• The KIPS Transactions:PartB
• /
• v.13B no.4 s.107
• /
• pp.439-448
• /
• 2006

Previous recognition/clustering algorithms such as Kohonen SOM(Self-Organizing Map), MLP(Multi-Layer Percecptron) and SVM(Support Vector Machine) might not adapt to unexpected input pattern. And it's recognition rate depends highly on the complexity of own training patterns. We could make up for and improve the weak points with lowering complexity of original problem without losing original characteristics. There are so many ways to lower complexity of the problem, and we chose a kernel concepts as an approach to do it. In this paper, using a kernel concepts, original data are mapped to hyper-dimension space which is near infinite dimension. Therefore, transferred data into the hyper-dimension are distributed spasely rather than originally distributed so as to guarantee the rate to be risen. Estimating ratio of recognition is based on a new similarity-probing and learning method that are proposed in this paper. Using CEDAR DB which data is written in cursive letters, 0 to 9, we compare a recognition/clustering performance of kSOM that is proposed in this paper with previous SOM.

• The KIPS Transactions:PartB
• /
• v.9B no.2
• /
• pp.202-214
• /
• 2002

In this paper, we propose a modified FCM (MFCM) algorithm to solve the problems of the FCM algorithm and the fuzzy clustering algorithm using an average intracluster distance (FCAID). The MFCM algorithm grants the regular grade of membership in the small size of cluster. And it clears up the convergence problem of objective function because its objective function is designed according to the grade of membership of it, verified, and used for clustering data. So, it can solve the problem of the FCM algorithm in different size of cluster and the FCAID algorithm in the convergence problem of objective function. To verify the MFCM algorithm, we compared with the result of the FCM and the FCAID algorithm in data clustering. From the experimental results, the MFCM algorithm has a good performance compared with others by classification entropy.

• Journal of KIISE:Databases
• /
• v.32 no.5
• /
• pp.498-508
• /
• 2005

A shape data is used one oi most important feature for image retrieval as data to reflect meaning of image. Especially, structural feature of shape is widely studied because it represents primitive properties of shape and relation information between basic units well. However, most structural features of shape have the problem that it is not able to guarantee an efficient search time because the features are expressed as graph or tree. In order to solve this problem, we generate curvature-based morphological graph, End design key to cluster shapes from this graph. Proposed this graph have contour features and morphological features of a shape. Shape retrieval is accomplished by stages. We reduce a search space through clustering, and determine total similarity value through pattern matching of external curvature. Various experiments show that our approach reduces computational complexity and retrieval cost.

• Journal of the Korea Society of Computer and Information
• /
• v.12 no.1 s.45
• /
• pp.33-42
• /
• 2007

Mobile ad-hoc network(MANET) can increase independence and flexibility of network because it consists of mobile node without the aid of fixed infrastructure. But, Because of unrestriction for the participation and breakaway of node, it has the difficulty in management and stability which is a basic function of network operation. Therefore, to solve those problems, we suggest a distributed weighted clustering algorithm from a manageable and stable point of view. The suggested algorithm uses distributed weighted clustering algorithm when it initially forms the cluster and uses a concept which is distributed gateway and sub-cluster head to reduce the re-clustering to the minimum which occurs mobile nodes after forming the cluster. For performance evaluation, We compare DCA and WCA with the suggested algorithm on the basis of initial overhead, resubscriber rate and a number of cluster.

• The KIPS Transactions:PartC
• /
• v.15C no.2
• /
• pp.119-124
• /
• 2008

The LEACH is a study method of a clustering base that was representative in many routing techniques that was suggested in sensor network environment. This is suggested technique to manage the sensor network group which consisted of many sensors in efficiency. However, it does not consider energy remaining quantity of the cluster header to manage cluster group, the problem where the cluster group is able to destroyed on the middle which transmits data. We are going to propose to solve such a problem by this paper for new clustering technique to accomplish management for a cluster group. It uses the technique which it proposes from the dissertation which it sees and are to the data transfer and the control also the connection directivity of the cluster group to improve also the lag time diminishes, test result it will lead and the possibility which it will know it was.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.12 no.3
• /
• pp.231-238
• /
• 2002

As a modeling method where the merits of fuzzy inference system and evolutionary computation are put together, evolutionary fuzzy modeling performs global approximate optimization. The paper proposes fuzzy clustering as fuzzy rule generation process which is one of the most important steps in evolutionary fuzzy modeling. With application of fuzzy clustering into the experiment or simulation results, fuzzy rules which properly describe non-linear and complex design problem can be obtained. The efficiency of evolutionary fuzzy modeling can be improved utilizing the membership degrees of data to clusters from the results of fuzzy clustering. To ensure the validity of the proposed method, the real design problem of an automotive inner trim is applied and the global approximation is achieved. Evolutionary fuzzy modeling is performed for several cases which differ in the number of clusters and the criterion of rule selection and their results are compared to prove that the proposed method can provide proper fuzzy rules for a given system and reduce computation time while maintaining the errors of modeling as a satisfactory level.

