计算机工程与应用 ›› 2014, Vol. 50 ›› Issue (3): 129-134.

• 图形图像处理 • 上一篇    下一篇

小麦冠层图像H分量的K均值聚类分割

黄  芬1,于  琪1,姚  霞2,商贵艳2,朱  艳2,伍艳莲1,黄  宇2   

  1. 1.南京农业大学 信息科学与技术学院,南京 210095
    2.南京农业大学 国家信息农业工程技术中心,南京 210095
  • 出版日期:2014-02-01 发布日期:2014-01-26

K-means clustering segmentation for H weight of wheat canopy image

HUANG Fen1, YU Qi1, YAO Xia2, SHANG Guiyan2, ZHU Yan2, WU Yanlian1, HUANG Yu2   

  1. 1.College of Information Science and Technology, Nanjing Agricultural University, Nanjing 210095, China
    2.National Information Technology Center of Agricultural Engineering, Nanjing Agricultural University, Nanjing 210095, China
  • Online:2014-02-01 Published:2014-01-26

摘要: 大田环境下小麦冠层图像具有光照不均匀、背景复杂及阴影遮挡等特点,经典图像分割算法存在精度低、过分割等问题,提出一种基于HSI空间下H分量的K均值聚类算法。使用[R+G-B]归一化处理RGB空间下的彩色图像,以抑制其B分量;将归一化图像进行RGB到HSI的颜色空间转化;根据光照是否均匀,使用K均值聚类算法对彩色图像的H分量进行不同的聚类处理,经形态学开运算及去噪处理获得最终目标图像。实验表明,该方法对不同施氮量、不同光照、不同生长时期小麦冠层图像的分割效果较好,相对基于Lab空间的K-means聚类分割,该方法可一定程度避免过分割现象;相对基于H分量的Otsu算法,对光照不均匀图像分割更完整,对复杂背景图像分割更精确。

关键词: 小麦冠层图像分割, HSI颜色空间, H分量, K均值聚类

Abstract: Wheat canopy image under the natural light has the feature of nonuniform illumination, complicated background with shadows. A K-means clustering algorithm based on HSI color space is proposed to conquer the problem of low accuracy and over segmentation existing in classic image segmentation algorithm. To restrain B weight, [R+G-B] is used to normalize color images in RGB space. After transforming the normalized image from RGB to HSI color space, the different methods of K-means cluster used to the H weight depend on whether the sunlight is uniform or not. The final image is gained after using mathematical morphology and noise-removal process. The experiments show that compared with K-means cluster processing in Lab space and Otsu algorithm based on H weight, the method based on H weight can avoid over segmentation and has accurate segmentation results in different N fertilization, different illumination and different periods.

Key words: wheat canopy image segmentation, HSI color space, H weight, K-means clustering