IQSP 2025 Program
Excel to HTML
MONDAY 3 FEBRUARY 2025
Objective Imaging Performance and Simulation I
Session Chair: Peter Burns, Burns Digital Imaging LLC
08:30 - 10:30
Grand Peninsula A
08:30IQSP-238
Estimation of effective SFR variation due to sampling phase and correction for sub-pixel SFR measurement in ISO12233, Robin Jenkin, NVIDIA (US) [view abstract]
Imaging sensors commonly use a sub-pixel array to as a strategy to extend the total dynamic range of exposures while minimizing temporal artifacts. In this scheme, a pixel is divided spatially into two or more sub-pixel cells of varying size. The smaller cells are less sensitive to light, therefore extending the exposure by saturating at much higher light levels than the larger cells.Sub-pixels are of the same order of size (~1/3 pixel) as the typical binning spacing (~1/4 pixel) used in the ISO12233 SFR measurement technique. Because of this, the SFR yielded for these sub-pixels is lower than expected and requires correction for the binning. This paper demonstrates the error via simulation of images and measurement using ISO12233. Subsequently a correction term is derived and shown to be effective removing this effect, Figure 1.The author has previously derived expressions yielding an estimation of the variation in SFR due to misalignment between the signal and sampling grid of the sensor. This effect is exacerbated for pixels with low fill-factors, i.e. sub-pixels, and also increases the potential for aliasing. The work is reviewed and practical implications for safety critical camera systems discussed. Further modification of ISO12233 is suggested to yield the variation in SFR that may be expected for a typical camera module.
08:50IQSP-239
Angle- and noise-robust spatial frequency response-based resolution analysis in natural scenes, Seungwan Jeon, Samsung Electronics Co,. Ltd (Republic of Korea); Subin Han, Samsung Electronics Co,. Ltd (Republic of Korea); Yu Gyeong Lee, Samsung Electronics Co,. Ltd (Republic of Korea); KiChul Park, Samsung Electronics Co,. Ltd (Republic of Korea); Sung-Su Kim, Samsung Electronics Co,. Ltd (Republic of Korea) [view abstract]
Evaluating spatial frequency response (SFR) in natural-scene images is crucial for understanding camera system performance and its implications for image quality in various applications, including machine learning and automated recognition. Traditionally, SFR assessments relied on chart-based method, which restricts its applicability in typical scene images. This reliance on predetermined test patterns limits the ability to evaluate camera performance in natural environments effectively. Fortunately, the introduction of Natural Scene Spatial Frequency Response (NS-SFR) has made it possible to evaluate the camera performance in natural environments without the need for charts [1]. The NS-SFR method primarily relies on the ISO 12233 e-SFR [2], focusing on detecting and extracting suitable step edges within images to assess camera performance. This approach is structured in multiple stages. The process involves applying the Canny algorithm for edge detection, extracting step-edge like regions of interests (ROIs) from the detected edges, filtering the ROIs based on criteria such as angle, contrast, size, and linearity, isolating the edges using pixel stretching techniques, averaging the line spread functions (LSFs) from the edge ROIs, and finally calculating the mean MTF.
10:30 – 11:00 and 15:00 – 15:30 Coffee Breaks
Objective Imaging Performance and Simulation II
Session Chair: Patrick Denny, University of Limerick
15:30 - 17:30
Grand Peninsula A
16:10IQSP-240
Building end-to-end deblur image quality evaluation simulation for Hybrid-EVS-CIS sensor images, Daisuke Saito, OmniVision Technologies (US); Kamal Rana, OmniVision Techinologies (US); Zhiyao Yang, OmniVision Techonologies (US); Wei Zhang, OmniVision Techonologies (US); Bo Mu, OmniVision Techonologies (US); Eiichi Funatsu, OmniVision Techonologies (US) [view abstract]
Event-based Vision Sensor (EVS) generates pixel-level and low-latency event data that is useful for reconstructing temporal components of images in image deblurring. In this kind of development for new application, we need to know how EVS hardware parameters affect the image quality (IQ) to determine hardware specifications before starting design. To realize this approach, it is beneficial to build an End-to-End IQ-simulation which runs from hardware simulations to natural scene IQ evaluation. Previously, we developed Hybrid-EVS-CIS simulator to generate synthesis images fed into an image deblur block. We evaluated blurriness with Blurred Edge Width metric (BEW). In this paper, we extend this End-to-End IQ-simulation to natural scene. We propose an IQ evaluation scheme using Video Multi-method Assessment Fusion (VMAF) and BEW to evaluate both of blurriness and noise as a building block of End-to-End IQ-simulation. We proved that VMAF was applicable not only to videos but also to deblur images. We also introduced a method to assess pixel-speed, which affects blurriness, by utilizing the correlation between VMAF and BEW. It was the last piece to realize End-to-End IQ-simulation to be used in EVS hardware design with checking both of images and their IQ metric value.
