【作者】 张娜；

【导师】 白雪冰；

【作者基本信息】 东北林业大学 ， 林业工程自动化， 2013， 硕士

【摘要】 由于背景的复杂性,树木图像的分割相对比较困难,本文对树木图像的分割进行研究,从而为林业立体视觉测量、林木农药精确对靶施用、基于图像的树木可视化重建、树木的生长状态评估、树种自动识别与分类等课题提供基础数据和技术支撑。本文以静态彩色树木图像为研究样本,基于Matlab软件,分别使用三种方法实现树木图像的分割。基于RGB彩色空间,利用绿色植物过绿因子数值远远大于背景人造物过绿因子数值的特点,选取适当的过绿因子作为图像分割阈值。试验表明基于过绿因子的方法可以较好的实现树木图像分割。引入了马尔可夫(Markov)随机场理论,首先对树木彩色图像建立MRF-MAP分割模型,并用EM算法对观测场模型的参数进行估计,然后采用条件迭代算法和模拟退火算法对分割问题进行优化求解,最终确定了彩色树木图像分割的类别数和试验参数,并将得到的分割结果进行比较。在上述研究的基础上,研究初始标记场对MRF-MAP模型分割结果的影响,分别采用过绿特征和K-均值分割作为初始标记场,并研究了惩罚因子设定方式对MRF-MAP模型分割结果的影响,其惩罚因子分别采用固定不变和随程序运行不断增加的方式,然后将其得到的分割结果进行对比。最后将过绿特征的阈值分割结果作为初分割参数,结合马尔可夫随机场理论,将过绿马尔科夫随机的分割结果和马尔科夫随机场的分割结果进行比较。试验结果表明：无论基于马尔可夫(Markov)随机场的分割法还是基于过绿马尔可夫(Markov)随机场的分割法,都可应用在树木图像分割上。更多还原

【Abstract】 Relatively difficult due to the complexity of the background, trees image segmentation, image segmentation of the trees, so that stereo vision measurement for forestry, forest pesticides precisely target application visualization reconstruction, image-based trees, tree growth statusassessment of tree species in the automatic identification and classification of topics such as basic data and technical support. Static color images of trees as samples, respectively, using three methods trees image segmentation based on Matlab software.Based on the RGB color space, the use of green plants over the green factor value is far greater than the background creation of green factor values, select the appropriate over the green factor as image segmentation threshold. Test that green factor method can achieve trees image segmentation.Introduced Markov random field theory, first established trees color image segmentation model of MRF-MAP, and using EM algorithm to estimate the parameters of the model of the observation field, and then using conditional iterative algorithm and simulated annealing algorithm to optimize the segmentation problem solving, and ultimately determine the number of categories and test parameters of the color tree image segmentation, and the segmentation results were compared.On the basis of these studies, research the initial mark field MRF-MAP model segmentation results, with green features and K-means segmentation as initial mark field, and the penalty factor setting mode MRF-MAPthe model segmentation results, the penalty factor were fixed and with the program to run continuously in an increasing manner, and then comparing the obtained segmentation results. Finally, over the green features threshold segmentation results as the initial segmentation parameters, combined with Markov random segmentation results over the Green Markov random field theory and Markov random field segmentation results compare.The results showed that: both based on Markov random field segmentation method is based on Green Markov with airport segmentation method can be applied on the tree image segmentation更多还原