摘要
叶片图像分割是自动化分类的先决步骤。提出一种基于L*a*b*彩色空间和局域动态阈值的叶片图像分割方法。该方法首先把叶片的RGB图像转换为L*a*b*图像;然后针对b*通道图像,估算出目标叶片所在的局部矩形范围;在此范围内,统计获得像素值分布直方图;最后利用最大类间方差阈值分割法,先算出局部矩形范围内的阈值继而进行全图的分割。实验结果表明:该方法对已采集的叶片图像,包括颜色偏暗的叶片的图像,均具有良好的分割效果。由于锁定目标叶片所处的局部矩形范围,找到了适应于目标叶片分割的阈值,从而更好地实现了南天竺等叶片图像的分割。此外,分割过程不包含数学形态学的开闭运算,使得叶片边缘的细节得以完整保留。
Abstract Leaf image segmentation is a prerequisite step for automatic classification of leaves. We present a leaf image segmentation method, which is based on L＊ a＊ b＊ col0ur space and the local dynamic threshold. First, the RGB image of leaf is converted to an L＊ a＊ b＊ one. Then in terms of the channel b ＊ image, the rectangular local range, where the target leaf is located, is estimated. Afterwards, the dis- tribution histogram of pixel values within this range is got by counting. Finally the algorithm of threshold segmentation based on maximum be- tween-class variance is used to calculate the threshold in rectangular local range first and followed by performing the segmentation on the entire image. Experimental results have shown that this method brings about desirable effects on the segmentation of the leaf image samples, inclu- ding those that are relatively dark. As the rectangular local range for the target leaf is identified, and the threshold value suitable for segmentation is worked out, thus better segmentation on the leaf, such as that of Common Nandina, is implemented. Besides, the segmentation excludes the mathematical morphological operation of openings and closings so that the details of the leaves edge can be preserved completely.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第1期232-235,共4页
Computer Applications and Software
基金
广东省教育科研"十一五"规划研究项目(2009tjk082)
广东省建设中医药强省科研课题(20111233)
广东中医药大学教育研究课题(1055)
作者简介
高理文，讲师，主研领域：智能化检测。
林小桦，讲师。
