Human Vision and Electronic Imaging 2026 Program

MONDAY 2 MARCH 2026

Perception of Augmented and Mixed Reality I (Joint Session with Stereoscopic Displays and Applications and Engineering Reality of Virtual Reality)

Session Chair: Alex Chapiro, Meta
08:30 - 10:30
Grand Peninsula A

08:30HVEI-207
HVEI KEYNOTE: How human vision science shapes future displays, Qi Sun, NYU (US) [view abstract]

Wearable display systems offer groundbreaking opportunities for rendering virtual content, sensing physical environments, and precisely tracking human behavior. With rapid advances in display optics, sensing, and machine learning, the hardware software pipeline of displays is undergoing a fundamental transformation. However, this transformation also demands a new understanding of how human perception and cognition interact with visual computing systems. In this talk, I will present our recent research on modeling visual perception and behavior to develop efficient and user-centric display systems. We explore how perceptual constraints can guide the optimization of rendering and sensing pipelines, enabling XR systems that allocate computation only where the human visual system is most sensitive. By combining psychophysical modeling, computational imaging, and generative methods, our work aims to align machine efficiency with human perception paving the way toward adaptive, perceptually grounded wearable displays that enhance realism, comfort, and performance. In particular, I will discuss how human visual acuity in color and luminance connect to the power consumption of near eye displays to answer the fundamental question "what's the visual quality gain per Watt of power usage?", and how perceptually guided algorithms may enable significant battery life extension. Additionally, I will share new insights on the necessity to further understand the consequential human behaviors --- such as how fast do we react to visual targets --- to avoid negative effects and risks for users using wearable displays in daily life.

09:20HVEI-208
Enhancing digital rear-view mirrors in trucks using overlayed graphics, Felicia Toerner, Lulea University of Technology (Sweden); Edith Shaodong Rosen, Lule University of Technology (Sweden); Shirin Rafiei, RISE Research Institutes of Sweden (Sweden); Bo Schenkman, RISE Research Institutes of Sweden (Sweden); Anders Djupsjoebacka, RISE Research Institutes of Sweden (Sweden); Boerje Andren, RISE Research Institutes of Sweden (Sweden); Kjell Brunnstroem, RISE Research Institutes of Sweden (Sweden) [view abstract]

As the automotive industry becomes more digitalized, safety and driver experience demand increase. One development is the shift from traditional side mirrors to digital ones a.k.a. camera monitoring systems, which offer better aerodynamics, wider views, and improved visibility. However, depth perception remains a challenge, and drivers may distrust systems perceived as unclear or unreliable. This study explores how different camera monitoring systems augmentations affect distance estimation during overtaking. Thirty participants viewed video clips across three road scenes using four interface types: one baseline and three with AR enhancements. They estimated vehicle distances, rated uncertainty, and shared preferences. The concept of using distance lines and vehicle outlines yielded the best results in both accuracy and user experience. It was seen as clear, reliable, and modern. A learning effect was noted, as baseline estimations improved after exposure to the augmented interface. The study recommends intuitive depth cues, accessible colors, and well-timed visual elements. Future research should explore sound cues, symbolic warnings, and long-term user acceptance.

09:40HVEI-209
The impact of blur on motion parallax and binocular disparity, Sophie Kergassner, Universita della Svizzera italiana (Switzerland); Piotr Didyk, Universit della Svizzera italiana (Switzerland) [view abstract]

Foveated rendering is a key technique for reducing computational load in immersive display systems by lowering image quality in the peripheral visual field while preserving high fidelity in the fovea. While its impact on spatial detail is well understood, its influence on other visual qualities such as depth from motion parallax remains unclear. In this work, we investigate how foveated rendering affects motion-based depth perception across the visual field. Building on our previous work on binocular disparity, we use a comparable experimental setup to isolate motion parallax as the sole depth cue and measure depth discrimination thresholds under varying levels of blur and eccentricity. Our initial results show that depth from motion is significantly impaired even by mild foveation, with stronger impairments at higher levels of blur. These findings suggest that motion-based depth cues may be more sensitive to foveated rendering than disparity cues, which we previously found to be largely unaffected in our earlier work. This research provides new insights into how different depth cues are affected by foveation and can guide the design of perceptually optimized foveation strategies for VR and AR applications.

