基于显著性检测的黄瓜叶部病害图像分割算法

Segmentation Algorithm of Cucumber Leaf Disease Image Based on Saliency Detection

摘要:

针对复杂背景下黄瓜叶部病害分割精度不高的问题，提出了一种基于显著性检测的黄瓜叶部病害图像分割算法。首先利用超像素将黄瓜图像分块，获取黄瓜叶片的边缘，并提出了一种超像素间权重计算方法和显著种子选取方法；然后通过流形排序计算显著图，对得到的显著图进行阈值分割，得到二值图像；再将二值图像与原图像进行掩码运算，得到黄瓜病害叶片；最后利用超绿特征和数学形态学对病害叶片进行分割得到病斑。对常见的黄瓜病害(白粉病、褐斑病、霜霉病、炭疽病)图像进行测试，结果表明该算法与Otsu算法和k-means算法相比，有效解决了冗余分割问题，错分率均在5%以内，算法平均执行时间均小于4.000ms，分割效果更加精确，为后续构建黄瓜病害自动识别系统奠定了基础。

Abstract:

In order to solve the problems of low accuracy of cucumber leaf disease image segmentation in complex background, a new segmentation algorithm of cucumber leaf disease image based on saliency detection (SCLDSD) was proposed. The proposed algorithm mainly consists of two parts: saliency detection in cucumber disease image which is used to get the leaf extraction and image segmentation which is used to get cucumber leaf disease. The algorithm first used the superpixel segmentation method to divide the cucumber image into blocks, got the edge of cucumber leaf preferably, and proposed a new method to calculate the weights among different superpixels. Then the algorithm used Harris points and convex hull to select saliency seeds. After using manifold ranking to compute the saliency map, the threshold segmentation was adopted on the obtained saliency map to get the binary map. At last, the cucumber disease leaf and background of the original image were separated by adding the binary map to the original image. In order to obtain the disease parts, ExG was used to expand the disparity of green parts and lesion parts and then threshold was used to carry out the segmentation. Finally, the morphological operation was processed in order to obtain fuller lesion. The proposed algorithm was tested on common cucumber disease images. The experimental result shows that the algorithm effectively solves the redundant segmentation and its more accurate with the error rate less than 5% and the average execution time of the algorithm less than 4.000ms in segmentation. From the results it can be concluded that the algorithm verifies the feasibility and practicality of the saliency detection algorithm in processing of disease images. Meanwhile it lays the foundation for the subsequent establishment of the automatic identification system of cucumber disease.