• Proceedings of the Korean Information Science Society Conference
• /
• 2003.04c
• /
• pp.334-336
• /
• 2003

얼굴인식 연구는 얼굴표현에 집중하여 연구가 이루어져 왔으며, 이에 따라 최종 인식단계에서는 간단한 분류기가 적용되었다. 그러나, 보다 강건한 분류 성능을 얻기 위해서는 복잡한 분류기를 도입할 필요성이 있다. 복잡한 분류기를 사용하면서, 얼굴인식의 특성상 온라인 학습을 가능하게 하기 위해서는 이진 분류기 분해/결합 방법을 도입한다. 이러한 기법으로는 OPC와 PWC가 대표적이다. 본 방법의 성능은 기본이진 분류기의 성능에 좌우되기 때문에 SVM을 적용한다. 본 논문에서는 대표적인 방법인 OPC와 PWC의 장단점을 설명하고, 거절 조건을 갖는 이들 분류기들의 순차 결합 방법을 제안한다. ORL얼굴데이터 베이스에 대한 실험에서는, 제안하는 방법을 통해 얼굴인식 성능을 향상시킬 수 있음을 보인다.

• Journal of Korean Society of Transportation
• /
• v.21 no.2
• /
• pp.119-130
• /
• 2003

SP data have been widely used in assessing new transport policies and transport related plans. However, one of criticisms of using SP is that respondents may show different reaction between hypothetical experiments and real life. In order to overcome the problem, combination of SP and RP data has been suggested and the combined methods have been being developed. The purpose of this paper is to suggest a new SP and RP combined method using error component method and to verify the method. The error component method decomposes IID extreme value error into non-IID error component(s) and an IID error component. The method estimates both of component parameters and utility parameters in order to obtain relative variance of SP data and RP data. The artificial SP and RP data was created by using simulation and used for the analysis, and the estimation results of the error component method were compared with those of existing SP and RP combined methods. The results show that regardless of data size, the parameters of the error component method models are similar to those assumed parameters much more than those of the existing SP and RP combined models, indicating usefulness of the error component method. Also the values of time for error component method are more similar to those assumed values than those of the existing combined models. Therefore, we can conclude that the error component method is useful in combining SP and RP data and more efficient than the existing methods.

• Journal of KIISE
• /
• v.43 no.3
• /
• pp.362-369
• /
• 2016

This study is directed toward the design of a hybrid algorithm for syllable-based Korean POS tagging. Previous syllable-based works on Korean POS tagging have relied on a sequence labeling method and mostly used only a machine learning method. We present a new algorithm integrating a machine learning method and a pre-analyzed dictionary. We used a Sejong tagged corpus for training and evaluation. While the machine learning engine achieved eojeol precision of 0.964, the proposed hybrid engine achieved eojeol precision of 0.990. In a Quiz domain test, the machine learning engine and the proposed hybrid engine obtained 0.961 and 0.972, respectively. This result indicates our method to be effective for Korean POS tagging.

• Proceedings of the Korean Society Of Semiconductor Equipment Technology
• /
• 2006.10a
• /
• pp.35-38
• /
• 2006

디스플레이 산업에 이용되는 적층판(layered plates)의 결합공정 중 냉각공정에서 열팽창계수의 차이로 인해 잔류응력이 발생하고 심하면 적층판에 크랙(crack)이 발생한다. 본 연구에서는 적층판의 결합공정을 대상으로 현상을 분석하고 이 과정을 시뮬레이션하는 해석 프로그램을 개발하였다. 또한 이를 토대로 향후의 새로운 제품에 대해서도 크랙과 같은 문제점을 최소화 할 수 있는 신뢰성 있는 공정 셋업을 제시하기 위해 차원감소법(dimension reduction method)과 근사화 방법인 반응표면법(response surface method), 순차적 근사최적화 기법(Sequential Approximate Optimization, SAO)을 이용하여 신뢰성기반의 강건최적설계를 하였다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2005.07a
• /
• pp.721-723
• /
• 2005

상위 단계 합성에서의 자원 공유 (또는 바인딩)는 최종회로 결과의 질에 심각한 영향을 미치는 중요한 작업들 가운데 하나이다. 기존의 자원 공유 기법의 목적은 주어진 자원 제약 조건하에서 회로의 지연 시간을 최소화하거나, 주어진 지연시간 조건하에서 하드웨어의 복잡도 (예: 회로 면적)를 최소화하는 것이다. 본 논문에서는, 자원 공유 문제를 회로에서 소모되는 전력 소모를 줄이는 각도에서 보고 있다. 구체적으로, 전압 배정 작업을 자원 공유 문제와 결합하여, 두 개의 작업을 통합된 방식으로 해결함으로서 회로 결과에서의 소모되는 전력 소모량을 충분히 그리고 효과적으로 줄이는데 목표를 두고 있다. 벤치마크를 사용한 실험에서 우리는 제안한 방법을 사용하면, 기존의 순차적인 자원 공유와 전압 배정 적용 방식보다 $0.7\%-16\%$ 더 적은 전력 소모를 가짐을 알 수 있었다.