Color Imaging I

Session Chair: Damon Chandler, Ritsumeikan University
15:30 - 17:30
Grand Peninsula A

15:30HVEI-210
Efficient ultra-high-resolution hyperspectral reconstruction using a patched input spatial-spectral transformer, Jinhyeok An, Samsung Electronics (Republic of Korea); Wonkyung Jung, Samsung Electronics (Republic of Korea); Jintae Jang, Samsung Electronics (Republic of Korea); Jeongwook Lee, Samsung Electronics (Republic of Korea); Sung-Su Kim, Samsung Electronics (Republic of Korea); Yitae Kim, Samsung Electronics (Republic of Korea) [view abstract]

Transformers, which have demonstrated remarkable performance improvements in natural language processing, have been increasingly adopted in computer vision tasks since the introduction of the Vision Transformer (VIT). In hyperspectral image (HSI) reconstruction, Transformer-based models have gained popularity due to their ability to capture global dependencies. While these models alleviate certain limitations of convolutional neural networks (CNNs), their computational complexity scales quadratically with spatial resolution, making ultra-high-resolution reconstruction infeasible. Spectral Transformer variants have been proposed to mitigate the spatial resolution burden, yet they still face challenges in handling ultra-high-resolution imagery.In this work, we propose a ''Patched Input Spatial-Spectral Transformer (PssT)'' that efficiently reconstructs HSIs from ultra-high-resolution RGB images. The model integrates a spatial transformer before spectral processing, enabling global context awareness while maintaining computational efficiency through in-model patch partitioning. Although performance slightly decreases for low-resolution inputs compared to state-of-the-art (SOTA) models, our method achieves the highest reconstruction quality for ultra-high-resolution inputs, outperforming existing approaches in PSNR while significantly reducing memory consumption.

15:50HVEI-211
Do photobooks need additional (ISO) standardization?, Reiner Fageth, CEWE Stiftung & Co. KGaA (Germany); Birte Stadtlander, CEWE Striftung & Co. KGaA (Germany) [view abstract]

Photobooks designed with personal images are still growing worldwide. They are used to document holidays, exhibitions, the growing up of babies, family gatherings, personal hobbies as well as weddings and are very often used as highly appreciated gifts. Also, the shift from digital cameras (point & shoot as well as digital SLRs) to smartphones did not slow down this trend. They are used for compelling storytelling as well as preserving the images. For customers, important factors include image quality, print quality, image permanence (of both cover and pages), binding stability, and - last not least - a wide range of available products with respect to sizes, binding possibilities, cover types and different papers including silver halide. As members of the ISO TC/42 Photography WG 2 experienced bad binding quality in the US, they suggested to verify if it is beneficial to establish a standard for photobook permanence/ durability so that consumers can expect a very well-produced product, which is normally not very cheap and will fulfill the desire to preserve the images in a well-functioning product. The proposal is referred to ISO 18952 - Imaging Materials Photo Books Characteristics for long-term permanence." In order to objectify the permanence issues a survey was commonly designed and numerous manufacturers and distributors on three continents (North America, Europa and Asia) were asked to participate. This paper will show the questions, the results and the derived results for the ISO committee also referred to above listed existing standards.

16:10HVEI-212
The preservation and repackaging of color and tonal metadata, Coleman Earlywine, University of Kentucky (US); Henry Dietz, University of Kentucky (US) [view abstract]

Modern raw workflows depend as much on metadata as on pixel values. In earlier work we showed that super-resolution outputs from parsek could display objectionable color and tonal shifts when opened in raw processors due to missing or mismatched metadata. Here we present a practical remedy, KYDNG: a DNG repackaging path that embeds raw image data together with camera consistent metadata, so processors treat the file like the original raw capture. Our implementation writes raw specific metadata tags, the correct CFA pattern and repeat dimensions, and derives camera color information directly from the source raw (ColorMatrix1, AsshotNeutral, White/Black level, make/model, DNG versioning), while also embedding a JPEG preview. We demonstrate that RawTherapee and similar tools then recognize the file as raw and apply their existing camera pipelines, restoring expected color and tonality without bespoke profiles. This approach generalizes to multiple cameras and super-resolution methods by delegating color science to the downstream raw processor via faithful metadata replication and retrieval.

16:30HVEI-213
Closed-Loop Color Refinement in Camera ISP via Post-ISP Color Feedback, Huanzhao Zeng (US) [view abstract]

A closed-loop color feedback algorithm that leverages post-ISP statistics to improve camera color quality is presented. Unlike traditional approaches, which evaluate white balance and color early in the pipeline and tune individual modules in isolation, the proposed method assesses color near the end of the ISP pipeline, compares it against target perceptual colors, and feeds the resulting deviations back to upstream processing blocks. This enables dynamic adjustment of AWB and color-related parameters to achieve desired perceptual color outcomes. The framework addresses key limitations of conventional color processing, including (1) evaluating AWB in the raw domain where perceived color cannot be reliably assessed, (2) the inability of fixed color-tuning parameters to compensate for deviations introduced by other ISP blocks, and (3) the lack of coordinated color evaluation across modules. We further demonstrate an application of this framework for skin-tone improvement. The system takes face regions, filters non-skin pixels, computes representative skin color statistics, compares them with target skin colors, and derives adjustment parameters that update color tunings for the current or subsequent frame. This example illustrates the flexibility and effectiveness of the proposed closed-loop approach for perceptually guided color enhancement or accurate color reproduction.

