一种用于植物叶片图像骨架提取的去毛刺方法

Pruning method for skeletonization of plant leaf images

为了获取能描述植物叶片区域主要分支结构的骨架,提出了一种植物叶片图像骨架中的毛刺去除方法。方法基于修剪后骨架所重构区域的质量判定骨架中的毛刺分支。重构区域与原区域的面积差指标被作为主控指标,以保证修剪不会明显影响区域的宏观形状;提出了突出部分骨架长度显著性指标作为保护性指标,以保护对于叶片识别分类具有重要意义的尖细状边缘细节不被剪除。完整的剪除过程由贪婪策略控制。在实际图像上的实验结果表明,所提方法能有效除去毛刺,并能得到更为准确的叶片区域骨架。

A skeleton pruning method for plant leaf image is proposed in order to obtain a skeleton representation for de- scribing major branches of the leaf regions. The method utilizes the region reconstruction quality of the pruned skeleton to determine the spurious branches in the skeleton. The difference between areas of the reconstructed region and the original region is used as the main control criterion to guarantee that the pruning does not affect the macro appearance of the re- gion. An index of the length of skeleton segment in spike region is proposed as the protective criterion to avoid pruning out thin and elongated boundary details, which are important for leaf recognition and classification. The pruning process as a whole is controlled by a greedy strategy. The experimental results on real-world images show that the proposed method can prune spurious branches effectively and give more accurate leaf region skeletons.