• 한국컴퓨터정보학회논문지
• /
• 제26권1호
• /
• pp.27-35
• /
• 2021

본 연구에서는 레이블 임베딩의 분포를 반영하는 딥러닝 모형을 위한 새로운 스태킹 앙상블 방법론을 제안하였다. 제안된 앙상블 방법론은 기본 딥러닝 분류기를 학습하는 과정과 학습된 모형으로 부터 얻어진 레이블 임베딩을 이용한 군집화 결과로부터 소분류기들을 학습하는 과정으로 이루어져 있다. 본 방법론은 주어진 다중 분류 문제를 군집화 결과를 활용하여 소 문제들로 나누는 것을 기본으로 한다. 군집화에 사용되는 레이블 임베딩은 처음 학습한 기본 딥러닝 분류기의 마지막 층의 가중치로부터 얻어질 수 있다. 군집화 결과를 기반으로 군집화 내의 클래스들을 분류하는 소분류기들을 군집의 수만큼 구축하여 학습한다. 실험 결과 기본 분류기로부터의 레이블 임베딩이 클래스 간의 관계를 잘 반영한다는 것을 확인하였고, 이를 기반으로 한 앙상블 방법론이 CIFAR 100 데이터에 대해서 분류 성능을 향상시킬 수 있다는 것을 확인할 수 있었다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제8권9호
• /
• pp.3211-3230
• /
• 2014

Recently, sparse coding has achieved remarkable success in image representation tasks. In practice, the performance of clustering can be significantly improved if limited label information is incorporated into sparse coding. To this end, in this paper, a novel semi-supervised algorithm, called constrained sparse concept coding (CSCC), is proposed for image representation. CSCC considers limited label information into graph embedding as additional hard constraints, and hence obtains embedding results that are consistent with label information and manifold structure information of the original data. Therefore, CSCC can provide a sparse representation which explicitly utilizes the prior knowledge of the data to improve the discriminative power in clustering. Besides, a kernelized version of our proposed CSCC, namely kernel constrained sparse concept coding (KCSCC), is developed to deal with nonlinear data, which leads to more effective clustering performance. The experimental evaluations on the MNIST, PIE and Yale image sets show the effectiveness of our proposed algorithms.

• Journal of information and communication convergence engineering
• /
• 제11권3호
• /
• pp.179-184
• /
• 2013

Traditional clustering methods are usually based on the bag-of-words (BOW) model. A disadvantage of the BOW model is that it ignores the semantic relationship among terms in the data set. To resolve this problem, ontology or matrix factorization approaches are usually used. However, a major problem of the ontology approach is that it is usually difficult to find a comprehensive ontology that can cover all the concepts mentioned in a collection. This paper proposes a new document clustering method using semantic features and fuzzy relations for solving the problems of ontology and matrix factorization approaches. The proposed method can improve the quality of document clustering because the clustered documents use fuzzy relation values between semantic features and terms to distinguish clearly among dissimilar documents in clusters. The selected cluster label terms can represent the inherent structure of a document set better by using semantic features based on non-negative matrix factorization, which is used in document clustering. The experimental results demonstrate that the proposed method achieves better performance than other document clustering methods.

• 전기학회논문지
• /
• 제67권2호
• /
• pp.293-300
• /
• 2018

We propose a moving object extraction and tracking method for improvement of animal identification and tracking technology. First, we propose a method of merging separated moving objects into a moving object by using FCM (Fuzzy C-Means) clustering algorithm to solve the problem of moving object loss caused by moving object extraction process. In addition, we propose a method of extracting data from a moving object and a method of counting moving objects to determine the number of clusters in order to satisfy the conditions for performing FCM clustering algorithm. Then, we propose a method to continuously track merged moving objects. In the proposed method, color histograms are extracted from feature information of each moving object, and the histograms are continuously accumulated so as not to react sensitively to noise or changes, and the average is obtained and stored. Thereafter, when a plurality of moving objects are overlapped and separated, the stored color histogram is compared with each other to correctly recognize each moving object. Finally, we demonstrate the feasibility and applicability of the proposed algorithms through some experiments.

