Abstract: The method based on combining Edgeflow and normal ISODATA has the disadvantage of the low processed speed and the inaccurate result, the method based on combining Canny and improved ISODATA is proposed for image segmentation. Color space of RGB is converted to color space of Lab, Canny algorithm is applied to get the initial over-segmentation image with the L of the color space Lab, and ISODATA is applied to get the conglomeration image in the color space of Lab. Then, the result based on ISODATA algorithm is applied to solve over-
segmentation problem caused by Canny. Experimental results show that the method proposed for image segmentation in this paper is efficient to conglomerate image segmentation.

Key words: conglomerate image, color space, image-segmentation, Canny, ISODATA

摘要： 针对基于边缘流和传统ISODATA（迭代自组织的数据分析算法）相融合的算法对砾岩图像进行分割时存在速度慢、分割不准确的问题，提出了一种融合Canny和改进ISODATA聚类相结合的砾岩图像分割算法。该算法将图像从RGB色彩空间转换到Lab色彩空间，利用Canny算子对砾岩图像的L分量进行边缘提取，得到过分割图像；运用改进的ISODATA聚类算法进行聚类，得到聚类图像，消除了Canny算子的过分割问题。在砾岩图像的分割应用中，该算法取得了较好的分割效果。

关键词: 砾岩图像, 颜色空间, 图像分割, Canny算子, ISODATA聚类