Hyunsuk Ko
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Hyunsuk Ko has a total of 60 co-authors. Here are the top co-authors by number of shared publications:
- Hui Yong Kim
School of Computing, Kyung Hee University, Yongin, Republic of Korea
5 shared publications - Alan Conrad Bovik
The University of Texas at Austin
3 shared publications - Dae Yeol Lee
Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin, TX, USA
3 shared publications Jongho Kim
Media Coding Research Section, Daejeon, ETRI, South Korea
3 shared publicationsJuhyun Lee
3 shared publications- Yongrok Kim
Department of Electrical and Electronic Engineering, Hanyang University, Ansan, South Korea
3 shared publications
Latest Publications
- 31 July 2026
- Published by The Korean Institute of Broadcast and Media Engineers in Journal of Broadcast Engineering
- Vol. 31 (4) , 782-792
Enhancing Time–Frequency Characteristics in Passive Sonar Data Generation via Self-Conditioned Diffusion
- 31 July 2026
- Published by The Korean Institute of Broadcast and Media Engineers in Journal of Broadcast Engineering
- Vol. 31 (4) , 782-792
Enhancing Time–Frequency Characteristics in Passive Sonar Data Generation via Self-Conditioned Diffusion
- Open Access
- 23 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 22 (6) , 1-23
Adaptive Neural In-Loop Filtering via Boundary-Aware Skipping and Time-Distortion Optimization
Neural network-based in-loop filters (NNILFs) have recently been integrated into emerging video coding frameworks to enhance reconstruction quality beyond the capabilities of traditional signal processing filters. However, the high computational complexity of these filters significantly increases encoding and decoding time, hindering their real-time deployment. In this article, we propose an efficient neural in-loop filter control framework that improves decoding efficiency while preserving coding performance. Our approach introduces boundary strength-aware control to selectively skip NNILF operations on visually less critical regions. In addition, a time-distortion optimization (TDO) strategy is presented to adaptively manage the tradeoff between visual distortion and inference time during encoding. Experimental results on the NNVC reference software show that the proposed methods reduce decoding time by up to 23% and 21% for the random access (RA) and low-delay B (LDB) configurations, respectively. Apart from a minimal luma loss only in RA, chroma gains were also achieved, with Y: 0.10% U: −0.01% V: −0.14% for RA and Y: 0.00% U: −1.34% V: −0.94% for LDB. These results show that the proposed control framework offers a practical and effective solution for integrating NN-based tools into future video coding systems.
- 23 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 22 (6) , 1-23
Open AccessAdaptive Neural In-Loop Filtering via Boundary-Aware Skipping and Time-Distortion Optimization
Neural network-based in-loop filters (NNILFs) have recently been integrated into emerging video coding frameworks to enhance reconstruction quality beyond the capabilities of traditional signal processing filters. However, the high computational complexity of these filters significantly increases encoding and decoding time, hindering their real-time deployment. In this article, we propose an efficient neural in-loop filter control framework that improves decoding efficiency while preserving coding performance. Our approach introduces boundary strength-aware control to selectively skip NNILF operations on visually less critical regions. In addition, a time-distortion optimization (TDO) strategy is presented to adaptively manage the tradeoff between visual distortion and inference time during encoding. Experimental results on the NNVC reference software show that the proposed methods reduce decoding time by up to 23% and 21% for the random access (RA) and low-delay B (LDB) configurations, respectively. Apart from a minimal luma loss only in RA, chroma gains were also achieved, with Y: 0.10% U: −0.01% V: −0.14% for RA and Y: 0.00% U: −1.34% V: −0.94% for LDB. These results show that the proposed control framework offers a practical and effective solution for integrating NN-based tools into future video coding systems.
- Open Access
Low-Rank Curvature for Zeroth-Order Optimization in LLM Fine-Tuning
We introduce LOREN, a curvature-aware zeroth-order (ZO) optimization method for fine-tuning large language models (LLMs). Existing ZO methods, which estimate gradients via finite differences using random perturbations, often suffer from high variance and suboptimal search directions. Our approach addresses these challenges by: (i) reformulating the problem of gradient preconditioning as that of adaptively estimating an anisotropic perturbation distribution for gradient estimation, (ii) capturing curvature through a low-rank block diagonal preconditioner using the framework of natural evolution strategies, and (iii) applying a REINFORCE leave-one-out (RLOO) gradient estimator to reduce variance. Experiments on standard LLM benchmarks show that our method outperforms state-of-the-art ZO methods by achieving higher accuracy and faster convergence, while cutting peak memory usage by up to 27.3% compared with MeZO-Adam.
Open AccessLow-Rank Curvature for Zeroth-Order Optimization in LLM Fine-Tuning
We introduce LOREN, a curvature-aware zeroth-order (ZO) optimization method for fine-tuning large language models (LLMs). Existing ZO methods, which estimate gradients via finite differences using random perturbations, often suffer from high variance and suboptimal search directions. Our approach addresses these challenges by: (i) reformulating the problem of gradient preconditioning as that of adaptively estimating an anisotropic perturbation distribution for gradient estimation, (ii) capturing curvature through a low-rank block diagonal preconditioner using the framework of natural evolution strategies, and (iii) applying a REINFORCE leave-one-out (RLOO) gradient estimator to reduce variance. Experiments on standard LLM benchmarks show that our method outperforms state-of-the-art ZO methods by achieving higher accuracy and faster convergence, while cutting peak memory usage by up to 27.3% compared with MeZO-Adam.