• Proceedings of the Korea Multimedia Society Conference
• /
• 2002.05c
• /
• pp.223-228
• /
• 2002

숫자를 인식하는 방법에는 여러 가지가 있지만 단일 인식기를 구성하는 경우보다 다중 인식기를 이용하는 방법이 뛰어나다는 연구 발표가 있었다. 그래서 다중 인식에 대한 연구가 활발히 진행되고 있는데, 다중 인식기를 이용하는 방법에는 크게 직렬 조합형과 병렬 조합형이 있는데, 직렬 조합형은 인식기를 파이프라인 처럼 구성하여 순차적으로 인식하는 방법이고, 병렬조합형은 인식기를 병렬로 구성하여 인식기들의 결과를 조합하여 얻어내는 방법이다. 본 논문에서는 무제약 필기체 숫자를 인식하기 위한 병렬 파이프라인 다중 인식기의 구조와 결합 방법을 제안 하고자 한다. 조선대학교 필기체 숫자 데이터를 이용하여 실험한 결과 기존의 방법보다 비교적 높은 인식률을 나타내었다.

• The Korean Journal of Applied Statistics
• /
• v.21 no.2
• /
• pp.275-290
• /
• 2008

The size of microarray gene expression data is very big and its observation process is also very complex. Thus missing values are frequently occurred. In this paper we propose the sequential partial least squares(SPLS) regression fitting method to estimate missing values for time course gene expression data that has correlations among observations over time points. The SPLS method is to combine the sequential technique with the partial least squares(PLS) regression fitting method. The usefulness of method proposed is evaluated through some simulation study for three yeast time course data.

• Journal of the Institute of Electronics Engineers of Korea TC
• /
• v.41 no.9
• /
• pp.39-45
• /
• 2004

A design method is proposed for the sequence detection with fixed decision delay with less hardware complexity using the concept of the Voronoi diagram and its dual, the Delaunay tessellation. This detector design is based on the Fixed Delay Tree Search (FDTS) detection. The FDTS is a computationally efficient sequence detection algerian and has been shown to achieve near-optimal performance in the severe Intersymbol Interference (ISI) channels when combined with decision feedback equalization and the appropriate channel coding. In this approach, utilizing the information contained in the Voronoi diagram or equivalently the Delaunay tessellation, the relative location of the detector input sequence in the multi-dimensional Euclidean space is found without any computational redundancy, which leads to a reduced complexity implementation of the detector.

• Journal of Broadcast Engineering
• /
• v.9 no.1
• /
• pp.43-53
• /
• 2004

In this paper, we propose a robust deinterlacing algorithm which combines edge dependent interpolation (EDI) and global motion compensation (GMC). Generally, EDI algorithm shows a visually better performance than any other deinterlacing algorithm using one field. However, due to the restriction of information in one field, a high duality progressive image from Interlaced sources cannot be acquired by intrafield methods. On the contrary, since algorithms based on motion compensation make use of not only spatial information but also temporal information, they yield better results than those of using one field. However, performance of algorithms based on motion compensation depends on the performance of motion estimation. Hence, the proposed algorithm makes use of mixing process of EDI and GMC. In order to obtain the best result, an adaptive thresholding algorithm for detecting the failure of GMC is proposed. Experimental results indicate that the proposed algorithm outperforms the conventional approaches with respect to both objective and subjective criteria.

• Journal of KIISE
• /
• v.42 no.7
• /
• pp.901-909
• /
• 2015

In this paper, to build a parallel corpus between Korean and English in Wikipedia. We proposed a method to find similar sentences based on language resources and topic modeling. We first applied language resources(Wiki-dictionary, numbers, and online dictionary in Daum) to match word sequentially. We construct the Wiki-dictionary using titles in Wikipedia. In order to take advantages of the Wikipedia, we used translation probability in the Wiki-dictionary for word matching. In addition, we improved the accuracy of sentence similarity measuring method by using word distribution based on topic modeling. In the experiment, a previous study showed 48.4% of F1-score with only language resources based on linear combination and 51.6% with the topic modeling considering entire word distributions additionally. However, our proposed methods with sequential matching added translation probability to language resources and achieved 9.9% (58.3%) better result than the previous study. When using the proposed sequential matching method of language resources and topic modeling after considering important word distributions, the proposed system achieved 7.5%(59.1%) better than the previous study.