计算机科学与探索 ›› 2018, Vol. 12 ›› Issue (3): 494-501.DOI: 10.3778/j.issn.1673-9418.1708034

• 理论与算法 • 上一篇    下一篇

优化加权核K-means聚类初始中心点的SLIC算法

杨  艳,许道云+   

  1. 贵州大学 计算机科学与技术学院,贵阳 550025
  • 出版日期:2018-03-01 发布日期:2018-03-08

SLIC Algorithm Based on Optimizing Initial Center Point of Weighted Kernel K-means Clustering

YANG Yan, XU Daoyun+   

  1. College of Computer Science and Technology, Guizhou University, Guiyang 550025, China
  • Online:2018-03-01 Published:2018-03-08

摘要: 超像素是近年来快速发展的一种图像预处理技术,被广泛应用于计算机视觉领域。简单线性迭代聚类(simple linear iterative clustering,SLIC)算法是其中的一种图像预处理技术框架,该算法根据像素的颜色和距离特征进行聚类来实现良好的分割结果。然而,SLIC算法尚存在一些问题。基于优化加权核K-means聚类初始中心点,提出一种新的SLIC算法(WKK-SLIC算法)。算法基于图像像素之间的颜色相似性和空间相似性度量,采用超像素分割的归一化割公式,使用核函数来近似相似性度量。算法将像素值和坐标映射到高维特征空间中,通过对该特征空间中的每个点赋予适当的权重,使加权K均值和归一化割的目标函数的优化在数学上等价。从而通过在所提出的特征空间中迭代地应用简单的K-means聚类来优化归一化割的目标函数。在WKK-SLIC算法中,采用密度敏感的相似性度量计算空间像素点的密度,启发式地生成K-means聚类的初始中心以达到稳定的聚类结果。实验结果表明,WKK-SLIC算法在评估超像素分割的几个标准上优于SLIC算法。

关键词: 超像素, 超像素分割, 加权核K-means, 密度, 初始中心点

Abstract: Superpixel is a kind of image preprocessing technology which has been developed rapidly in recent years and is widely used in the field of computer vision. Simple linear iterative clustering (SLIC) algorithm is a kind of image preprocessing technology framework. The algorithm achieves a good segmentation result by clustering based on the color and distance characteristics of pixels. However, some problems still exist in the SLIC algorithm. This paper proposes a new SLIC algorithm (WKK-SLIC algorithm) based on optimizing initial center point of weighted kernel K-means clustering. The algorithm is based on the similarity metric that measures the color similarity and space proximity between image pixels, uses the normalization formula of the superpixel segmentation and uses a kernel function to approximate the similarity metric. The algorithm maps the values and coordinates of pixels into the high-dimensional feature space. By assigning an appropriate weight to each pixel in the feature space, the objective function optimization of the weighted K-means and the normalized cuts is mathematically equivalent. Thus the objective function of normalization is optimized by iteratively applying simple K-means clustering in the proposed feature space. The algorithm uses density-sensitive similarity measure to calculate the density of spatial pixels, and heuristically generates the initial center of K-means clustering to achieve stable clustering results. The experimental results show that the algorithm outperforms the SLIC in several commonly used metrics of superpixel segmentation.

Key words: superpixel, superpixel segmentation, weighted kernel K-means, density, initial center point