基于特征学习的建筑物自动识别算法研究

针对单幅航拍灰度图像的建筑物自动识别问题,给出了一种基于类特征学习的算法结构.利用建筑物顶部的高亮度特性,采用基于灰度的OTSU方法实现图像的粗略分割;然后采用二阶高斯马尔可夫随机场（GMRF）模型描述已分割图像,得到目标/背景类的6维特征向量;以此特征向量及粗分割的类标记为训练集,训练支持向量机（SVMs）分类器学习两类特征,再用训练过的分类器重新分割图像;最后采用基于先验知识的规则对分割区域进行建筑物/非建筑物判定;理论分析和实验结果表明了算法的有效性.

Learning Based Extraction of Building Objects from High Resolution Satellite Images

A learning based building detection scheme is presented, the process start with a two-dimension OTSU segmentation, then the segmented regions are modeled as second order GMRF model, the model parameters are estimated and used to train a SVM classifier, the trained SVM classifier segmenting the image in a more accurate result, finally, the system output is verified on priori-information. Theoretic analysis and experimental results verify the effectiveness of the scheme.