颜色阈值与聚类结合的自动油菜花图像分割

Automated segmentation of oilseed rape flowers with combing color threshold and clustering

提出了一种结合HSI(Hue,Sturation,Intensity)颜色阈值和K均值聚类的自然光照下油菜花图像自动分割方法。首先将原始图像转换到HSI颜色空间,然后在H通道中通过颜色阈值将得到候选目标区域。原始RGB(Red,Green,Blue)图像中候选目标区域内的像素转换为Lab颜色空间,最后,采用K均值聚类算法对候选目标区域进行分割。聚类过程改进了阈值化结果,而阈值过程使聚类在具有固定类别数的简化数据中实现,克服了聚类问题中结果依赖聚类数的问题。对于目标敏感颜色阈值和具有固定聚类数的聚类的组合,可以简化油菜花的分割,而常规的颜色分割方法则依赖复杂的多阈值搜索或聚类数的确定。实验中测试了三十张油菜种子的图像,最终得到的分割结果的平均分数即F 1度量为84%。

In order to improve the segmentation accuracy of oilseed rape flowers under natural illumination, an automated segmentation method combing HSI color threshold and K-means clustering is proposed in this paper.The original image is firstly converted into HSI color space.Then a candidate target region is located by color threshold in H channel.Pixels within the candidates target region in the original RGB image are converted to Lab color space.Finally, K-means clustering algorithm is used to segment the localized image.The clustering process improves the result by threshold, while the threshold process makes the clustering implemented in simplified data with a fixed number of categories.Thus an automated and accurate segmentation can be achieved.For the combination of object sensitive color threshold and clustering with fixed clustering number, oilseed rape flowers segmentation is simplified while general color segmentation methods depend on complex multiple threshold search, or the cluster number determination.In the experiment, thirty images of oilseed rape flowers are tested.The average F1 score of the segmentation result reach 84%.