基于改进OS-SIFT的可见光与SAR图像自动配准

Automatic Registration of Optical and SAR Images Based on Improved OS-SIFT

针对尺度不变特征变换(SIFT)算法配准可见光和合成孔径雷达(SAR)图像时性能较差的问题,提出了一种基于改进光学-SAR图像的SIFT(OS-SIFT)可见光和SAR图像配准算法。首先,利用非线性扩散滤波构建可见光和SAR图像的非线性扩散尺度空间,并采用多尺度Sobel算子和多尺度指数加权均值比算子分别计算可见光和SAR图像的一致性梯度信息。然后,用图像分块策略剔除尺度空间第一层后对尺度空间进行分块,在一致性梯度信息的基础上提取Harris特征点,得到稳定且均匀的点特征。基于梯度位置和方向直方图模板构建描述符并对其进行归一化处理,以克服影像间的非线性辐射差异。最后,利用欧氏距离进行特征匹配,并采用快速抽样一致性算法剔除误匹配。实验结果表明,相比联合位置、尺度和方向的SIFT算法和OS-SIFT算法,本算法的匹配率有明显提高,均方根误差也相对较低。

Aiming at the problem of poor performance of the scale-invariant feature transform algorithm when registering optical and synthetic aperture radar images,this paper proposes an improved optical and SAR scaleinvariant feature transform based on registration algorithm for optical and SAR images.First,the nonlinear diffusion filter is used to create the nonlinear diffusion scale space of optical and SAR images,and the multiscale Sobel operator and the ratio of exponentially weighted averages operator are used to compute the consistent gradient information of optical and SAR images,respectively.Then,the image block strategy is adopted,the scale space is divided into blocks after skipping the first layer of the scale space,and Harris feature points are extracted based on consistent gradient information to obtain stable and uniform point features.To overcome the nonlinear radiation difference between the images,the gradient location and orientation histogram descriptor template are used to build the descriptor.Finally,for feature matching,the Euclidean distance is used and the fast sample consensus algorithm is used to eliminate mismatches.The experimental results show that compared with the scale-invariant feature transformation algorithm combining position,scale,and direction and the OS-SIFT algorithms,the algorithm’s matching rate is considerably improved,and the root mean square error is relatively low.