Suk-Ju Kang’s research while affiliated with Sogang University and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (83)


Performance comparison with state-of-the-art perceptual-driven SISR methods on commonly used benchmark datasets for scaling factor 4. The proposed method demonstrates superior perceptual SR performance, as verified by LPIPS, and comparable PSNR and SSIM values, simultaneously.
Performance comparison of methods with different components for scaling factor 4. The best performance is highlighted in red, while the second-best one is highlighted in blue (PSNR ↑/SSIM ↑/LPIPS ↓).
Frequency Domain-Based Super Resolution Using Two-Dimensional Structure Consistency for Ultra-High-Resolution Display
  • Article
  • Full-text available

October 2024

·

8 Reads

Journal of Imaging

Yu Lim Seo

·

Suk-Ju Kang

·

Yeon-Kug Moon

Recent advancements in the field of super resolution (SR) have seen the adoption of generative adversarial networks (GANs) for realistic images. In this case, when performing with low-resolution (LR) images, several challenges arise due to the loss of high-frequency details from high-resolution (HR) images, potentially leading to unwanted distortions in the generated SR images. Our paper presents a novel solution by using two-dimensional structure consistency (TSC) for image analysis. The TSC serves as a mask, enabling adaptive analysis based on the unique frequency characteristics of different image regions. Furthermore, a mutual loss mechanism, which dynamically adjusts the training process based on the results filtered by the TSC-based mask, is introduced. Additionally, the TSC loss is proposed to enhance our model capacity to generate precise TSC in high-frequency regions. As a result, our method effectively reduces distortions in high-frequency areas while preserving clarity in regions containing low-frequency components. Our method outperforms other SR techniques, demonstrating superior results in both qualitative and quantitative evaluations. Quantitative measurements, including PSNR, SSIM, and the perceptual metric LPIPS, show comparable PSNR and SSIM values, while the perceptual SR quality is notably improved according to the LPIPS metric.

Download






Improving Gaze Tracking in Large Screens With Symmetric Gaze Angle Amplification and Optimization Technique

January 2023

·

104 Reads

IEEE Access

Many gaze tracking applications focus on use in personal devices such as mobile phones and PCs. However, gaze tracking in large screens poses challenges because with an increase in screen size, gaze tracking in the edge region decreases owing to the restricted range of human eye movement. In addition, as large screens are often exposed to the public, anyone can use the gaze tracking module. This makes it difficult to apply personalized calibration as in personal devices. To acquire accurate gaze in the edge region, we propose a novel approach—symmetric angle amplifying function—for the gaze angle, which amplifies angles when a user is looking at the edge area of the large screen. Our function is designed particularly for the case where the screen is divided into grid-shaped regions. Furthermore, for the better user experience, we optimize neural networks using the network-optimization framework and also propose a center gravity function that pulls gaze coordinates presented on the screen to the predefined center of the region to compensate for the person-wise difference in movement of the human eyes. Experimental results revealed the superiority of the proposed methods over the baseline and different types of fitting functions. The gaze tracking module serves as a part of an aggregated system and is implemented for use in autonomous vehicles.




Heatmap Assisted Accuracy Score Evaluation Method for Machine-Centric Explainable Deep Neural Networks

January 2022

·

263 Reads

·

5 Citations

IEEE Access

Junhee Lee

·

Hyeonseong Cho

·

Yun Jang Pyun

·

[...]

·

There have existed many studies about the explainable artificial intelligence (XAI) that explains the logic behind the complex deep neural network called a black box. At the same time, researchers have tried to evaluate the explainability performance of various XAIs. However, most previous evaluation methods are human-centric, that is, subjective, where they rely on how much the results of explanation are similar to what people’s decision is based on rather than what features actually affect the decision in the model. Their XAI selections are also dependent of datasets. Furthermore, they are focusing only on the output variation of a target class. On the other hand, this paper proposes a robust heatmap assisted accuracy score (HAAS) scheme over datasets that helps selecting machine-centric explanation algorithms to show what actually leads to the decision of a given classification network. The proposed method modifies the input image with the heatmap scores obtained by a given explanation algorithm and then puts the resultant heatmap assisted (HA) images into the network to estimate the accuracy change. The resultant metric ( HAAS ) is computed as a ratio of accuracies of the given network over HA and original images. The proposed evaluation scheme is verified in the image classification models of LeNet-5 for MNIST and VGG-16 for CIFAR-10, STL-10, and ILSVRC2012 over totally 11 XAI algorithms of saliency map, deconvolution, and 9 layer-wise relevance propagation (LRP) configurations. Consequently, for LRP1 and LRP3, MINST showed largest HAAS values of 1.0088 and 1.0079, CIFAR-10 achieved 1.1160 and 1.1254, STL-10 had 1.0906 and 1.0918, and ILSVRC2012 got 1.3207 and 1.3469. While LRP1 consists of ϵ\epsilon -rules for input, convolutional, and fully-connected layers, LRP3 adopts a bounded-rule for an input layer and the same ϵ\epsilon -rules for other layers as LRP1. The consistency of evaluation results of HAAS and AOPC has been compared by means of Kullback-Leibler divergence, ensuring that HAAS is the more robust evaluation method than AOPC independently of datasets since HAAS has much lower average divergence of 0.0251 than AOPC of 0.3048. In addition, the validity of the proposed HAAS scheme is further investigated through the inverted HA test that employs inverted HA images made up with inverted heatmap scores and estimates the accuracy degradation caused by applying them to the network. The XAI algorithms with largest HAAS results experience biggest accuracy degradation in the inverted HA test.


