基于YCbCr颜色空间和Fisher判别分析的棉花图像分割研究

Image Segmentation of Cotton Based on YCbCcr Color Space and Fisher Discrimination Analysis

棉花的分割是采棉机器人研究的关键技术。本文分别在HSV、HIS和YCbCr颜色空间下,首先根据棉花的颜色信息与背景颜色信息的差距,对样本图像中的各个对象(棉絮、棉枝、土壤等)分类;其次根据分类结果分别提取各类在各颜色空间下的样本像素值;再根据类间离散度最大和类内离散度最小的准则计算出Fisher判别向量和各类的质心;最后按照像素值离各质心最近的准则进行图像分割。结果表明,在YCbCr颜色空间下产生的分割噪声最小,选取此颜色空间,采用贴标签的方法自适应去噪。实验仿真表明,本方法可有效避免阳光直射和阴影的干扰,对各种情况都能准确分割,分割准确率达90.44%。

For cotton harvesting robot,the cotton image segmentation is one of the key technologies.In this paper,under HSV,HIS,and YCbCr color spaces respectively,according to the difference between cotton color and background color,the various objects（cotton batting,cotton branches,soil etc.） in the sample images were classified,and then the pixel value of every category in different samples was extracted based on the classification result.In the following,the rule that the dispersion is biggest between different classes and smallest within the same class was used to calculate the Fisher discrimination vector and the center of mass in every class.Finally,image segmentation was carried out based on the criterion of pixel value close to the center of mass.The result showed that the least segmentation noise was obtained in the YCbCr color space,in which the method of labeling for self-adapting denoising was need.The simulation showed that the cotton could be separated exactly from the background by the above algorithm whether the cotton was exposed to the sunlight or the shadow.A total of that 136 cotton images were segmented with an accuracy of 90.44% in YCbCr color space.