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 8930-8938
Phase Distribution Matters: On the Importance of Phase Distribution Alignment (PDA) in Holographic Applications
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 8930-8938
Open AccessPhase Distribution Matters: On the Importance of Phase Distribution Alignment (PDA) in Holographic Applications
- Open Access
- 3 October 2025
- Published by Elsevier in Ocean Engineering
- Vol. 342
Reinforcement learning-based automated target motion analysis in underwater environments
- 3 October 2025
- Published by Elsevier in Ocean Engineering
- Vol. 342
Open AccessReinforcement learning-based automated target motion analysis in underwater environments
- Open Access
- 23 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 168087-168102
Subjective and Objective Quality Evaluation of Super-Resolution Enhanced Broadcast Images on a Novel SR-IQA Dataset
Super-Resolution (SR) is essential for displaying low-quality broadcast content on high-resolution screens. Recently, SR methods have been developed that not only increase resolution while preserving the original image information but also enhance the perceived quality. However, evaluating the quality of SR images generated from low-quality sources, such as SR-enhanced broadcast content, is challenging due to the need to consider both distortions and improvements. Additionally, assessing SR image quality without original high-quality sources presents another significant challenge. Unfortunately, there has been a dearth of research specifically addressing the Image Quality Assessment (IQA) of SR images under these conditions. In this work, we introduce a new IQA dataset for SR broadcast images in both 2K and 4K resolutions. We conducted a subjective quality evaluation to obtain Mean Opinion Score (MOS) for these SR images and performed a comprehensive human study to identify key factors influencing perceived quality. Finally, we evaluated the performance of existing IQA metrics on our dataset. This study reveals the limitations of current metrics, highlighting the need for a more robust IQA metric that better correlates with the perceived quality of SR images. The proposed dataset and the subjective evaluation platform are publicly available at https://sites.google.com/hanyang.ac.kr/ivml/datasets/sreb
- 23 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 168087-168102
Open AccessSubjective and Objective Quality Evaluation of Super-Resolution Enhanced Broadcast Images on a Novel SR-IQA Dataset
Super-Resolution (SR) is essential for displaying low-quality broadcast content on high-resolution screens. Recently, SR methods have been developed that not only increase resolution while preserving the original image information but also enhance the perceived quality. However, evaluating the quality of SR images generated from low-quality sources, such as SR-enhanced broadcast content, is challenging due to the need to consider both distortions and improvements. Additionally, assessing SR image quality without original high-quality sources presents another significant challenge. Unfortunately, there has been a dearth of research specifically addressing the Image Quality Assessment (IQA) of SR images under these conditions. In this work, we introduce a new IQA dataset for SR broadcast images in both 2K and 4K resolutions. We conducted a subjective quality evaluation to obtain Mean Opinion Score (MOS) for these SR images and performed a comprehensive human study to identify key factors influencing perceived quality. Finally, we evaluated the performance of existing IQA metrics on our dataset. This study reveals the limitations of current metrics, highlighting the need for a more robust IQA metric that better correlates with the perceived quality of SR images. The proposed dataset and the subjective evaluation platform are publicly available at https://sites.google.com/hanyang.ac.kr/ivml/datasets/sreb
- 17 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (3) , 5073-5086
Progressive Multiview Instance Matching: Occlusion-Robust Approach With Initial Segmentation Enhancement
The growing demand for immersive media has led to active research on multi-view images. However, challenges such as occlusion and the lack of publicly available multi-view datasets hinder progress in this field. To address these issues, we introduce a novel dataset, SMIIM (Synthetic Multi-view Images for Instance Matching), designed for multi-view instance matching. We also propose a progressive matching algorithm, a three-stage process that effectively handles occlusion. Finally, we improve the image segmentation results by refining the masks generated by existing networks with our matching results. Compared to existing instance matching algorithms, our method not only provides faster processing time but also significantly enhances performance, achieving an average ID matching accuracy of 97.5% and a mean Intersection over Union (mIoU) improvement of 19.8% not only on our dataset but also on standardized datasets and real-world datasets. Multi-object matching in multi-view environments is rare, making our research a valuable contribution to this field. The SMIIM dataset will be released to facilitate further research and development in the field of multi-view image processing.
- 17 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (3) , 5073-5086
Progressive Multiview Instance Matching: Occlusion-Robust Approach With Initial Segmentation Enhancement
The growing demand for immersive media has led to active research on multi-view images. However, challenges such as occlusion and the lack of publicly available multi-view datasets hinder progress in this field. To address these issues, we introduce a novel dataset, SMIIM (Synthetic Multi-view Images for Instance Matching), designed for multi-view instance matching. We also propose a progressive matching algorithm, a three-stage process that effectively handles occlusion. Finally, we improve the image segmentation results by refining the masks generated by existing networks with our matching results. Compared to existing instance matching algorithms, our method not only provides faster processing time but also significantly enhances performance, achieving an average ID matching accuracy of 97.5% and a mean Intersection over Union (mIoU) improvement of 19.8% not only on our dataset but also on standardized datasets and real-world datasets. Multi-object matching in multi-view environments is rare, making our research a valuable contribution to this field. The SMIIM dataset will be released to facilitate further research and development in the field of multi-view image processing.