16:30IQSP-241
JIST-first ACCEPTED: Performance of Automatic License Plate Recognition Systems on Distorted Images, Nikola Plavac; Seyed Amirshahi; Marius Pedersen; Sophie Triantaphillidou [view abstract]
Automatic License Plate Recognition (ALPR) systems are essential for various applications, including law enforcement, traffic management, and access control. However, their performance can be significantly affected by images distorted by adverse environmental conditions and the imaging pipeline. Three different ALPR systems are used to evaluate their robustness to different distortions using images from six well-known ALPR datasets. Two groups of distortions are the focus of the study: simulated weather conditions (rain, brightness, fog, frost, and snow), and modeled camera read noise in the simulated imaging pipeline. The findings indicate that certain weather distortions drastically reduce the accuracy of ALPR systems, with the accuracy of the systems approaching zero in some cases. Read noise also negatively impacts performance, even at minimal levels. The sensitivity to the introduced distortions varied between different models and datasets. The results underscore the need for robust ALPR system designs that can handle diverse and challenging capturing conditions.
TUESDAY 4 FEBRUARY 2025
Image Quality Challenges for AI, CG, and HDR
Session Chair: Nicolas Bonnier, Apple Inc.
08:50 - 10:30
Grand Peninsula C
08:50IQSP-242
Keynote: Quality versus utility assessment in the era of artificial intelligence, [view abstract]
09:30IQSP-243
3D CG image quality assessment in vision and language based on stable diffusion, Norifumi Kawabata, Computational Imaging Lab (Japan) [view abstract]
In this study, we first generated an image database by adjusting parameters using Stable Diffusion, which is a deep learning model that is also used for image generation based on text input and images. And then, we carried out experiments to evaluate the 3D CG image quality from the generated database, and discussed the quality assessment of the image generation model.
09:50IQSP-244
Analysis of a new HDR dataset of laboratory scenes images using the ICtCp color space, Elodie Souksava, DXOMARK Image Labs (France); Francois-Xavier Thomas, 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]
This paper is the continuation of a previous work [1] which aimed to develop a color rendering model using ICtCp color space, to evaluate SDR and HDR-encoded content. However, the model was only tested on an SDR image dataset. The focus of this paper is to provide an analysis of a new HDR dataset of laboratory scenes images using our model and additional color rendering visualization tools. The new HDR dataset, captured with different devices and formats in controlled laboratory setups, allows the estimation of HDR performances, encompassing several key aspects including color accuracy, contrast, and displayed brightness level, in a variety of lighting scenarios. The study provides valuable insights into the color reproduction capabilities of modern imaging devices, highlighting the advantages of HDR imaging compared to SDR and the impact of different HDR formats on visual quality.[1] Objective color characterization of HDR videos captured by smartphones: laboratory setups and analysis framework, A. Tigranyan et al. (Electronic Imaging 2024)
10:10IQSP-245
Printing image re-conversion method for adopting to standard dynamic range (SDR) images converted from twilight ultra-high-dynamic-range (UHDR) images, Sakuichi Ohtsuka, International College of Technology, Kanazawa (Japan); Shoko Hira, Kagoshima University (Japan); Saki Iwaida, Kagoshima Tenmonkan Medical College (Japan) [view abstract]
Printing display images normally proceed well within the SDR gamut supported by color management systems (CMS), like ICM used in Windows OS. Twilight vision, unfortunately, complicates this process, as it is quite difficult to convert images captured by UHDR systems as their contrast ratios (CR) sometime exceeded approx. 10^6:1; SDR images have CR values of approx. 100:1. A typical example is the reproduction of lunar texture. Quite recently, we resolved this problem by employing just global tone mapping (GTM) (https://doi.org/10.1002/sdtp.17455). Even with the use of GTM, emissive displays like LCD, not projection display or printed matter, are strongly recommended to prevent image quality degradation. This study addresses the reasons why CMS fails to handle twilight vision material well and proposes enhanced GTM for printing emissive display images. The main point is the difference between the perception responses demonstrated in daytime and in twilight, even after the UHDR material is converted into SDR images.