• International Journal of Computer Science & Network Security
• /
• 제22권3호
• /
• pp.29-36
• /
• 2022

In deep learning classification tasks, most models frequently assume that all labels are available for the training datasets. As such strategies to learn new concepts from unlabeled datasets are scarce. In fingerprint classification tasks, most of the fingerprint datasets are labelled using the subject/individual and fingerprint datasets labelled with finger type classes are scarce. In this paper, authors have developed approaches of classifying fingerprint images using the majorly known fingerprint classes. Our study provides a flexible method to learn new classes of fingerprints. Our classifier model combines both the clustering technique and use of deep learning to cluster and hence label the fingerprint images into appropriate classes. The K means clustering strategy explores the label uncertainty and high-density regions from unlabeled data to be clustered. Using similarity index, five clusters are created. Deep learning is then used to train a model using a publicly known fingerprint dataset with known finger class types. A prediction technique is then employed to predict the classes of the clusters from the trained model. Our proposed model is better and has less computational costs in learning new classes and hence significantly saving on labelling costs of fingerprint images.

• 디지털산업정보학회논문지
• /
• 제19권4호
• /
• pp.1-11
• /
• 2023

Clustering analysis is used in various fields including customer segmentation and clustering methods such as k-means are actively applied in the credit card customer segmentation. In this paper, we summarized the input features selection method of k-means clustering for the case of the credit card customer segmentation problem, and evaluated its feasibility through the analysis results. By using the label values of k-means clustering results as target features of a decision tree classification, we composed a method for prioritizing input features using the information gain of the branch. It is not easy to determine effectiveness with the clustering effectiveness index, but in the case of the CH index, cluster effectiveness is improved evidently in the method presented in this paper compared to the case of randomly determining priorities. The suggested method can be used for effectiveness of actively used clustering analysis including k-means method.

• 한국항행학회논문지
• /
• 제14권2호
• /
• pp.239-246
• /
• 2010

본 논문은 비음수 행렬 분해와 퍼지 관계를 이용한 새로운 문서군집 방법을 제안한다. 제안된 방법은 비음수 행렬 분해된 의미특징을 이용하여 군집 레이블과 군집의 대표 용어들을 선택함으로서 문서군집의 내부구조를 더 잘 표현할 수 있으며, 퍼지 관계 값을 이용한 군집은 문서군집에 유사하지 않은 문서를 더 잘 구분함으로써 문서군집의 성능을 높일 수 있다. 실험결과 제안방법을 적용한 문서군집방법이 다른 문서군집 방법에 비하여 좋은 성능을 보인다.

• 정보처리학회논문지B
• /
• 제17B권2호
• /
• pp.177-182
• /
• 2010

본 논문은 주성분 분석과 퍼지 연관을 이용한 새로운 문서군집 방법을 제안한다. 제안된 방법은 주성분 분석의 의미특징을 이용하여 군집 레이블과 군집의 대표 용어들을 선택하기 때문에 문서군집의 내부구조를 더 잘 표현할 수 있다. 또한 퍼지연관 값을 이용한 군집은 문서군집에 유사하지 않은 문서를 더 잘 구분함으로써 문서군집의 성능을 높일 수 있다. 실험결과 제안방법을 적용한 문서군집방법이 다른 문서군집 방법에 비하여 좋은 성능을 보인다.

• International journal of advanced smart convergence
• /
• 제7권2호
• /
• pp.33-37
• /
• 2018

In this study, we propose an iterative clustering algorithm that automatically clusters a set of voice signal data without a label into an optimal number of clusters and generates hmm model for each cluster. In the clustering process, the likelihood calculations of the clusters are performed using iterative hmm learning and testing while varying the number of clusters for given data, and the maximum likelihood estimation method is used to determine the optimal number of clusters. We tested the effectiveness of this clustering algorithm on a small-vocabulary digit clustering task by mapping the unsupervised decoded output of the optimal cluster to the ground-truth transcription, we found out that they were highly correlated.

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• 제27권2호
• /
• pp.146-159
• /
• 2023

This paper proposes a hybrid algorithm combining K-means clustering and watershed algorithms for flower segmentation and counting. We use the K-means clustering algorithm to obtain the main colors in a complex background according to the cluster centers and then take a color space transformation to extract pixel values for the hue, saturation, and value of flower color. Next, we apply the threshold segmentation technique to segment flowers precisely and obtain the binary image of flowers. Based on this, we take the Euclidean distance transformation to obtain the distance map and apply it to find the local maxima of the connected components. Afterward, the proposed algorithm adaptively determines a minimum distance between each peak and apply it to label connected components using the watershed segmentation with eight-connectivity. On a dataset of 30 images, the test results reveal that the proposed method is more efficient and precise for the counting of overlapped flowers ignoring the degree of overlap, number of overlap, and relatively irregular shape.