基于LBP滤波和ACWE的葡萄病害图像分割方法

Image segmentation method for grape disease based on LBP filter and active contour without edges

针对葡萄病害的特征在不同发病期的多变性,提出了一种新的无监督葡萄病害图像分割方法。该方法使用局部二进制模式(LBP)算子描述葡萄叶纹理特征,采用正常叶片训练出的LBP直方图作为滤波模板,在LBP域中滤除葡萄叶片表面纹理,增强了病害区域与其它区域的对比度,再通过无边界主动轮廓算法分割病害区域。在RGB、HSV和L*a*b*等三种彩色空间中对9种分量下的葡萄病害进行分割的实验,发现在G、V、L*、a*等4个分量下,本方法对于不同发病期的葡萄叶斑病的分割准确率较高,且对不同光线影响有较好的鲁棒性。实验还表明,与其它5种基于主动轮廓算法的对比分割方法相比,本方法具有更高的准确率。

Aiming at the variability of grape disease characteristics in different progression,a novel unsupervised segmentation method was presented,in which local binary pattern（LBP） was used to describe the grape leaves textures.The histogram of LBP from normal grape leaves was used as filter template.The contrasts between the disease regions and the others were strengthened by filtering the leaf veins in LBP domain.Finally,the algorithm of active contour without edges was employed to segment the disease regions.The segmentation tests of the grape disease under 9 channels among RGB,HSV and L^＊a^＊b^＊ color spaces were taken.（1） on G,V,L^＊ and a^＊ channels,the proposed method has high accuracy for segmentation of leaf spots in different progress and is robust to the changing illumination,（2） the segmentation accuracy outperforms other five comparative methods based on active contour algorithm.