluate the main effects of flicker frequency and semantic content. Subsequently, Wilcoxon signed-rank tests were performed on PSE and JND at each flicker frequency to compare the semantic and scrambled conditions. Experimental data from each participant are available on the Open Science Framework (https://osf.io/egvmf). Results and discussion Flicker frequency (2, 4, and 6 Hz) had a robust effect on PSE (Figure 3). This effect was supported by a BFincl of 14.04 for flickers without luminance modulation (Experiment 1a) and more than 1,000 for flickers with luminance modulation (Experiment 1b), indicating strong to extreme evidence. By contrast, the JND was not influenced by flicker frequency (Figure 4), with BFincl values providing anecdotal evidence for the null effect (0.64 for flickers without luminance modulation [Experiment 1a]; 0.53 for flickers with luminance modulation [Experiment 1b]). In contrast, semantic information did not modulate either the PSE (Figure 3) or the JND (Figure 4), with the BFincl providing anecdotal evidence in favor of the null effect (PSE: BFincl = 0.62 for flickers without luminance modulation [Figure 3a] and 0.42 for flickers with luminance modulation [Figure 3b]; JND: BFincl = 0.40 for flickers without luminance modulation [Figure 4a] and 0.46 for flickers with luminance modulation [Figure 4b]). Consistent with this overall pattern, follow-up Wilcoxon signed-rank tests conducted at each flicker frequency revealed no significant differences between the scrambled and semantic conditions for either the PSE or JND. For conditions without luminance change, Wilcoxon signed-rank tests revealed no significant differences in the PSE between conditions at static (V = 7; p = 0.529), 2 Hz (V = 6; p = 0.107), 4 Hz (V = 19; p = 0.944), or 6 Hz (V = 18.5; p = 1.00). The JND likewise showed no significant differences at static (V = 13; p = 0.529), 2 Hz (V = 15; p = 0.726), 4 Hz (V = 12; p = 0.441), or 6 Hz (V = 22; p = 0.624). For conditions with luminance change, the PSE again showed no significant effects at 2 Hz (V = 30; p = 0.106), 4 Hz (V = 18; p = 1.00), or 6 Hz (V = 12; p = 0.441). The JND was also non-significant at 2 Hz (V = 25; p = 0.362), 4 Hz (V = 23; p = 0.529), and 6 Hz (V = 18.5; p = 1.00). Collectively, these findings indicate that, although higher flicker frequencies reliably enhanced FITD, as reflected by the increasing PSE, temporal sensitivity remained unchanged across frequencies. Furthermore, the lack of differences between scrambled and semantic conditions shows that semantic content did not modulate the FITD effect. The PSE for scrambled (blue) and semantic stimuli (red). The duration of the standard stimulus was 2.5 seconds. A PSE equal to 2.5 seconds indicates that the comparison stimulus was perceived as equally long as the static standard. A PSE greater than 2.5 seconds indicates that the comparison stimulus was perceived as longer than the static standard. (a) PSE for scrambled and semantic stimuli without luminance modulation (Experiment 1a). (b) PSE for scrambled and semantic stimuli with luminance modulation (Experiment 1b). JND for scrambled (blue) and semantic stimuli (red). JND values correspond to the temporal sensitivity in each condition. (a) JND for scrambled and semantic stimuli without luminance modulation (Experiment 1a). (b) JND for scrambled and semantic stimuli with luminance modulation (Experiment 1b). Experiment 2 Methods Participants Eight participants were recruited for Experiment 2 (mean age, 18.63 ± 0.89 years). None of them had participated in Experiment 1. Apparatus The apparatus in Experiment 2 was identical to that in Experiment 1. Stimuli The stimuli in Experiment 2 were the same as those in Experiment 1. Procedure Participants were asked to rate the subjective saliency of the stimuli on a scale from 1 to 10. Saliency was defined as the property of an image or visual information that strongly attracts attention or stands out prominently, and participants were explicitly informed of this definition before the experiment. Each trial began with a fixation display presented for a randomly determined duration between 1.0 and 1.5 seconds, followed by a blank screen lasting 0.3 to 0.6 seconds. Subsequently, a stimulus was presented for 2.5 seconds, after which participants used a mouse to rate its saliency by clicking one of ten buttons (1–10) displayed on the screen. They were instructed to assign higher ratings to stimuli that stood out more compared with stimuli in other trials and to use the full rating scale from 1 to 10. There were 16 images × 14 conditions, resulting in a total of 224 trials. All trials were randomized and then divided into two blocks (112 trials per block). Before these, 28 dummy trials (2 images, 1 face and 1 house, across 14 conditions) were administered in the same manner as the main experiment and treated as an initial block to establish the task criteria, without the participants’ awareness. Analysis The subjective saliency for each condition was calculated by averaging the ratings provided by each participant. We