16:50HVEI-237
A perceptual uniformity error metric for standard and high dynamic range colour spaces, Maryam Azimi, University of Cambridge (United Kingdom); Minjung Kim, University of Cambridge (United Kingdom); Graham Finlayson, University of East Anglia (United Kingdom); Rafal Mantiuk, University of Cambridge (United Kingdom) [view abstract]

In perceptually uniform colour spaces, the perceptual differences in colour pairs are approximately the same as the Euclidean distance between them. Uniformity is of great importance in applications such as gamut mapping where the perceptual difference between original and mapped colour needs to be minimised. Ideally, in a perceptually uniform colour space, the locus of constant JND around different colour samples should be the unit sphere. While several perceptually uniform colour spaces for SDR and HDR have been proposed, there is no standardized uniformity metric with respect to which we might judge whether one space is more uniform than another. In this paper, we propose and develop such a uniformity metric. Importantly, our approach takes into account changes in all three directions of a colour space including luminance and this is in contradistinction to prior art that focuses mainly on the colour signal (separate from luminance). The proposed metric can be based on any perceptual colour difference metric that models JNDs.

TUESDAY 3 MARCH 2026

Color Imaging II

Session Chair: Bernice Rogowitz, Visual Perspectives
08:30 - 09:30
Grand Peninsula A

08:30HVEI-214
Perceptual study of real-world colors on ultra-WCG displays, Farnaz Agahian, (US); Sofie R. Herbeck, Dale Stolitzka [view abstract]

This study builds upon our previous work, where we analyzed the range of real-world colors and identified images containing colors that exceed the boundaries of legacy color gamuts such as sRGB and DCI-P3, making them difficult for traditional displays to render accurately. In our current research, we conducted a series of visual experiments to evaluate perceptual differences and viewer preferences when such images are displayed on ultra-WCG displays compared to standard-gamut displays. Our findings indicate that observers could consistently distinguish between images shown on an ultra-WCG display and the same images calibrated to sRGB. The perceptual difference between DCI-P3 and ultra-WCG was notably smaller, resulting in lower detection rates that were more content-dependent. Overall, observers showed a strong preference for the ultra-WCG display, regardless of the viewing condition or the image content.

08:50HVEI-215
The bi-monocularity of human color sensation and implication for color filter settings in eye-based devices, Charles Wu, Perception and Cognition Research (US) [view abstract]

It is well established that a small percent of the color deficiency population is monocular (Judd, 1948; Broackes, 2010). On the other hand, it is also well known that under certain conditions, binocular fusion of colors (including brightness) does occur. Considering the two sides together, human color sensation is bi-monocular. Furthermore, relating the La Hire phenomenon about the physiological blind spot to the neuroanatomical finding that the blind spot is represented in V1-L4, we can infer that V1-L4 is the neural substrate for color sensations in the human brain. This neural substrate is bi-monocular in that the excitatory neurons there are monocular in terms of thalamic inputs but the two eyes monocular neurons inhabit there side by side: Together they can represent binocular information. In short, bi-monocularity is a prominent attribute of color vision worthy of further investigation. This quality of color sensation has an obvious and important implication for devices that contain eye-based displays: For example, presently, all the commercially available VR headsets (e.g., Apple s Vision Pro and Meta s Quest products) do not have separate color filter settings for the two eyes of an individual user: This feature is worthy of enhancement for the uniocular color deficient population.

09:10HVEI-216
Refining half-space methods for metamer mismatch bodies via ensemble vertex selection, Alexander Forsythe, Simon Fraser University (US); Brian Funt, Simon Fraser University (Canada) [view abstract]

Existing half-space methods for calculating Metamer Mismatch Bodies rely on a one-shot random sampling of a 6D hypersphere. However, uniformity on this initial hypersphere in k-space does not translate to a uniform vertex distribution in the final 3D sensor-response space. Furthermore, we demonstrate that a single sampling run is not exhaustive. We introduce a novel two-stage ensemble method to address these limitations. First, an ensemble generation process, which combines multiple stochastic runs, produces a more comprehensive vertex super-set. Second, an ensemble of Farthest-Point Samplings intelligently selects a final subset, which is both dense and exhibits significantly improved uniformity, providing a high-fidelity sampling ideal for accurate visualization and analysis.

