文献类型：期刊文献

英文题名：A novel weighted semi-supervised clustering algorithm and its application in image segmentation

作者：Li, Zhaofeng[1]; Liu, Lanqi[2]

第一作者：Li, Zhaofeng

机构：[1] College of Information Engineering, Henan Institute of science and technology, Xinxiang, Henan, China; [2] Henan University of Economics and Law, Zhengzhou, Henan, 450046, China

第一机构：College of Information Engineering, Henan Institute of science and technology, Xinxiang, Henan, China

年份：2016

卷号：17

期号：10

起止页码：6.1-6.7

外文期刊名：International Journal of Simulation: Systems, Science and Technology

收录：EI(收录号：20164302933023);Scopus(收录号：2-s2.0-84991585088)

语种：英文

外文关键词：Clustering algorithms - Gaussian distribution - Iterative methods - Matrix algebra - Pixels

摘要：In this paper we propose a novel weighted semi-supervised clustering algorithm and then study on how to apply it in the problem of image segmentation. We explain how to obtain weights of the semi-supervised clustering algorithm using the number of unlabeled data samples and the number of data samples. After defining the data sample weights, the next task is to obtain the cluster labels by optimizing a contingency matrix, and then cluster labels can be randomly initialized and updated in an iterative mode. Afterwards, based on the weighted semi-supervised clustering algorithm, novel image segmentation is given. In our proposed method, each image is represented by a D-dimensional random vector, and each pixel is drawn independently from the mixture density. Next, experts are invited to evaluate the quality of image segmentation using the training dataset. Afterwards, each segment of the images in the training dataset is assigned a class label. Then, a one-to-one mapping function between mixture model components and the segment classes is defined. Furthermore, each pixel of an image is assigned to the labelled component of the mixture which has the highest posterior probability. Finally, the image segmentation result can be obtained using the class label of the mixture component. Experimental results demonstrate that the proposed image segmentation algorithm can achieve both high segmentation accuracy and low computation cost. ? 2016, UK Simulation Society. All rights reserved.