then performed a Bayesian repeated-measures analysis of variance in JASP (version 0.19.3; JASP Team, 2025) and computed BFincl for the effects of flicker frequency and semantic content. In addition, differences in subjective saliency between the scrambled and semantic conditions at each frequency were evaluated with Wilcoxon signed-rank tests. The experimental data for each participant are available on the Open Science Framework (https://osf.io/egvmf). Results and discussion Flicker frequency strongly affected subjective saliency (Figure 5), with very strong evidence from the BFincl (BFincl = 84.60 for flickers without luminance modulation; BFincl = 54.09 for flickers with luminance modulation). However, unlike the PSE and JND in Experiment 1, subjective saliency was significantly influenced by semantic content, with the BFincl, indicating strong evidence for flickers without luminance modulation (BFincl = 33.53) (Figure 5a) and anecdotal evidence for flickers with luminance modulation (BFincl = 2.42) (Figure 5b). In accordance with this overall effect, subjective saliency differed significantly between the scrambled and semantic conditions at all tested frequencies. For the conditions without luminance change, differences were significant at static (V = 0; p = 0.014), 2 Hz (V = 0; p = 0.014), 4 Hz (V = 0; p = 0.014), and 6 Hz (V = 0; p = 0.014). For the conditions with luminance change, significant differences were observed at 2 Hz (V = 3; p = 0.042), 4 Hz (V = 3; p = 0.042), and 6 Hz (V = 1; p = 0.021). Together, these results indicate that subjective saliency increased with both flicker frequency and the presence of semantic information. Subjective saliency of scrambled (blue) and semantic stimuli (red). (a) Subjective saliency of scrambled and semantic stimuli without luminance modulation. (b) Subjective saliency of scrambled and semantic stimuli with luminance modulation. General discussion The present study aimed to examine the contribution of higher-order visual processing to time perception by testing whether semantic information modulates FITD, a robust and widely studied temporal illusion. The FITD refers to the perceptual expansion of time induced by temporally modulated visual stimuli, and previous research has suggested that its magnitude is largely determined by the subjective perception of temporal changes (Herbst et al., 2013; Li et al., 2020). At the same time, a prior study (Houshmand Chatroudi & Yotsumoto, 2025) reported no significant difference in FITD magnitude between scrambled and semantic stimuli, leading the authors to conclude that the ventral visual pathway does not contribute to the FITD. However, their study relied on smoothly modulated flickers, which may have generated only weak flicker percepts, potentially insufficient to fully engage the mechanisms underlying temporal dilation. Given that the FITD is primarily driven by subjective perceptual experience (Herbst et al., 2013; Li et al., 2020), weak or subtle flicker signals may fail to elicit the full extent of the dilation effect. This limitation is particularly relevant in evaluating the contribution of the ventral pathway because if FITD is primarily driven by subjective perceptual experience, sufficiently salient flicker percepts of semantic stimuli (e.g., faces and houses) may be necessary for any contribution of semantic processing to be reflected in the observed effect. To address this issue, the present study used square-wave modulated flickers, which produce more abrupt and pronounced changes. These stronger flicker percepts are expected to enhance the subjective saliency of temporal modulation and potentially reveal contributions from ventral stream processing that were not evident under smoothly modulated conditions. Furthermore, because the prior study examined only a single temporal frequency (2 Hz), the absence of semantic effects reported previously leaves the possibility that the null result stemmed not from a genuine lack of semantic influence, but rather from properties specific to low-frequency stimulation. When only low-frequency flicker is used, semantic-related contributions may remain undetectable simply because the flicker signal is too slow to engage the relevant mechanisms. To address this limitation and clarify whether the previously observed null effect generalizes beyond 2 Hz, the present study examined flickers at three temporal frequencies (2, 4, and 6 Hz). By testing a wider range of stimulation rates, this design allowed us to determine whether the lack of semantic modulation observed earlier reflects an intrinsic characteristic of the FITD or merely a consequence of relying on a single, low flicker frequency. In addition to broadening the range of temporal frequencies, a key feature of the present study is that it measured saliency as a subjective experience for both semantic content and flicker frequency. Whereas previous research, such as Herbst et al. (2013), examined the relationship among flicker frequency, perceptual saliency, and temporal distortion primarily in terms of the vividness