10:30 – 11:00 Coffee Break
Perceptually Based Quality I
Session Chair: Susan Farnand, Rochester Institute of Technology
11:00 - 12:20
Grand Peninsula B
11:00IQSP-246
Evaluating the impact of tinted eyewear on spatial-chromatic contrast sensitivity, Likhitha Nagahanumaiah, Rochester Institute of Technology (US); Dr. Susan Farnand, Rochester Institute of Technology (US); Dr. Christopher Thorstenson, Rochester Institute of Technology (US) [view abstract]
This research explores the effect of various eyewear lenses, designed with varied transmittance properties, on human visual perception. These lenses are developed to enhance contrast between spatial-chromatic patterns like cyan-red and magenta-green compared to lenses with more uniform transmittance. The study evaluates participants' accuracy and response times in identifying contrast patterns, aiming to understand how different eyewear configurations affect these visual metrics. Two experiments were conducted: the first adjusted spatial frequencies to determine visibility thresholds with different eyewear, while the second utilized a 4-alternative forced-choice (4-AFC) method to measure participants' ability to identify contrast patterns. Preliminary results indicate that eyewear with varied transmittance enhances contrast sensitivity more effectively than uniform transmittance lenses, providing insights for optimizing color-enhancing eyewear for improved visual performance.
15:00 – 15:30 Coffee Break
Perceptually Based Quality II
Session Chair: Damon Chandler, Ritsumeikan University
15:30 - 17:30
Grand Peninsula B
16:50IQSP-247
Image quality assessment for natural scene portraits: An industrial application, Daniela Carfora Ventura (France); Gabriel Pacianotto Gouveia, DXOMARK (France); Hoang-Son Nguyen, DXOMARK (France); Jianqiang Sky Zhou, DXOMARK (France); Nicolas Chahine, DXOMARK (France); Sira Ferradans, DXOMARK (France) [view abstract]
Portraits are one of the most common use cases in photography, especially in smartphone photography. However, evaluating portrait quality in real portraits is costly, inconvenient, and difficult to reproduce. We propose a new method to evaluate a large range of detail preservation rendering on real portrait images. Our approach is based on 1) annotating a set of portrait images grouped by semantic content using pairwise comparison 2) taking advantage of the fact that we are focusing on portraits, use cross-content annotations to align the quality scales 3) training a machine learning model on the global quality scale. On top of providing a fine-grained wide range detail preservation quality output, numerical experiments show that the proposed method correlates highly with the perceptual evaluation of an image quality expert.
17:10IQSP-248
Evaluating the influence of eyewear on perception of small color difference in reflective samples, Shuyi Zhao, Rochester Institute of Technology (US); Susan Farnand, Rochester Institute of Technology (US); Christopher Thorstenson, Rochester Institute of Technology (US) [view abstract]
Various types of eyewear are being developed and implemented in a wide range of applications. However, the impact of eyewear on color vision has been insufficiently investigated and remains an area of ongoing exploration. This study aims to investigate the impact of different types of eyewear on color perception, specifically focusing on the ability to discern reflective samples with small color differences. First, perceived color differences for pairs having small color differences are predicted through modeling. These predictions are then tested through psychophysical experiments using the method of scaling to assess differences in observers' color difference judgments when wearing different eyewear. Two sets of stimuli are prepared: one consisting of several pairs of adjacent Munsell samples differing only in hue and the other containing parametric pairs (colors with similar L*a*b* values under a specific light source but different spectral reflectance) generated by modeling 17 pigments using the Kubelka-Munk theory. This research can not only enhance the understanding of how eyewear affects color perception but also provide a robust method for evaluating the performance of eyewear in color discrimination tasks, especially with small color difference stimuli.