AI & XR for Education (Joint Session with Stereoscopic Displays and Applications and Engineering Reality of Virtual Reality)

Session Chair: Bjorn Sommer, Royal College of Art
11:40 - 12:40
Grand Peninsula D

11:40SDA-340
Virtual reality in engineering education: A survey of applications, trends, and challenges, Rojin Manouchehr, University of Nevada, Reno (US); Sergiu Dascalu, University of Nevada, Reno (US) [view abstract]

The use of Virtual Reality (VR) in education, specifically in engineering, is rapidly increasing due to the immersive and interactive learning environments provided by them, that go beyond traditional classroom methods. While most studies report benefits such as enhanced engagement and improved learning outcomes, the literature remains fragmented across different domains and methodologies. Synthesizing representative works is essential to capture both persistent challenges and emerging opportunities as this is a growing field. This work presents a survey of VR in engineering education, drawing on representative studies from 2012–2025. We group applications into three categories: conceptual and visualization learning, procedural and safety training, and gamified or collaborative approaches. Our analysis highlights consistent benefits of VR for engagement and visualization, identifying unresolved issues such as evaluation rigor, scalability, and curricular integration. This survey highlights the importance of continued synthesis in a rapidly expanding research area.

12:00SDA-341
MAIVE: A multi-agent AI-driven immersive virtual reality environment for astronomy education, Francia Fuentes Riesco, Colorado State University (US); Marie Vans, Colorado State University (US) [view abstract]

12:20SDA-342
A comparative study on memory strategy adaptation in XR vocabulary learning, Nicko Caluya, Ritsumeikan University (Japan); Damon Chandler, Ritsumeikan University (Japan); Minxu Yang, Ritsumeikan University (Japan) [view abstract]

With the rapid advancement of extended reality (XR) technologies in education particularly Augmented Reality (AR) and Virtual Reality (VR) there is a growing need to understand how these platforms influence cognitive learning processes. Vocabulary acquisition, a core aspect of second-language learning, heavily relies on memory strategies. However, it remains unclear how platform-specific features such as contextual anchoring in AR or spatial immersion in VR interact with different strategies to affect learning outcomes.This study examines how two key memory strategies semantic association and spatial positioning perform in AR and VR environments during second-language vocabulary learning. Specifically, it investigates whether a strategy platform compatibility exists, where certain strategies may be more effective depending on the platform s cognitive affordances.

Perception of Augmented and Mixed Reality II

Session Chair: Nicko Caluya, Ritsumeikan University
15:30 - 17:30
Grand Peninsula A

15:30HVEI-217
The influence of driver age on judging the distance of approaching vehicles in digital rear-view mirrors, Gabriele Pifferi, University of Trento (Italy); Shirin Rafiei, RISE Research Institutes of Sweden AB (Sweden); Bo Schenkman, RISE Research Institutes of Sweden AB (Sweden); Irene Sperandio, University of Trento (Italy); Anders Djupsjoebacka, RISE Research Institutes of Sweden AB (Sweden); Boerje Andren, RISE Research Institutes of Sweden AB (Sweden); Kjell Brunnstroem, RISE Research Institutes of Sweden AB (Sweden) [view abstract]

The automotive industry continues to innovate with technologies aimed at enhancing safety and comfort, such as Advanced Driver-Assistance Systems (ADAS) and Camera Monitor Systems (CMS). CMS, a digital alternative to traditional side mirrors, offers benefits like improved aerodynamics and visibility but also introduces challenges in depth perception and user adaptability. This study investigates how driver age influences the ability to judge distance and make overtaking decisions using CMS. Building on Thulinsson & Soederlund (2024), the study replicates their design while adding age as a key variable. Participants viewed 36 realistic driving scenarios on a 65-inch display, simulating CMS perspectives with varying camera heights and fields of view (FOV). Two tasks were assessed: Distance Judgment (Dist) and Last Safe Gap (LSG). Results are expected to show that wider FOVs improve distance accuracy, while narrower FOVs increase underestimation and safety margins, especially among older drivers. The study also explores subjective experiences, including confidence, realism, and attitudes toward CMS. Findings aim to inform age-inclusive CMS design by identifying perceptual and behavioral differences across demographics, ultimately contributing to safer and more user-friendly vehicle interfaces.

15:50HVEI-218
Cross-modal interaction of color and texture in mixed reality: Evidence for visual modulation of tactile perception, Elena Fedorovskaya, Rochester Institute of Technology (US); Alireza Rabbanifar, Rochester Institute of Technology (US); Pratheep Chelladurai, Rochester Institute of Technology (US); Mekides Abebe, Rochester Institute of Technology (US) [view abstract]

Visual color and tactile texture cues jointly shape our perception of objects and surfaces. Previous research shows that visual color can influence perception of materials and alter how touch is experienced (Zaidi, 2011; Ludwig et al., 2013; Chylinski, et al., 2015). Understanding such cross-modal effects is especially relevant for immersive extended-reality (XR) systems, where accurate object identification relies on the integration of multiple sensory inputs. In this study, participants wearing augmented-reality headsets touched concealed texture samples while viewing visual images of textures that varied in color. Participants judged the perceived similarity between the felt and seen textures. Results revealed that the color of visually presented textures systematically biased tactile identification: hues semantically associated with hard or smooth surfaces exerted the strongest influence on perceived texture. These findings highlight the role of color touch interactions in shaping multisensory perception within XR environments and have implications for the design of more perceptually coherent mixed reality experiences.