or intensity of temporal changes, the current study places temporal and semantic factors in parallel by assessing both through subjective saliency ratings. This design enables a direct comparison of their respective contributions to FITD and provides a unified test of whether two different forms of saliency exert similar or dissociable influences on temporal perception. However, an important limitation of the present study is that temporal bias and subjective saliency were assessed in independent samples of participants. Consequently, the relationship between these two measures was not directly examined within the same individuals, but was instead inferred at the group level. Although the parallel assessment of temporal and semantic saliency provides a useful basis for comparison, the absence of within-participant data limits the strength of conclusions regarding their correspondence. Based on the conceptual structure proposed by Houshmand Chatroudi and Yotsumoto (2025), our conditions were designed to differentiate processing pathways. Semantic flickers were expected to strongly engage higher-order regions along the ventral visual pathway, particularly those selective for faces and houses, including the fusiform face area and the parahippocampal place area, while steadily activating early visual areas (V1/V2) through continuous low-level visual input (Koenig-Robert et al., 2015). Although early visual cortex is necessarily driven by the basic visual properties of the stimuli, they demonstrated that semantic flickers do not elicit pronounced periodic responses in V1/V2, in contrast with higher-order ventral regions. In contrast, semantic luminance flickers were designed to strongly drive lower-level visual processing in early visual areas (Di Russo et al., 2007; Houshmand Chatroudi & Yotsumoto, 2025; Koenig-Robert et al., 2015; Pastor, Artieda, Arbizu, Valencia, & Masdeu, 2003) while simultaneously engaging higher-order ventral regions involved in semantic processing. Scrambled flickers and scrambled luminance flickers were created by removing semantic information from the semantic and semantic luminance flickers, respectively. A further methodological constraint concerns the range of flicker frequencies examined, which was restricted to 2, 4, and 6 Hz. This is because semantic flickers produced a much weaker flickering percept and appeared nearly static at higher frequencies. This constraint may have prevented a comprehensive assessment of the full frequency range across which temporal distortion and early visual area responses are maintained, as previously observed for luminance flickers (Herbst et al., 2013). Nonetheless, Koenig-Robert and VanRullen (2013) reported that significant frequency-tagged responses to semantic flicker were observed at frequencies between 1.4953 and 4.3243 Hz, but not at 6.9700 Hz or higher. Thus, although the highest frequency tested in the present experiment was 6 Hz, this upper limit is sufficient for assessing the involvement of higher-order ventral pathway regions in the FITD. These results showed that the magnitude of the FITD increased with higher flicker frequencies for both scrambled and semantic stimuli, regardless of the presence or absence of luminance modulation. This increase corresponded with higher subjective saliency ratings for higher-frequency stimuli, suggesting that temporal distortion is modulated by the perceptual saliency associated with flicker frequency. Previous studies have shown that neural activity, as reflected by blood oxygen level-dependent signal increases, in early visual areas increases with temporal frequency up to approximately 10 Hz for luminance-driven flickers (Chai et al., 2019; Gentile et al., 2024; Singh, Kim, & Kim, 2003), whereas higher-order ventral visual regions such as the fusiform face area and the parahippocampal place area show maximal responses at lower temporal frequencies, around 6 Hz, when stimulated with face- or house-related flickers (Gentile & Rossion, 2014; McKeeff, Remus, & Tong, 2007). In light of these physiological findings, the present observation that both subjective saliency and FITD magnitude increased at higher flicker frequencies may reflect the contribution of stronger neural responses to the perceptual vividness of flicker and the temporal expansion. Importantly, however, FITD magnitude was unaffected by manipulations of semantic content, regardless of luminance modulation. This finding, consistent with the results reported by Houshmand Chatroudi and Yotsumoto (2025) at 2 Hz, suggests that higher-order regions along the ventral visual pathway do not play a substantial role in generating the FITD effect. Furthermore, because semantic information did not modulate the FITD, even under luminance-modulated conditions, it appears that additional engagement of higher-order ventral regions does not strengthen the temporal illusion beyond the level achieved by lower visual processing. Overall, these findings further clarify the relationship between neural activity, subjective saliency, and temporal distortion. Although