基于影像分割的树木特征点提取方法研究

The Method to Extract Feature Point for Trees Based on Image Segmentation

树木处于复杂的自然环境中,加之其本身的不规则性,使得影响树木影像检测和识别质量的参数难以确定.采用一种基于集群聚类的方法对树木影像进行影像分割,在聚类中心的定义方式上不是简单地以各类的灰度重心作为聚类中心,而是采用了一种最大—最小距离法进行聚类中心的动态选择.在聚类过程中除了合并过程还加入了类别分裂处理,对每一次迭代过程中标准差最大的类别分裂成新类,并将新类中的像元重新加入到距离判别过程中,成功地将树木从背景影像中分离出来.在此基础上选择合适的特征提取算法,分别对原始影像和分割结果进行特征点的提取,对提取结果进行比较,得到了更好的特征点提取结果,为后续的树木影像匹配等研究工作提供可靠的数据基元.

Trees is still in a complex natural environment, coupled with the irregularity of its own, mak- ing it difficult to determine parameters which affect quality of detection and identification of trees image. This paper used a kind of clustering method for trees image segmentation. For defining the cluster cen- ters, it is not simply with the center of gravity of gray on the various types as the cluster center, but uses a maximum and minimum distance method to select dynamically. And in addition to the merger in the clustering process, the classes split is joined, which is a process to split the class with the largest standard deviation into new classes in each step of iteration and to set the pixels of the new classes into the process of identification for distance. This separated the trees successfully from the background image. On this basis, it selected the appropriate feature extraction algorithm to extract feature points from the o- riginal image and segmentation results respectively, and to compare the extraction results for a better result of feature point extraction. This will provide a reliable data base element for follow-up of the image matching for trees and other researches.