16:10HVEI-219
Utility of color symbology in optical see-through head-mounted displays, Austin Erickson, KBR (US); Eric Seemiller, KBR (US); Marc Winterbottom, AFRL (US) [view abstract]

Imagery from optical see-through (OST) head-mounted displays (HMDs) is perceived as a blending of light emitted by the display added to the light from the user s physical environment, which can result in color distortions and desaturation of the virtual imagery. Due to these limitations, the user's ability to distinguish between colors shown on the display may be reduced compared to more traditional types of displays, which may impact the interpretation of the symbology, and potentially reduce performance. Further, individual variation in color perception may also impact the utility of color symbology in OST HMDs. In this paper, we present a user study that investigated the utility of color-coded symbology displayed on an OST Augmented Reality (AR) display within a flight simulator with variable scene imagery. In experiment 1, we examined color discrimination thresholds across four types of static backgrounds for color normal and color deficient participants. In experiment 2, we compared performance between the two groups in a dynamic flight simulator and investigated effects of symbology contrast and symbology color set on participant response times, accuracy, and eye behavior.

16:30HVEI-220
Understanding and exploiting the time course of chromatic adaptation for display power optimizations in virtual reality, Yuhao Zhu, University of Rochester (US); Ethan Chen, University of Rochester [view abstract]

We introduce a gaze-tracking--free method to reduce OLED display power consumption in VR with minimal perceptual impact. This technique exploits the time course of chromatic adaptation, the human visual system s ability to maintain stable color perception under changing illumination. To that end, we propose a novel psychophysical paradigm that models how human adaptation state changes with the scene illuminant. We exploit this model to compute an optimal illuminant shift trajectory, controlling the rate and extent of illumination change, to reduce display power under a given perceptual loss budget. Our technique significantly improves the perceptual quality over prior work that applies illumination shifts instantaneously. Our technique can also be combined with prior work on luminance dimming to reduce display power by 31% with no statistical loss of perceptual quality.

WEDNESDAY 4 MARCH 2026

Displays and High Dynamic Range

Session Chair: Alex Chapiro, Meta
08:30 - 10:30
Grand Peninsula A

08:30HVEI-221
HVEI KEYNOTE: How gaze direction and dynamics affect visual resolution, Martin Banks, University of California Berkeley (US) [view abstract]

Humans exhibit machine-like eye movements in space and time while performing demanding acuity tasks. To investigate these, we used an adaptive-optics imaging and display system to present ultra-sharp Vernier acuity stimuli briefly every two seconds while simultaneously measuring eye movements, including precisely where on the retina each stimulus fell. We found that drifts and microsaccades combined to confine the landing location of the anticipated stimulus to a tiny retinal region centered on the preferred retinal locus (PRL). The variance of landing location was smallest at the time of stimulus presentation and a few hundred milliseconds after. We correlated where the stimulus fell in space and time with correct or incorrect responses. The PRL and a small area around it, including the anatomical fovea, conferred the best acuity. Acuity declined consistently in the rare events in which the stimulus fell more than 7 10minarc from the PRL. We also found that acuity was best when the last microsaccade occurred sufficiently prior to stimulus presentation. Our findings reveal a highly evolved oculomotor system where gaze direction during fixation is rarely far enough from the PRL to compromise visual resolution when a person makes natural fixational eye movements.

09:20HVEI-222
Perceptual impact of peak luminance and contrast in direct view HDR display, Kenneth Chen, New York University (US); Yunxiang Zhang, New York University (US); Qi Sun, New York University (US); Alexandre Chapiro, Meta (US) [view abstract]

Characterization of a high dynamic range (HDR) display s performance can be largely defined by its contrast and peak luminance. Prior work has studied this question in a haploscopic HDR setup, but it is not obvious if those results are transferrable to a more traditional viewing setting, such as direct view. In this work, we conducted a study to measure user preference for different contrast and peak luminance parameters in a direct view scenario, and develop a perceptual just-objectionable-difference (JOD) scale to quantify preference scores. The data is used to develop a model that can drive display design.

09:40HVEI-223
Visual perception of mobile displays depending on the combination of the luminance of the display and the lighting conditions, Hijiri Okumura, Yokohama National University (Japan); Katsunori Okajima, Yokohama National University (Japan) [view abstract]

Mobile displays, such as smartphones and tablets, are used across a wide range of light conditions. Readability and visual comfort vary greatly depending on the surrounding light environment as well as the luminance of the display. To evaluate effects of the light environment, we developed a two-booth laboratory system capable of independently manipulating three factors: the illuminance on display surface, the luminance behind the display and the ambient illumination in the user's space. Participants viewed black text on a white smartphone screen under various light conditions, rating its readability, discomfort glare, and screen comfort across multiple luminance levels. The results demonstrated that all three factors had a significant effect on perception. Illuminance on the screen had the most powerful effect on readability, while the factors interacted, offsetting each other's effects. It also became clear that the user's light-dark adaptation state had an effect. These findings indicate that a display's visual characteristics are determined not by individual factors, but by their complex combination. Based on these results, we demonstrate the characteristics of human perception of the screen with respect to the three factors.

