基于形态学的黑背景下收获前棉花图像自动分割技术研究

Probe on the Morphology-Based Auto-Segmentation Technique for the Preharvest Cotton Images photoed on the Dark Background

以棉田实地采集的黑背景下收获前含铃壳棉花图像为样本(7级×9张/级),首先,对原始灰度图二值化以提取中心模板,进一步进行形态学高帽剪切预处理并提取其外围二值模板,两者叠加完成背景分割;试验结果表明:98.4%的含铃壳棉花二值模板图像被准确提取。其次,用形态学低帽剪切预处理后的二值图执行开启运算,以判别棉瓣间的连通性:⑴对于多连通的低品级棉花,直接提取未开启二值图中不超过4个区域的棉瓣二值模板,⑵对于单连通的高品级棉花,重新对原始灰度图二值化并提取最大区域的棉瓣二值模板;实验证明:基于棉瓣面积明显大于铃壳面积的棉瓣抽取算法是健壮的,74.6%的棉瓣二值模板图像被准确提取。

In this experiment, 63 sheets of photos, viz. 7 grades and 9 sheets per grade , of the preharvest cotton image with natural shape photoed on the dark background afield were selected as samplings. The cotton binary template with bracteoles can be extracted by the followed steps： （1）The original gray image was transformed to the binary image with the min-hreshold in the left side of trough span of its histogram, and the center binary template can be extracted, which may lose some bracteoles in the border of cotton. （2）By dint of tophat cutting preprocess, the bracteoles in the border of cotton became more bright. The enhanced gray image was transformed to the binary image with the Otsu＇s threshold, and the periphery binary template can be extracted. （3） The center binary template plus periphery binary template is the preharvest cotton binary template with bracteoles namely. The experimental result indicates that the preharvest cotton binary template with bracteoles reached high precision, viz. 98.4% The segmentation arithmetic between cotton and its bracteoles consists of three aspects followed： （1）As a result of the bothat preprocessing, the intensity troughs between cotton petals was found and the periphery bracteoles region showed a granule appearance using a small structuring element. The enhanced gray image was transformed to the binary image with the Otsu＇s threshold, and the area of every bracteole is far smaller than the area of each cotton petal, which is the key technology in this experiment. （2）By dint of opening operation using a big self-adaptive structuring element, the connectivity statistics of the cotton image, viz. the number of objects, can be obtained in the opened binary image to judge the grade of preharvest cotton. （3）To the lower grade of preharvest cotton with multipleconnectivty, cotton petals not exceeding four was extracted from the non-opened binary image in turn according to the above-mentioned key technology. To the higher grade of preharvest cotton with single-connectivity, the original gray image was transformed to the binary image with the Otsu＇s threshold renewedly to avoid losing area, and the max connective region was extracted from the newly binary image. The experimental result indicated that the better precision, viz. 74.6%, was obtained in the course of extracting the cotton binary template, and the arithmetic used for extracting cotton petals is robust.