摘要
针对自然图像分割中,由于单一的颜色空间难以表示复杂多变的场景信息以及目标与背景的低对比度等产生的过分割和误分割问题,提出了选择RGB、XYZ和LUV这3个颜色空间的增强图像进行基于层次聚类的融合分割的新方法。对Berkeley分割图像库中的多幅图像进行了多组分割实验,并与Mean-shift等多种经典分割方法进行了定性和定量的比较,实验结果表明本文方法的结果更符合手工标记的真实分割结果,在解决过分割和误分割方面具有明显的优势。
In order to eliminate over segmentation and false segmentation caused by difficulty of presenting complex scene information in a specific single color space and low contrast between object and background in natural image segmentation,this paper proposes a new segmentation method,which choses the enhanced images in RGB,XYZ and LUV color space of the original image and makes fusion segmentation based on hierarchical clustering.We carry out a variety of experiments on Berkeley segmentation database and compare with other classical segmentation methods,e.g.Mean-shift.The experiments demonstrate that the segmentation results of our method are more consistent with the ground truth and have distinct advantages in eliminating over segmentation and false segmentation.
出处
《电路与系统学报》
CSCD
北大核心
2011年第3期68-74,共7页
Journal of Circuits and Systems
基金
国家自然科学基金(60875012
60905005)
高等学校博士学科点专项科研基金(2009111110015)
关键词
图像分割
层次聚类
特征融合
多颜色空间
图像增强
image segmentation
hierarchical clustering
feature fusion
multi-color space
image enhancement