Recognition and Higher-Level Judgements

Session Chair: Kjell Brunnstroem, RISE
11:00 - 12:20
Grand Peninsula A

11:00HVEI-224
Judging correlation in line drawings and scatterplots, Bernice Rogowitz, Visual Perspectives (US) [view abstract]

A very common data analysis task is to judge the degree to which two variables are correlated. Anscombe s (1973) convincing demonstration (1) shows clearly that two variables with the same mean, standard deviation and correlation may have very different shapes (and meanings). Previously, in this conference, Rogowitz and Borrel (2023) showed that the same data, rendered differently, can produce very different impressions (2). In this paper, we show how the correlation between the same two variables can appear to be very different, depending on how these time series data are rendered. The left-hand graph in Figure 1, showing two time-series on the same axis, was created by a student in my data visualization course at Columbia University. Since the peaks and troughs in the line drawing version seem to track, he concluded that the two variables were highly correlated. Comparing this plot with the tick-by-tick scatterplot of corresponding data values revealed the low correlation, and led to this experiment. In this paper, we report a pilot experiment in which pairs of variables were presented either as two time-series plots on the same graph or as a scatterplot. 10 observers judged each graph and reported on the degree of correlation on a 7-value Likert scale (strong negative, moderate negative, weak negative, no correlation, weak positive, moderate positive, strong positive.) The also rated their degree of confidence on a 7-point scale. After each judgment, they were invited to talk about the shape of the data and their interpretation of what the data showed. The pairs of graphs were drawn from a real dataset of economic variables spanning 25 years, sampled quarterly. Seven pairs of plots were selected, representing a wide range from strong negative to strong positive correlation. Each plot was presented in counterbalanced order. We found that the scatterplot led to more accurate judgments of correlation, but that the time series plot provided deeper insight into the shape of the relationship between two variables. For example, two variables could have a similar shape over time and not have a high correlation if there were a lead or lag in the evolution of the data over time. We present the results of this pilot experiment and discuss quantitative and qualitative differences in how these plots are judged. We also discuss how differences in these rendering geometries lead the observer into different interpretations. For example, the two line drawings provide an intrinsic time scale, which provides instant insight into how the relationship between the two variables evolves over time. The scatterplot representation loses the temporal ordering of the data pairs, but the degree to which the points organize around a single line provides instant insight into the degree of correlation (Rensink, 2017). We also discuss how comparing sets of points in the two types of representations, using brushing, can enhance understanding (Becker and Cleveland, 1987).The choice of representation is not only a matter of adhering to good visualization practice. All the graphs in this experiment are created according to standard visualization guidelines. Yet, we observe that different choices in how the data are rendered can influence the types of judgments that can be drawn and degree to which those judgments accurately reflect patterns intrinsic to the data. Understanding how our high-level judgments are influenced by the shape of the data, not just the values they represent, is a promising avenue for future research.

11:20HVEI-225
Frequently selected orientations during visual production do not promote visual recognition, Griffin Newell, North Dakota State University (US); Laura Thomas, North Dakota State University (US); Benjamin Balas, [view abstract]

Do observers tap the same canonical representations to both recognize and sketch common objects? Drawings from Google s Quick Draw! database show a tendency for users to sketch common utensils (i.e., spoons, forks, & knives) at specific spatial orientations. However, it is unclear if 1) real-world sketchers show orientation preferences and 2) if viewing objects at preferred sketch orientations facilitates recognition. In Experiment 1, we aimed to determine if real-world sketchers favor specific spatial orientations by asking participants to sketch common utensils under varying time constraints. Participants in Experiment 2 attempted to quickly recognize sketches of the same utensils presented at various orientations. We found that the most common sketch orientations from Experiment 1 did not facilitate object recognition in Experiment 2. Our findings suggest the mental prototypes observers use to aid object recognition are distinct from factors influencing sketch orientations. Future work is needed to disentangle whether orientation preferences are due to drawing-related decision making and/or unique mental representations used for actively sketching as opposed to recognition.