Citations (50)


... Current anomaly detection focuses on several directions. Unsupervised methods include image reconstruction [21][22][23], using autoencoders to spot anomalies by comparing reconstructed and original images. Embedding-based methods learn the normal sample distribution, detecting anomalies by comparing new sample embeddings. ...

Reference:

CutPaste-ROI: An Industrial Anomaly Data Detection Method based on Self-supervised Learning
Semi-supervised Anomaly Detection with Reinforcement Learning
  • Citing Conference Paper
  • July 2022

... Metrics in this category are based on observing the change in the model's output when permuting the input based on a given attribution map. Several metrics aim to compute the correlation between the model's original probability output and the perturbed one [11] [52] or the correlation between probability drops and attribution scores on various points [3] [11], while others observe the change in model performance after perturbing the input [9] [39] [30]. ...

Heatmap Assisted Accuracy Score Evaluation Method for Machine-Centric Explainable Deep Neural Networks

IEEE Access

... Among these GANbased anomaly detection methods, the ways in which the output of the discriminator is used to distinguish the two types of data can also be roughly classified into two categories. One is used by models such as [31], which directly uses the discriminator to classify the input data, and the final output is the label. The other is used by models such as [2,35], which propose the concept of anomaly score. ...

Attention Map-guided Two-stage Anomaly Detection using Hard Augmentation

... Compared with image-based methods, hand skeletonbased methods mitigate the interference of complex backgrounds and various lighting conditions and are more robust while reducing computational costs and enabling realtime gesture interaction on mobile devices. Moreover, due to the development of low-cost depth cameras (i.e., Microsoft Kinect [14]or Intel RealSense [15]) and hand pose estimation algorithms [16]- [18], correct hand skeleton sequences can be easily obtained. These advantages have promoted skeletonbased gesture recognition in recent studies. ...

3D Hand Pose Estimation via Graph-Based Reasoning

IEEE Access

... One key advantage of using Web crawlers compared to prepared datasets is the ability to obtain timely real-world data, allowing ML models to better generalize to unseen scenarios. Since ML requires many examples to generalize effectively and the Web is a vast source of data, some works have used data from the Web to train ML algorithms (Koloveas et al., 2021;Lee & Kang, 2019;Meesad, 2021;Yanai, 2003a). ...

Web Scraping Crawling-based Automatic Data Augmentation for Deep Neural Networks-based Vehicle Classifications
  • Citing Conference Paper
  • January 2019

... In the research of S. Kul, S. Eken and A. Sayar [23], they conducted a vehicle tracking and classification system using vehicle traffic information with the help of video surveillance system, in which they had problems in classification due to different sizes and shapes of vehicles, while in the research of S. Y. Jo, N. Ahn, Y. Lee and S. J. Kang [24], it proposed a transfer learning based vehicle classification using a limited scale dataset, in which it was found that the range of colour layers helps to train the programming and affects the learning performance. Kang, proposed a transfer learning based vehicle classification using a limited scale dataset in which it was found that the range of colour layers helps to train programming and affects learning performance, while in this research it was possible to classify vehicles and calculate the speed at which they move using a dynamic behaviour of a particle at constant speed using videos obtained from the Mavic Air drone. ...

Transfer Learning-based Vehicle Classification
  • Citing Conference Paper
  • November 2018

... However, in order to overcome the problem that the traditional chessboard may cause uneven illumination of the calibration target under strong annular illumination, resulting in the failure of ordinary chessboard detection. Reference [55] proposed a calibration method based on the Charuco plate, which can solve the problems caused by the traditional calibration plate and can also be used to detect other saddle points. To date, there are many external calibration tools for LiDAR and cameras. ...

Charuco Board-Based Omnidirectional Camera Calibration Method

Electronics

... Marzougui et al. (2020) employed PPHT to detect lane markings, which achieved a correct detection rate (CDR) of 93.8% and cost 21.5 ms on a 640 × 480 pixels image with an Intel Core i7-2630QM CPU. Chang and Kang (2018) applied the PPHT to detect lane markings by removing lanes that didn't reach requirements of set angle or distance conditions. Chen et al. (2020) utilized PPHT to detect welding flame and achieved CDR of 90.2%. ...

Real-time Vehicle Detection and Tracking Algorithm for Forward Vehicle Collision Warning
  • Citing Article
  • October 2018

JSTS Journal of Semiconductor Technology and Science

... However, at the time of this writing, there is limited work investigating the extent to which network latency affects VR games. While studies addressing latency in VR exist [4,20,24,36,44,55], they are often focused around investigating motion-to-photon (MTP) latency, which is more of an issue for VR cloud-gaming, as opposed to locally-rendered games. ...

Time Sequential Motion-to-Photon Latency Measurement System for Virtual Reality Head-Mounted Displays

Electronics

... High dynamic range (HDR) images have wide application prospects in the fields of medical imaging, aerospace remote sensing, and cross-media color reproduction because of their higher dynamic range, wider color gamut, and richer details [1][2][3][4][5]. The dynamic range of HDR images must be mapped to the range of display devices, which is called tone mapping [2]. ...

Perceptual Brightness-based Inverse Tone Mapping for High Dynamic Range Imaging
  • Citing Article
  • July 2018

Displays