基于不一致性法的最优分割参数评价

Optimal Segmentation Parameters Evaluation Based on Discrepancy Measure

影像分割是基于对象分析方法的前提,分割结果的质量决定了后续分类的精度,分割参数的选择影响到分割结果的好坏。基于不一致性度量方法,考虑几何不一致性和算术不一致性,构建了新的最优分割参数评价指标FIX,实现最优分割参数的选择。选取高空间分辨率IKONOS影像上2种不同地物(池塘、农田)进行实验研究,表明不同地物具有不同的最优分割参数,为最优分割参数评价方法提供一种思路。

Image segmentation is the basis for object-based image analysis. The quality of segmented objects directly affect the accuracy of the subsequent classification. Meanwhile, the choice of image segmentation parameters will directly determine the segmentation results. The paper manages to construct a new optimal segmentation assessment index FIX based ondiscrepancy measure, considering the geometrical discrepancy and algorithmic discrepancy. Two different categories of land cover（pond, farmland） on the high-resolution IKONOS image are studied. The ex- perlment demonstrates that different categories of land cover have different optimal segmentation parameters and it provides a research method in this field.