• Journal of KIISE
• /
• v.43 no.1
• /
• pp.40-46
• /
• 2016

Digital road map generation is primarily based on artificial satellite photographing or in-site manual survey work. Therefore, these map generation procedures require a lot of time and a large budget to create and update road maps. Consequently, people have tried to develop automated map generation systems using GPS trajectory data sets obtained by public vehicles. A fundamental problem in this road generation procedure involves the extraction of representative trajectory such as main roads. Extracting a representative trajectory requires the base data set of piecewise line segments(GPS-trajectories), which have close starting and ending points. So, geometrically similar trajectories are selected for clustering before extracting one representative trajectory from among them. This paper proposes a new divide- and-conquer approach by partitioning the whole map region into regular grid sub-spaces. We then try to find similar trajectories by sweeping. Also, we applied the $Fr{\acute{e}}chet$ distance measure to compute the similarity between a pair of trajectories. We conducted experiments using a set of real GPS data with more than 500 vehicle trajectories obtained from Gangnam-gu, Seoul. The experiment shows that our grid partitioning approach is fast and stable and can be used in real applications for vehicle trajectory clustering.

• Journal of KIISE:Software and Applications
• /
• v.30 no.11
• /
• pp.1054-1061
• /
• 2003

Q-Learning, a representative algorithm of reinforcement learning, experiences repeatedly until estimation values about all state-action pairs of state space converge and achieve optimal policies. When the state space is high dimensional or continuous, complex reinforcement learning tasks involve very large state space and suffer from storing all individual state values in a single table. We introduce Q-Map that is new function approximation method to get classified policies. As an agent learns on-line, Q-Map groups states of similar situations and adapts to new experiences repeatedly. State-action pairs necessary for fine control are treated in the form of rule. As a result of experiment in maze environment and mountain car problem, we can achieve classified knowledge and extract easily rules from Q-Map

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2014.06a
• /
• pp.38-41
• /
• 2014

본 논문에서는 필기 문서 영상을 분석하여 단어 단위로 요소들을 분할하는 방법을 제안한다. 일반적으로 인쇄 문서에 비하여 필기 문서에서는 글자 간 간격이 일정하지 않을 뿐만 아니라 필기자 또는 작성된 언어에 따라 특성이 매우 다르게 나타나기 때문에 단어를 분리하는 것은 어려운 문제로 간주되었고 많은 연구가 진행되었다. 제안하는 방법은 이 문제를 해결하기 위하여 글자 획의 두께를 고려하여 정규화시킨 각 연결 요소간 간격과 간격 안에 존재하는 글자 픽셀의 수로 구성된 2 차원의 특징값을 추출하였다. 이 특징값을 바탕으로, 제안하는 방법은 k-평균 클러스터링을 이용하여 각 텍스트라인을 구성하는 연결 요소간 간격을 단어 사이의 간격과 단어 내부 글자간의 간격으로 분류하였다. ICDAR 2013 Handwriting Segmentation Contest 데이터베이스에 대한 실험 결과 제안하는 방법은 가장 우수한 성능을 나타내었다.

• Proceedings of the KIEE Conference
• /
• 2008.07a
• /
• pp.1889-1890
• /
• 2008

본 연구에서는 퍼지 추론 메커니즘에 기반 한 Polynomial RBF Neural Network(p-RBFNN)를 설계하고 얼굴인식 문제로 적용하여 분류기로서의 성능을 분석한다. 제안된 p-RBFNN 구조는 FCM 클러스터링에 기반 한 분할 함수를 활성 함수로 사용하며, 다항식 함수로 구성된 연결가중치를 사용함으로서 기존 신경회로망 분류기의 선형적인 특성을 개선한다. p-RBFNN 구조는 언어적 해석관점에서 "If-then"의 퍼지 규칙으로 표현되며 퍼지 추론 메커니즘에 의해 구동된다. 즉 조건부, 결론부, 추론부 세 가지의 기능적 모듈로 나뉘어 네트워크 구조가 형성된다. 조건부는 FCM 클러스터링을 사용하여 입력 공간을 분할하고, 결론부는 분할된 로컬 영역을 다항식 함수로 표현한다. 마지막으로, 네트워크의 최종출력은 추론부의 퍼지추론에 의한다. 또한 제안된 p-RBFNN을 얼굴인식 문제로 적용하여 성능을 분석한다.