Abstract: Traditional segmentation methods can obtain better result for these images which have distinct difference between the goal and background area. However these methods are difficult to obtain ideal disease speckle for diseased lentil leave images which have minor difference among normal leaves and disease speckles. So in this paper, it proposes a method that is suitable for diseased lentil leave images. This method has two stages including initial segmentation and secondary segmentation. Color gradient graph of these images is computed, then the Otsu algorithm is applied to eliminate lower gradient. Watershed algorithm is used to pre-segment the images, then a rough target zone based on zone area features is gained. FCM algorithm is applied to rough target zone. By analyzing difference between green alley of disease speckle and normal leaves, disease speckle is acquired. Experimental results show good effect of segmenting disease speckle with this method.

Key words: lentil diseased leaves, image segmentation, color gradient, watershed algorithm, Fuzzy C-Means（FCM）

摘要： 传统的分割方法针对目标和背景灰度值差距大的图像能得到较好的分割效果，但在对正常叶片和病斑灰度值相似度高的扁豆病害叶片图像分割时，难以得到理想的目标病斑。针对该问题，提出了一种适合正常叶片和病斑相似度高的图像剥离分割方法。该方法包括初始分割和二次分割两个步骤。初始分割是基于样本图片的彩色梯度图，采用最大类间标准方差与分水岭相结合的算法获得病斑粗略区域。二次分割是对粗略目标区域进行模糊C聚类分割得到目标病斑。实验结果表明，该剥离分割算法能提高病斑分割精确度，较好地分割出病斑目标。

关键词: 扁豆病害叶片, 图像分割, 彩色梯度, 分水岭, 模糊C聚类