WEDNESDAY 5 FEBRUARY 2025
Medical and Video Quality
08:50 - 10:30
Grand Peninsula A
08:50IQSP-249
Keynote: Camera technologies in minimally invasive surgery, Qiang Li, Intuitive Surgical (US) [view abstract]
Minimally invasive surgery (MIS) offers many benefits for patients, including the potential for smaller incisions and faster recovery. Endoscopes used in MIS have significantly benefited from advancements in camera technology. Key developments include miniaturized tip camera with pan and tilt capabilities, higher resolution sensors that provide sharper images and enhance visibility, stereoscopic camera that provides depth perception and spatial orientation for tool manipulation, and simultaneous color and NIR fluorescent imaging that provides surgeons with supplemental anatomic information in standard camera view.
09:30IQSP-251
Optimizing frame selection for improved video quality assessment through embedding similarity, Abderrezzaq Sendjasni, CNRS, Univ. Poitiers, XLIM, UMR 7252 (France); Mohamed-Chaker Larabi, CNRS, Univ. Poitiers, XLIM, UMR 7252 (France); Seif-Eddine Benkabou, LIAS, Univ. Poitiers (France) [view abstract]
No-reference (NR) perceptual video quality assessment (VQA) is a complex, unsolved, and important problem for social and streaming media applications. Efficient and accurate video quality predictors are needed to monitor and guide the processing of billions of shared, often imperfect, videos. Accurate VQA requires a strategic selection of frames that effectively represent overall perceptual quality without compromising computational efficiency. Traditional VQA methods, which often rely on uniform or random sampling, risk neglecting critical temporal features affecting perceived quality. In this work, we propose a similarity-preserving technique for frame selection, designed to maintain frame-level consistency and preserve perceptually relevant features throughout video sequences. The proposed selection preserves structural and semantic similarities within selected frames by analyzing the similarity among frames in the embedding space.To this end, frame embeddings are extracted using ResNet-50, capturing high-level visual representations. The selection algorithm then evaluates similarity among embeddings based on distance metrics and residual learning to identify perceptually important frames. Key frames are ranked and retained according to their contribution to the quality assessment task, with emphasis on those that exhibit significant variations in perceptual features.
09:50IQSP-252
Enhancing automatic visual quality in lossy image compression using advanced HVS metrics and adaptive parameter optimization, Oleksandr Zemliachenko (US) [view abstract]
As the demand for high-resolution digital imagery continues to surge, efficient lossy compression techniques that preserve visual fidelity have become increasingly essential. This paper addresses the challenge of automatically achieving specified visual quality levels in the compression of still images and video frames by leveraging advanced human visual system (HVS)-based metrics. We focus on metrics such as PSNR-HVS-M and MSSIM, which more accurately reflect human perception compared to traditional measures like PSNR. An adaptive, iterative compression framework is proposed, enabling automatic adjustment of compression parameters to meet desired visual quality thresholds with minimal computational overhead. Experimental results on a diverse set of grayscale and color images demonstrate that intelligent initialization of compression parameters significantly reduces the number of iterations required. Additionally, we explore cutting-edge non-iterative compression methods, including those utilizing deep neural networks, evaluating their effectiveness in attaining visual quality objectives. Our comprehensive analysis provides valuable insights into the performance of various compression algorithms, offering practical guidelines for their implementation in the evolving field of electronic imaging.
10:30 – 11:00 Coffee Break
Skin Tone Capture and Image Quality I
Session Chair: Jonathan Phillips, imatest
11:00 - 12:20
Grand Peninsula A
11:00IQSP-253
Keynote: Objective and subjective methods to quantify skin color and pigmentation; recent results and insights., Wim Verkruysse, Philips Medical Systems (Netherlands) [view abstract]
Objective and subjective pigmentation measurement methods (PMM) were evaluated in two studies with 12 and 21 subjects, respectively with balanced enrollment of light and dark subjects. Objective PMM included a colorimeter (Konica Minolta, CM700d), and two melanin meters (Courage &Khazaka, SPA and MX18). Subjective PMM included three skin color scales : Fitzpatrick, Von Luschan and Pantone SkinTone guide. Measurements were on four anatomical locations: forehead, dorsal forearm, ventral forearm and inner upper arm. Precision was much better for objective PMM than for subjective PMM: about 5% and 20% of the entire pigmentation range, respectively. Similarly, operator bias was negligible (< 2%) for the objective PMM while for the subjective PMM it was considerable: > 15%. All four operators matched swatches from the three scales. People with light skin were matched with swatches that were objectively lighter than the skin, and vice versa.
