关键词: 图像分割, K\, 均值聚类, 模板匹配, 形状上下文, 黄瓜叶片

Abstract: In order to realize automatic identification of cucumber disease leaves in the complex background， target leaves should be segmented from the complex background first to facilitate the subsequent feature extraction and disease recognition. For this purpose， K\|means clustering algorithm was initially used to remove the non\|green parts of the image， and then the approach based on LOG operator was proposed to select the candidate leaf areas. Finally， template matching was conducted based on shape context. During the matching process， the position， size and direction of the leaves were firstly identified via the detection of the growing point and apex of leaves to improve the matching efficiency， along with the search for the optimal matching based on superpixel to reduce the search complexity. To evaluate the feasibility of the proposed segmentation approach， 20 images of cucumber diseased leaves were segmented， and the result was compared with manual segmentation. It was shown that the proposed segmentation approach could extract images with cucumber diseased leaves from the complex background， and the average segmentation accuracy rate was 947%， which built a solid foundation for the subsequent feature extraction of cucumber lesion.

Key words: image segmentation, K\, means clustering, template matching, shape context, cucumber leaves