| 刘鑫磊，梁 耀，黎双文，钟伟镇，李丰果.复杂背景下重叠椭圆形叶片的分割算法[J].广东农业科学,2019,46(5):149-157 |
|
查看全文
HTML
复杂背景下重叠椭圆形叶片的分割算法 |
| Segmentation Algorithm of Overlapping Elliptical Leaves in Complex Background |
| |
| DOI：10.16768/j.issn.1004-874X.2019.05.021 |
| 中文关键词: 图像分割 复杂背景 凹点检测 凸点检测 贪婪算法 |
| 英文关键词: Image segmentation complex background pit detection bump detection greedy algorithm |
| 基金项目:国家自然科学基金（11575064）；广东省自然科学基金（2016A030313433） |
|
| 摘要点击次数: 1864 |
| 全文下载次数: 571 |
| 中文摘要: |
| 【目的】提出一种复杂背景下重叠椭圆形叶片的分割算法。【方法】利用超绿算法将重叠叶片主
体及部分绿色背景分割出来；通过选择最大连通区域选择重叠叶片的主体，去除叶柄后利用凹点检测和凸点检
测获得叶片重合部位的凹点和叶片的尖端点；利用获得的凹点与尖端点，截取上层叶片未重叠区域的部分边缘
点并以上层叶片的两个尖端点所确定的直线作为对称轴将其翻转至重叠区域作为贪婪算法的初始点寻找重叠区
域的边缘；对获得的边缘进行平滑和凸包处理，即可获得目标叶片的边缘，从而较为正确地分割出上层叶片。
对于重叠的下层叶片，则以整个重叠叶片减去分割出的上层叶片，即可得到下层叶片未被覆盖的区域；将下层
叶片完整的一半以其尖端点形成的直线为对称轴进行翻转，即可得到下层叶片的补全图。【结果】该算法可以
实现复杂背景椭圆形重叠叶片的分割，且平均错分率在 3.0% 以下。【结论】针对复杂背景下椭圆形重叠叶片的
分割，提出结合凸点检测和凹点检测的贪婪算法可以较为准确地分割目标叶片，平均错分率在 3% 以下，即分
割准确率在 97% 以上。 |
| 英文摘要: |
| 【Objective】A segmentation algorithm of overlapping elliptical leaves in a complex background was
proposed.【Method】Firstly, the main body of the overlapping leaves and part of the green background was segmented
through the super-green algorithm. The body of the overlapping leaves was selected by selecting the maximum connected
area. The concave point of overlapping parts and the tips of the leaves were obtained via pit detection and bump detection
after removing the petiole. The line defined by the two tips of upper-layer leaf and the edges of the non-overlapping areas
of the upper leaf was overturned to the overlapping area as the initial point of the greedy algorithm to find the edge of the
overlapping area.The edge of the target leaves could be obtained so that the upper leaves were segmented correctly through
smoothing and convex hull processing on the obtained edges. As for overlapping lower leaf, the uncovered area of the lower
leaf could be obtained by subtracting the upper blade segmented from the entire overlapping leaf. The complement of the
lower leaves could be obtained through inverting half of the lower leaf with the line formed by its tips as the symmetric axis.
【Result】The proposed algorithm can segment elliptical overlapping leaves in a complex background, with the average error rate below 3.0%. 【Conclusion】Aiming at the segmentation of overlapping elliptical leaves in complex background, the
greedy algorithm combined with bump detection and pit detection can segment target leaf more accurately, and the average error
rate is below 3%, scilicet, the segmentation accuracy is above 97%. |
|
查看/发表评论 下载PDF阅读器 |