11:40HVEI-226
Enhancing emotion estimation accuracy through integrated analysis of heart rate and pupil signals, Tsukasa Yano, Chiba University (Japan); Midori Tanaka; Takahiko Horiuchi, (Japan) [view abstract]

Texture and Generative AI

Session Chair: Ben Balas, North Dakota State University
15:30 - 17:30
Grand Peninsula A

15:30HVEI-230
HVEI KEYNOTE: What we've learned about visual attention, Ruth Rosenholtz, MIT (US) [view abstract]

Early in the study of visual attention, it appeared promising that understanding of preattentive and attentional processes could provide a unifying explanation of a wide range of visual phenomena, by elucidating a critical capacity limit faced by visual processing. However, researchers have uncovered significant anomalies, frustrating hopes of a single predictive mechanism. This state of affairs requires rethinking visual attention from the ground up. This talk provides my take on the critical phenomena to consider in search of a unifying theory. Commonalities between these phenomena suggest not only new ways of thinking of capacity limits and the mechanisms for dealing with those limits, but also visual perception itself and the contents of visual awareness. This rethinking of visual attention points to a new possibility of a unifying theory, in which all perception results from performing a task, and tasks face a limit on complexity.

16:20HVEI-231
A texture coloring task reveals differing representations of positive and negative natural textures, Molly Setchfield, North Dakota State University (US) [view abstract]

style="font-size: small"> lang="en-US">Textures, unlike objects, tend to be well-described computationally via statistical features that capture distributions of oriented contrast at multiple scales and correlations between wavelet-like coefficients. Our understanding of the representations used in the human visual system to discriminate and recognize textures remains incomplete, however. How do specific image statistics contribute affect the appearance of natural textures? Here, we asked participants to complete a texture interpolation task with original and contrast-negated versions of natural textures as a means of examining this question. Contrast negation is a particularly useful manipulation of texture statistics for our purposes insofar as it preserves a wide range of low-level edge statistics while dramatically altering mid- to high-level aspects of appearance. We find that negated textures are completed with different contrast statistics and different spatial frequency content, indicating key differences in how these textures are perceived.

16:40HVEI-232
Real or synthetic? An evaluation of AI-generated audio-visual content, Devi Klein, Dolby Laboratories Inc. (US); Anustup Choudhury, Dolby Laboratories, Inc. (US); Jaclyn Pytlarz, Dolby Laboratories Inc. (US); Scott Daly, Dolby Laboratories Inc. (US) [view abstract]

Generative AI (GenAI) models provide the means for scalable automated multimedia content creation. However, these models often produce video and audio clips replete with artifacts which reduces the content s perceived realism. This study evaluates human perceptual sensitivity to such distortions using a 2-interval forced-choice task with eye tracking. Participants (N=36) viewed paired real and GenAI video/audio clips and decided which of the two was more realistic. This procedure was repeated across four experimental conditions to isolate the relative weighting of visual and auditory cues in the decision-making process. Our results showed that participants were most accurate in the visual-only condition. On the other hand, GenAI audio was more difficult to distinguish from real audio. Gaze analysis revealed that violations of biological motion were the most salient visual artifacts (i.e., attracting the highest fixation rates). Subtler issues, such as texture blending and inaccurate representation of small complex objects, were gazed at less. Our findings contribute a multisensory assessment of GenAI content, highlight the relative perceptual strengths of GenAI audio, and provide a framework for quantifying artifact salience.

Friends of HVEI End-of-Day Discussion and Demos

Join us for an HVEI wrap-up in an informal atmosphere.
17:30 - 19:00
Regency A

THURSDAY 5 MARCH 2026

Perception and Image Quality I (Joint Session with Image Quality and System Performance)

Session Chair: Lukas Krasula, Adobe
08:30 - 09:40
Grand Peninsula A

08:30HVEI-233
Ongoing activities in the Video Quality Experts Group (VQEG), Kjell Brunnstroem, RISE Research Institutes of Sweden AB (Sweden); Ioannis Katsavounidis, Meta (US) [view abstract]

The Video Quality Experts Group (VQEG) was established to unite experts in subjective and objective video quality assessment. Since its first meeting in 1997, VQEG has focused on advancing video quality assessment methods and validating new objective quality metrics for standardization. VQEG is open to all interested parties without any fees or membership requirements, and its activities are documented and submitted to relevant ITU Study Groups or published in scientific journals.Examples of recent progress include: 1) The Immersive Media Group (IMG) has been developing a test plan, based on a large-scale cross-lab experiment, to evaluate the Quality of Experience (QoE) of immersive interactive communication systems, in collaboration with ITU-T, resulting in the Recommendation ITU-T Rec. P1321. 2) The 5G Key Performance Indicators (5GKPI) group is studying the relationship between 5G/6G network performance and video service QoE. They are preparing a VQEG Whitepaper on QoE management in telecommunication networks, which will be published soon.