- Open Access
Automated cell tracking using 3D nnUnet and Light Sheet Microscopy to quantify regional deformation in zebrafish
Light Sheet Microscopy (LSM) in conjunction with embryonic zebrafish, is rapidly advancing three-dimensional, in vivo characterization of myocardial contractility. Preclinical cardiac deformation imaging is predominantly restricted to a low-order dimensionality image space (2i) or suffers from poor reproducibility. In this regard, LSM has enabled high throughput, non-invasive 4i (3d+time) characterization of dynamic organogenesis within the transparent zebrafish model. More importantly, LSM offers cellular resolution across large imaging Field-of-Views at millisecond camera frame rates, enabling single cell localization for global cardiac deformation analysis. However, manual labeling of cells within multilayered tissue is a time-consuming task and requires substantial expertise. In this study, we applied the 3i nnU-Net with Linear Assignment Problem (LAP) framework for automated segmentation and tracking of myocardial cells. Using binarized labels from the neural network, we quantified myocardial deformation of the zebrafish ventricle across 4-6 days post fertilization (dpf). Our study offers tremendous promise for developing highly scalable and disease-specific biomechanical quantification of myocardial microstructures.
Open AccessAutomated cell tracking using 3D nnUnet and Light Sheet Microscopy to quantify regional deformation in zebrafish
Light Sheet Microscopy (LSM) in conjunction with embryonic zebrafish, is rapidly advancing three-dimensional, in vivo characterization of myocardial contractility. Preclinical cardiac deformation imaging is predominantly restricted to a low-order dimensionality image space (2i) or suffers from poor reproducibility. In this regard, LSM has enabled high throughput, non-invasive 4i (3d+time) characterization of dynamic organogenesis within the transparent zebrafish model. More importantly, LSM offers cellular resolution across large imaging Field-of-Views at millisecond camera frame rates, enabling single cell localization for global cardiac deformation analysis. However, manual labeling of cells within multilayered tissue is a time-consuming task and requires substantial expertise. In this study, we applied the 3i nnU-Net with Linear Assignment Problem (LAP) framework for automated segmentation and tracking of myocardial cells. Using binarized labels from the neural network, we quantified myocardial deformation of the zebrafish ventricle across 4-6 days post fertilization (dpf). Our study offers tremendous promise for developing highly scalable and disease-specific biomechanical quantification of myocardial microstructures.
- 6 February 2024
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 133
Facial image deblurring network for robust illuminance adaptation and key structure restoration
- 6 February 2024
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 133
Facial image deblurring network for robust illuminance adaptation and key structure restoration
- 1 November 2023
- Published by Optica Publishing Group in Optics Express
- Vol. 31 (23) , 38854-38877
Performance analysis of versatile video coding for encoding phase-only hologram videos
In recent decades, holographic technology has made significant progress with the development of novel hologram generation methods and three-dimensional rendering devices. Nevertheless, the substantial size of holographic data presents a significant challenge to its practical applications and thus necessitates the implementation of an efficient coding solution. In this study, we evaluate the efficiency of various coding tools within the state-of-the-art video coding standard, Versatile Video Coding, for encoding video of computer-generated phase-only hologram. Specifically, we examine the coding performance of transform/in-loop filter/screen-content coding tools. Through extensive encoding experiments and various statistical analyses, we investigated the limitations of existing standard codecs that do not account for the unique signal characteristics of phase-only holograms (POHs). The effects of coding artifacts on the visual quality of numerical reconstructions rendered from compressed POHs are also analyzed in detail. These comprehensive performance evaluations will provide valuable insights for developing efficient coding strategies for POH videos.
- 1 November 2023
- Published by Optica Publishing Group in Optics Express
- Vol. 31 (23) , 38854-38877
Performance analysis of versatile video coding for encoding phase-only hologram videos
In recent decades, holographic technology has made significant progress with the development of novel hologram generation methods and three-dimensional rendering devices. Nevertheless, the substantial size of holographic data presents a significant challenge to its practical applications and thus necessitates the implementation of an efficient coding solution. In this study, we evaluate the efficiency of various coding tools within the state-of-the-art video coding standard, Versatile Video Coding, for encoding video of computer-generated phase-only hologram. Specifically, we examine the coding performance of transform/in-loop filter/screen-content coding tools. Through extensive encoding experiments and various statistical analyses, we investigated the limitations of existing standard codecs that do not account for the unique signal characteristics of phase-only holograms (POHs). The effects of coding artifacts on the visual quality of numerical reconstructions rendered from compressed POHs are also analyzed in detail. These comprehensive performance evaluations will provide valuable insights for developing efficient coding strategies for POH videos.