11:40IQSP-254
ISO TC42 process for the selection of representative skin tones, Dietmar Wueller, Image Engineering (Germany); Ken Parulski, aKAP Innovation, LLC. (US); Rita Hofman, Psinex Ventures GmbH (Switzerland) [view abstract]
Members of several working groups within the ISO technical committee 42 (photography) have begun addressing the important topic of providing guidance for which skin tones to use for image quality testing in various photographic applications. Skin tones need to be accurately corrected in digital cameras and properly displayed and printed on various softcopy and hardcopy devices, and the permanence of the printed colors needs to be determined. During the 2023 plenary meeting of TC 42 in Japan, a working group was initiated to develop such guidance and report back during the upcoming 2025 plenary meeting in Berlin.The group has investigated existing skin tone stydies, including Fitzpatrick, Von Luschan, L'Oreal, PERLA, Monk, Pantone ST, Massey NIS, Verkruysse, Holm and Wueller. It is currently working on an ISO Technical Report that documents the work and results in recommendations regarding which colors and approaches to use for ISO related applications.
12:00IQSP-255
Improving image equity: Representing the diversity of skin tones in photographic test charts for digital camera characterization, Megan Borek, Imatest (US) [view abstract]
Accurate representation of diverse skin tones in photography has been a longstanding challenge due to biases toward lighter skin in traditional reference materials used for film and digital photography, such as Kodak's �Shirley� cards and the Fitzpatrick scale. These and other tools, such as the ColorChecker Classic, have offered limited ranges of skin tones and do not capture the full diversity of human skin, including variations in shades, undertones, and exposure behavior. In this study, we evaluate the application of the 10-point Monk Skin Tone Scale, developed by Harvard's Dr. Ellis Monk, to camera testing and characterization using printed skin tone charts. The Monk scale is tested in multiple configurations, including uniform and gradient patches, as well as in color-matched printed faces for testing cameras with facial detection capabilities. We compare the reference CIE LAB values and reflectance spectra of these printed targets to those of other commonly used skin tone references, and to real human skin data. Additionally, we assess the performance of these printed targets in mixed scenes containing both the charts and human subjects, analyzing differences in exposure, white balance, and color saturation. This research identifies limitations and strengths of current printed skin tone scales and charts in representing actual human skin tones, and provides recommendations for improving equitable camera calibration and characterization protocols.
Image Quality and System Performance Posters (with lunch)
12:20 - 14:00
The Grove
12:20IQSP-256
Exploiting skin melanin network for skin pigmentation classification, Wim Verkruysse, Philips Medical Systems (Netherlands); Karl van Bree, Philips Medical Systems (Netherlands); Shakith Fernando, Philips Medical Systems (Netherlands) [view abstract]
Recently skin pigmentation concerns have been raised for accuracy across different skin tone types. With current skin tone classification methods, robust skin type and melanin index detection is challenging, as it requires calibration against many factors including temperature, waveguide, ambient light, color reference, and synthetic skin tone type reference. The proposed system differentiates and quantifies skin type and melanin index by exploiting the variance in skin structures and skin pigmentation network across skin types. Our result with a small study shows skin structures pattern is a robust, color independent method for skin tone classification.
15:00 – 15:30 Coffee Break
Skin Tone Capture and Image Quality II
Session Chair: Bo Ding, Snap Inc.
15:30 - 16:30
Grand Peninsula A
15:30IQSP-257
Color and spectral matching in imaging performance evaluation, Peter Burns, Burns Digital Imaging (US); Don Williams, Image Science Associates (US) [view abstract]
Evaluation of camera or scanner image capture usually includes the selection of reference color test objects. For image capture, since the optical characteristics are known, the intended camera color-encoded image data are compared with the corresponding idea values. However, for critical scene content, the color patches may not sample the encoded color space efficiently. In addition the colorimetric sampling by the target, its spectral-reflectance characteristics may differ from those of important scene or object elements. We address the selection of color test patches for both colorimetric and spectral criteria. Results are shown for the imaging of human skin tones and include ananalysis of likely variability.
