基于光源不变图和聚类算法的叶片阴影中病斑的分割

Segmentation of shaded plant lesions using illuminant invariant image and clustering algorithm

【目的】提出了一种基于光源不变图的病斑分割方法,以提高病斑识别程序的准确性和稳定性。【方法】将阴影区和非阴影区视为不同光源照明,通过最小熵法计算原图的光源不变图,在该图上采用K均值聚类算法对病斑进行分割,以采集的病斑叶片为材料,对该方法的处理效果进行验证。【结果】比较光源不变图法和H分量法的处理结果后发现,采用光源不变图法处理病斑不同区域的平均差异较H分量法更低,仅为10.7%;聚类分割算法对使用光源不变图法处理病斑图像的分割准确率为95.0%,较H分量法具有更高的正确率,且误检率更低。【结论】采用光源不变图法对病斑图像处理的效果好、性能稳定,同一目标在不同光照条件下处理结果的一致性较高。

【Objective】This study proposed a segmentation method based on illuminant invariant image to improve the accuracy and stability of lesion recognition program.【Method】This method deemed shadow and non-shadow areas as different illumination sources and illuminant invariant image of the original map was calculated using entropy minimization method.Then the K-mean clustering algorithm was used to segment plant lesions in shaded and non-shaded areas based on illuminant invariant image.Experiments were also carried out to verify the effectiveness.【Result】Comparison of these two methods showed that the difference of grey value between the lesions in shadow or non-shadow areas when using illuminant invariant image method was 10.7%,much less than H component method.Clustering segmentation algorithm based on illuminant invariant image had the accuracy of 95.0%,better than H component method and the false reject rate was lower.【Conclusion】Illuminant invariant image method was better and more stable than H component method.It also had higher consistency for same lesion under different lighting conditions.