http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Robust Elastic-Net Subspace Representation
Kim, Eunwoo,Lee, Minsik,Oh, Songhwai IEEE 2016 IEEE TRANSACTIONS ON IMAGE PROCESSING - Vol.25 No.9
<P>Recently, finding the low-dimensional structure of high-dimensional data has gained much attention. Given a set of data points sampled from a single subspace or a union of subspaces, the goal is to learn or capture the underlying subspace structure of the data set. In this paper, we propose elastic-net subspace representation, a new subspace representation framework using elastic-net regularization of singular values. Due to the strong convexity enforced by elastic-net, the proposed method is more stable and robust in the presence of heavy corruptions compared with existing lasso-type rank minimization approaches. For discovering a single low-dimensional subspace, we propose a computationally efficient low-rank factorization algorithm, called FactEN, using a property of the nuclear norm and the augmented Lagrangian method. Then, ClustEN is proposed to handle the general case, in which the data samples are drawn from a union of multiple subspaces, for joint subspace clustering and estimation. The proposed algorithms are applied to a number of subspace representation problems to evaluate the robustness and efficiency under various noisy conditions, and experimental results show the benefits of the proposed method compared with existing methods.</P>
High-resolution Classifier Ensemble for Defect Inspection of Display Panels
Eunwoo Kim(김은우),Jaewon Kim(김재완) 제어로봇시스템학회 2021 제어·로봇·시스템학회 논문지 Vol.27 No.8
DL(Deep Learning) has been applied to various tasks, and image classification is one of the most active areas where DL is applied. While the majority of DL approaches have focused on achieving improvements in classification accuracy by innovating a network structure with standard datasets in low resolution, detailed features in a high-resolution dataset are crucial to enhance classification accuracy, especially for practical datasets in the industry. We proposed a DL classifier structure that fully utilizes high-definition information for an OLED (Organic Light-Emitting Diode) panel inspection, and, for verification, we performed the task of classifying the authenticity of the actual OLED panel defects. The authenticity inspection of panel defects requires a precise analysis of high-resolution information. We confirmed that the application of the proposed classifier structure improved the performance of the classification. The proposed method includes an object detection method optimized for panel inspection and displays stable performance by utilizing an ensemble structure. The proposed method has been applied and is being used in actual production lines.
Fabrication of Microgratings and their IR Diffraction Spectra
Kim, In Cheol,Choi, Eunwoo,Kim, Seong Kyu,Kang, Young Il,Kim, Taeseong,Bae, Hyo-Wook,Park, Do-Hyun Korean Chemical Society 2014 Bulletin of the Korean Chemical Society Vol.35 No.3
Microgratings whose diffracted field at a fixed angle generate IR spectra of $SF_6$ or $NH_3$ were fabricated by MEMS techniques for the purpose of IR correlation spectroscopy. Each micrograting was composed of 1441 reflecting lines in the area of $19.2{\times}19.2mm^2$. The depth profile of the line elements was determined with a gradient searching method that was described in our previous publication (J. Mod. Opt. 2013, 60, 324-330), and was discretized into 16 levels between 0 and $6.90{\mu}m$. The diffraction field from a given depth profile was calculated with Fraunhofer equation. The fabricated microgratings showed errors in the depth and the width within acceptable ranges. As the result, the diffracted IR spectrum of each micrograting matched well with its target reference spectrum within spectral resolution of our optical setup.
New fMRI analysis method for multiple stimuli using reference estimation
Kim, Eunwoo,Han, Yeji,Park, Hyunwook Wiley Subscription Services, Inc., A Wiley Company 2011 INTERNATIONAL JOURNAL OF IMAGING SYSTEMS AND TECHN Vol.21 No.4
<P><B>Abstract</B></P><P>The stimulation paradigms of a functional MRI (fMRI) usually consist of one or more stimulations and a resting state in the block‐based and event‐related designs. To localize the activation areas in the human brain, each voxel is statistically analyzed using the fMRI data measured with the stimulation. The conventional method can be inefficient for experiments with multiple stimuli because of measuring the resting‐state signals repeatedly, causing redundancy in the scanning process. Although the phase mapping method can be applied to reduce the redundant resting‐state measurements, there are still limitations in the detection of regions activated by multiple stimuli and the periodic sequence of the multiple stimuli. In this article, a new fMRI data analysis method is presented that enables the detection of functional activations without the resting‐state images. This method estimates the reference signal from the signals acquired during multiple stimuli, and a random sequence and various durations of the multiple stimuli can be applied. Therefore, it can be used in the event‐related design as well as the block‐based design. The results of simulation and fMRI experiments show that the proposed method can correctly detect the activation regions of multiple stimuli, even for overlap regions, and can reduce the imaging time by skipping the resting‐state imaging. © 2011 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 21, 315–322, 2011</P>
Eunwoo Kim,Minsik Lee,Chong-Ho Choi,Nojun Kwak,Songhwai Oh IEEE 2015 IEEE transactions on neural networks and learning Vol.26 No.2
<P>Low-rank matrix approximation plays an important role in the area of computer vision and image processing. Most of the conventional low-rank matrix approximation methods are based on the l(2)-norm (Frobenius norm) with principal component analysis (PCA) being the most popular among them. However, this can give a poor approximation for data contaminated by outliers (including missing data), because the l(2)-norm exaggerates the negative effect of outliers. Recently, to overcome this problem, various methods based on the l(1)-norm, such as robust PCA methods, have been proposed for low-rank matrix approximation. Despite the robustness of the methods, they require heavy computational effort and substantial memory for high-dimensional data, which is impractical for real-world problems. In this paper, we propose two efficient low-rank factorization methods based on the l(1)-norm that find proper projection and coefficient matrices using the alternating rectified gradient method. The proposed methods are applied to a number of low-rank matrix approximation problems to demonstrate their efficiency and robustness. The experimental results show that our proposals are efficient in both execution time and reconstruction performance unlike other state-of-the-art methods.</P>
DVB-S2 수신기를 위한 신호 대 잡음비 추정 하드웨어 설계
박은우(Eunwoo Park),이재웅(Jaeung Yi),김수성(Sooseong Kim),임채용(Chaeyong Im),여성문(Sungmoon Yeo),김수영(Sooyoung Kim) 대한전자공학회 2007 대한전자공학회 학술대회 Vol.2007 No.7
This paper presents an efficient and simple hardware design of signal to noise ratio (SNR) esto,atpr fpr DVB-S2 system. The estimator investigates the distribution of the received symbols by simply using two comparators and a counter, and calculates the address of an LUT where the corresponding SNR value is located. In this paper, we demonstrate the functional and timing simulation results of the FPGA implementation of proposed structure.
능동 롤 제어에 의한 차량 운동 성능 향상에 대한 연구
라은우(Eunwoo Na),김경호(Kyungho Kim),이상호(Sangho Lee),최장한(Janghan Choi),김충(Choong Kim),박준홍(Joonhong Park) 한국자동차공학회 2009 한국자동차공학회 부문종합 학술대회 Vol.2009 No.4
ARC (Active Roll Control) system operates the actuators either in the middle of the stabilizer bar (Rotary type) or at one end of stabilizer bar (Linear type) in order to reduce the roll angle. Here we apply ARC technology to a sedan and simulate it, in which the stabilizer bar stiffness changes along with the actuator stroke related to the lateral acceleration. Besides the roll angle reduction, it can improve the yaw rate behavior, too. The control logic of the linear ARC consists of three modules, 'Active Roll Compensation', 'Active Roll Damping' and 'Moment Distribution'.