15:50IQSP-258
Unveiling the role of color in skin segmentation: Analysis of augmentation techniques and color spaces, Mekides Assefa Abebe; Soroush Shahbaznejad; Nima Rabbanifar; Elena Fedorovskaya, [view abstract]
Recent advancements in artificial intelligence (AI) have significantly impacted many color imaging applications, including skin segmentation and enhancements. While state-of-the-art methods emphasize geometric augmentations to enhance model performance, the role of color-based augmentations and color spaces in improving skin segmentation accuracy remains underexplored. This study addresses this gap by systematically evaluating the impact of various color-based augmentations and color spaces on CNN-based skin segmentation models. We investigate the effects of color transformations�including brightness, contrast, saturation, and gamma adjust- ments�and explore color space conversions to YCbCr and CIELab. An existing semantic segmentation model was trained using these color augmentations on a custom dataset of 500 annotated skin images, encompassing diverse skin tones and lighting conditions. Our findings reveal that color-based augmentations significantly improve the model's ability to accurately segment skin regions, especially at the presence of over- and under-exposure problems. Additionally, color space conversions to YCbCr and CIELab offer performance comparable to the traditional RGB space, when coupled with color augmentation, suggesting that they have the potential to enhance accuracy and robustness when further supported by augmentation techniques. However, significant performance discrepancies based on skin tones were identified, highlighting challenges in achieving consistent segmentation across all skin types. Overall, this research bridges the gap left by current state-of-the-art methods, providing important insights to advance the performance and inclusivity of skin segmentation models.
16:10IQSP-259
Video health monitoring; possibilities and challenges in the context of tissue optics and skin color., Wim Verkruysse, Philips Medical Systems (Netherlands) [view abstract]
Cameras are increasingly used as measurement devices in health care. Vital signs such as heart rate and respiration rate can be measured and even more challenging ones are being explored such as arterial oxygen saturation (aka SpO2) and core-body-temperature. Current and future possibilities and remaining challenges will be presented from a technical perspective (camera technology) and a tissue optics and/or physiological perspective.
Display Image Quality Evaluation
Session Chair: Bo Ding, Snap, Inc.
16:30 - 17:10
Grand Peninsula A
16:30IQSP-260
Evaluation of lightness preference of images on OLED and QLED displays, Eddie Pei , Rochester Institute of Technology (US); Hosub Lee, Samsung Research America (US); Elena Fedorovskaya, Rochester Institute of Technology (US); Susan Farnand, Rochester Institute of Technology [view abstract]
This research investigates the influence of lightness, lightness contrast, and display types on image preference and perception. Building on previous studies that highlight the importance color attributes in shaping image quality, we explore these attributes using CIECAM16 color space calculations. Four experiments were conducted on OLED and QLED displays, in which participants selected their preferred image and rated its quality relative to a reference. Early findings suggest that lightness and display type significantly impact viewer preference. The study aims to enhance understanding of image quality factors across different displays and cultural contexts, with potential applications in optimizing color reproduction.
16:50IQSP-261
On a novel technique to quantify local contrast preservation in HDR scenes, Pu Zheng, DXOMARK (France); Francois-Xavier Thomas, DXOMARK (France); Claudio Greco, DXOMARK (France); Frederic Guichard, DXOMARK (France) [view abstract]
This work provides a novel glass-to-glass metric of local contrast, useful in the context of image quality evaluation of HDR content. This metric, called Local-Contrast Gain (LCG), uses the opto-optical transfer function (OOTF) of the imaging system and its first derivative to compute the incremental ratio between contrast in the scene and contrast on the display. In order to be perceptually meaningful, we chose Weber's definition of contrast. To know the OOTF in analytical form and to make the measurement robust to the uncertainty of measurements of the ground truth, we rely on a model that we propose and that expands upon our previously published work. We provide experimental validation of our metric on a variety of target charts, both reflective and transmissive, both in isolation and within complex setups spanning more than seven EVs.