both higher flicker frequency and the presence of semantic information enhanced subjective saliency, only higher frequency increased the magnitude of the FITD. This dissociation suggests that the subjective vividness or saliency of a stimulus does not directly explain temporal dilation. Nevertheless, this finding does not necessarily invalidate previous studies that have attributed FITD to the subjective saliency of visual stimuli (Herbst et al., 2013; Li et al., 2020). Those studies characterized the saliency of flickering stimuli as the perceived magnitude of temporal changes, which is known to peak at around 8 to 15 Hz (Kelly, 1961; Shady, MacLeod, & Fisher, 2004). On the contrary, saliency in the present study was defined as the degree to which a stimulus attracts attention or stands out from other stimuli, allowing us to examine the effects of flicker frequency and semantic information within the same framework. From this perspective, the increase in saliency for faster flickering stimuli may reflect the perceptual vividness of temporal changes, whereas the higher saliency of semantic stimuli may reflect their stronger impression or greater novelty relative to the scrambled condition, leading to higher ratings of attentional capture and visual prominence. Therefore, the absence of an FITD enhancement for semantic stimuli, despite their higher saliency, does not necessarily contradict the saliency account of FITD, because these two forms of saliency may be qualitatively distinct. Conversely, the finding that higher-frequency flickers enhanced both subjective saliency and FITD is consistent with the earlier studies. Building on this interpretation, the present results further suggest that the FITD is closely linked to the level of neural activation within the visual hierarchy. Specifically, the FITD increased with higher flicker frequency across all conditions, including those primarily engaging lower visual areas (scrambled flicker and scrambled luminance flicker), those engaging higher-order areas (semantic flicker), and those engaging both (semantic luminance flicker). This pattern is consistent with the view that the temporal dilation effect scales with the overall strength of visual cortical activation rather than with the type of regions involved (Houshmand Chatroudi & Yotsumoto, 2025). Nevertheless, semantic information itself did not modulate FITD, regardless of whether higher-order regions were engaged alone or together with lower visual areas. This finding implies that the recruitment of higher-order ventral regions contributes little to the expansion of perceived duration beyond what is driven by activity in early visual cortex. Therefore, enhancement of the FITD is primarily attributable to increased neural activity in lower visual areas (V1/V2), whereas higher-order visual areas play a minimal and non-causal role in this process. In sum, the effect of visual modulation on perceived time appears to depend more strongly on low-level stimulus properties and early sensory activity than on higher-order processing, including semantic representations. Conclusions Collectively, the present findings suggest that the magnitude of FITD scales with the strength of low-level temporal modulation and the associated activation of early visual areas, rather than with higher-order representational processing. At the same time, however, prior research has shown that semantic and conceptual factors can influence perceived duration under other circumstances. For instance, images of motorcycles are judged to last longer than bicycles, suggesting that conceptual or action-related meaning can modulate temporal judgments (Mioni et al., 2015). Likewise, Ma et al. (2024) demonstrated that more memorable images, which are associated with stronger activation in ventral visual regions, tend to be perceived as longer in duration. In addition, static images that imply motion have also been shown to produce time dilation, despite the absence of physical movement (Yamamoto & Miura, 2012). These findings collectively indicate that higher-order visual representations, including semantic meaning, memorability, and implied motion, can shape duration perception when temporal judgments are sensitive to representational factors. In contrast with these findings, the absence of semantic modulation in the present study is theoretically informative. Although semantic structure reliably increased subjective saliency, it did not alter the magnitude of the FITD. This dissociation suggests that the FITD is governed by mechanisms that are functionally distinct from those underlying semantic- or memorability-based distortions of time. Rather than reflecting the engagement of ventral stream representations, the FITD appears to depend predominantly on early sensory processing and the intensity of temporal dynamics. Thus, although higher-order meaning and memorability can influence perceived duration in some contexts, the present results indicate that the FITD is largely constrained by low-level visual activity and temporal modulation itself.