08:50HVEI-234 IQSP KEYNOTE: The influence of image semantic complexity on the performance of image quality metrics, Marius Pedersen, Norwegian University of Science and Technology (Norway); Peiyuan Zhang, Zhejiang Wanli University (China (Mainland)); Xinwei Liu, Zhejiang Wanli University (China (Mainland)); Sophie Triantaphillidou, Norwegian University of Science and Technology (Norway) [view abstract]

Image quality assessment has been a longstanding area of research, with significant efforts dedicated to developing objective metrics that can reliably predict perceived image quality. While numerous image quality metrics have been proposed, ranging from traditional handcrafted approaches to modern machine learning-based models, their evaluation typically relies on statistical comparisons with subjective human ratings where correlation is reported as the main way to assess their performance. In this study, we explore an additional dimension in image quality evaluation: the impact of image semantic complexity on metric performance. Specifically, we hypothesize that the number of distinct semantic categories within an image influences the accuracy of image quality metrics. We evaluate 8 state-of-the-art image quality metrics across 2 benchmark datasets, analyzing performance variations with respect to image semantic complexity (category count), based on two vision-language models. Our findings reveal that for some image quality metrics there is an impact of semantic complexity and outliers. This could suggest that content-aware evaluation may be crucial for future image quality research.

09:20HVEI-235
JIST-first-2025-007: Quality evaluation of contrast-enhanced images: Central Asians' perspectives, Altynay Kadyrova, KIMEP University; Marius Pedersen, Norwegian University of Science and Technology (Norway) [view abstract]

Culture can play a significant role in evaluating image quality. Therefore, we considered one of the least studied origin regions of observers, the impact of Central Asian culture on image quality evaluation. More specifically, we investigated how they evaluate the quality of contrast-enhanced images. We found that observers evaluations vary and can be divided into groups. These groups may have their individual preferences for the quality of contrast-enhanced images. Therefore, the personalization factor should be incorporated into the quality evaluation of (contrast) enhanced images. Furthermore, we obtained similar results with existing studies that current image quality metrics might not be enough to evaluate the quality of enhanced images. In addition, we introduced the Central Asian Contrast-Enhanced Image Quality Dataset (CACEIQD). Our dataset can be helpful for future research in the field of enhanced image quality evaluation.

Perception and Image Quality II (Joint Session with Image Quality and System Performance)

Session Chair: Lukas Krasula, Adobe
09:40 - 10:40
Grand Peninsula A

09:40HVEI-236
Method and findings for determining the just noticeable difference (JND) for an image sharpness metric, Tero Vuori, Microsoft (Finland); Jukka-Pekka Raunio, Microsoft (Finland); Ari Partinen, Microsoft (Finland) [view abstract]

In camera product development, where the goal is to achieve the best possible image quality and user experience, it is necessary to use both objective and subjective test methods. Both methods have a long tradition and their own advantages and disadvantages. Objective image quality measures are fast and efficient. They form the basis for, for example, camera product comparisons daily basis. Based on the results provided by the measure, it is possible to rank any camera products easily. However, how big is the noticeable difference between two products for a user if such an objective measure is used? When is the difference significant? Or does the user notice the measured difference at all? In this study, we wanted to get answers to these questions for our acutance measure which is used daily basis. We conducted numerous subjective tests in a controlled lab with carefully chosen stimuli. Based on these subjective results, we calculated the corresponding Just Noticeable Difference (JND) values for our acutance measure. This study presents methods and results for finding JND values for an objective acutance measure that can be more broadly generalized to all objective acutance measures and, in terms of the method, to all objective measures. Our result shows that one JND corresponds to about 0.02 acutance units which corresponds to a perceptual difference of approximately 2% in acutance units.

10:20HVEI-238
On a perceptually accurate visual noise metric for HDR imaging: Accounting for luminance adaptation and gradient effects, Hugo Masson, DXOMARK Image Labs (France); Francois-Xavier Thomas, DXOMARK Image Labs (France); Claudio Greco, DXOMARK Image Labs (France); Daniela Carfora Ventura, DXOMARK Image Labs (France); Mauro Patti, DXOMARK Image Labs (France); Benoit Pochon, DXOMARK Image Labs (France); Hoang-Phi Nguyen, DXOMARK Image Labs (France); Laurent Chanas, DXOMARK Image Labs (France); Frederic Guichard, DXOMARK Image Labs (France) [view abstract]

Current visual noise algorithms are primarily developed and calibrated using SDR images, which limits their accuracy in representing the actual noise perceived by humans in HDR content. One key factor often overlooked is the observer s luminance adaptation, especially when there is a significant contrast between the observed patch and its surrounding area. Moreover, the design of existing test charts, combined with increasingly sophisticated local tone mapping algorithms, introduces new challenges. A prominent issue is the presence of gradients in the final image, which significantly affect algorithmic measurements but have minimal impact on human perception for instance, a patch may register a Just Noticeable Difference (JND) of 6 despite appearing visually clean.This paper proposes a new direction for Visual Noise algorithms and lays the groundwork for future research. It presents findings on: A new HDR ruler for visual noise assessment; The impact of various factors (CSF, MTF, HPF, gradient correction, edge windowing) on algorithm performance; Evaluation of different color spaces for computing visual noise metrics