结合形状信息的SAR图像特征匹配方法

合成孔径雷达（Synthetic Aperture Radar, SAR）图像匹配是实现图像配准和定位导航等任务的重要途径之一。本文研究基于局部特征的SAR图像匹配，针对SAR图像特征匹配中错配率高的问题，引入邻域共识思想，结合以往被忽略的特征形状信息，提出一种新颖的误匹配滤除方法。所提方法主要包括形状信息引导的邻域构建和共识度量两个主要步骤。在邻域构建步骤中，本文利用形状信息确定一个仿射不变图像区域，从而构建对几何稳健的邻域。在共识度量步骤，利用形状信息获取局部变换矩阵，然后利用局部变换矩阵对邻域进行重投影，最后根据重投影误差评估匹配的共识度，并以此判断匹配是否应该被滤除。为了验证算法有效性，本文利用21对L,X和Ku三个波段下多角度机载SAR和多分辨率星载SAR图像与五个先进方法进行了对比实验。结果表明所提方法在精度、召回率和F1值上都优于所有对比算法，尤其对多分辨率数据最为有效。另外，通过搭配三种常用的特征进行实验，本文发现所提方法对特征类型不敏感，其中在SIFT特征上效果最为突出。但是，所提算法在效率上要略低于其他同类算法。

SAR Image Feature Matching Method Combined with Shape Information

? ?Synthetic Aperture Radar （SAR） image matching is the basis for tasks such as image registration and image-based localization. This paper studied the feature-based SAR image matching problem. To address the high mismatch ratio of SAR features， a novel neighborhood consensus method was developed with combination of the feature shape information. There are mainly two steps: neighborhood construction and consensus measurement， which are all based on the shape information. In the neighborhood building step， we exploited the shape information to build an affine-invariant neighborhood. In the consensus measurement step， the local transformation matrix， calculated from the shape information was used to re-project the neighbors to evaluate the geometric consensus. In the experiment， the proposed method was compared to five popular methods. Twenty-one pairs of multi-angle airborne SAR and multi-resolution spaceborne SAR images were used as the test data. The results showed that the proposed method considerably outperformed all other methods in precision， recall and F1-measure， especially for multi-resolution data. In addition， by experimenting with three common local features， this paper found that the proposed method was widely effective for different features. However， the efficiency of the proposed algorithm was relatively lower than other methods.