用于遥感图像建筑物目标分类的层次匹配核

摘要

This paper proposes a kernel function - hierarchical log-polar matching kernel which making use of the feature spatial information for building classification in remote sensing images. It extracts image local features, uses traditional clustering methods to quantize all feature vectors into several different types, and then partitions the image into multi-level increasingly fine log-polar "sub-regions (bins)". By computing histograms of local features found inside each sub-region in each level, the weighted multi-scale histograms are formulated. By summing all weighted multi-level histograms of each feature vector, the final hierarchical log-polar kernel is established. The building classification is done with a support vector machine (SVM) trained by using the "one-versus-all" rule. The experimental results on Caltech-256 database and real remote sensing images demonstrate the efficiency and effectiveness of the proposed kernel.%提出了一种利用图像特征空间信息的核函数——层次对数极坐标匹配核，用于遥感图像建筑物目标的分类。对图像进行特征提取，并将特征映射到已聚类好的“码本”中，量化为有限个类别。将图像由粗到细划分为多个层次的对数极坐标系下的“子区域(单元格)”。比对落入同一层次、同一“子区域(单元格)”的每类特征的直方图交集，建立加权的多尺度直方图，将多个特征多尺度直方图合并，得到最终的核函数，并利用“一对多”的支持向量机(support vector machine，SVM)完成建筑物的分类。对标准数据库Caltech-256和自建遥感图像数据集进行实验，结果证明了该核函数的有效性。