计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (14): 186-190.DOI: 10.3778/j.issn.1002-8331.1703-0254

• 图形图像处理 • 上一篇    下一篇

基于半监督的超像素谱聚类彩色图像分割算法

刘汉强,赵  静   

  1. 陕西师范大学 计算机科学学院,西安 710119
  • 出版日期:2018-07-15 发布日期:2018-08-06

Semi-supervised color image segmentation with  superpixels and spectral clustering

LIU Hanqiang, ZHAO Jing   

  1. School of Computer Science, Shaanxi Normal University, Xi’an 710119, China
  • Online:2018-07-15 Published:2018-08-06

摘要: 近年来谱聚类算法被广泛应用于图像分割领域,而相似性矩阵的构造是谱聚类算法的关键步骤。 针对传统谱聚类算法计算复杂度高难以应用到大规模图像分割处理的问题,提出了基于半监督的超像素谱聚类彩色图像分割算法。该算法利用超像素将彩色图像进行预分割,利用用户提供的少量标记信息构造预分割区域的基于半监督的模糊相似性测度,利用该相似性测度构造预分隔区域的相似性矩阵并通过规范切图谱划分准则对预分割区域进行划分得到最终的图像分割结果。由于少量标记信息和模糊理论的引入,提高了传统谱聚类的分割性能,对比实验也表明该算法在分割效果和计算复杂度上都有较大的改善。

关键词: 半监督, 超像素, 谱聚类, 模糊隶属度

Abstract: In recent years, spectral clustering algorithm is widely used in the field of image segmentation, and the construction of the similarity matrix is the key of spectral clustering algorithm. Due to the high computational complexity, spectral cluster algorithm is hard to be applied to the large scale image segmentation. Aiming at this problem, a semi-supervised color image segmentation with superpixels and spectral clustering is presented. Firstly, initial partition is performed by the superpixels method. Then the semi-supervised fuzzy similarity measure among the initial partition regions is constructed by utilizing the few label information. Finally, the similarity matrix of the initial partition regions is produced by this similarity measure, and these regions are grouped by normalized cut criterion. Because of the introduction of the label information and fuzzy theory, the experimental results show that the segmentation accuracy and computational complexity of the proposed algorithm paper are improved substantially.

Key words: semi-supervised, superpixels, spectral clustering, fuzzy membership