Aiming at the optical image affine transformation of the automatic registration, a coarse-to-fine remote sensing image automatic registration algorithm was proposed. Firstly, the input images were mapped for a local feature vector sets with translation, scaling and rotation invariant characteristic based on SIFT feature. According to the euclidean distance of the feature vector which is taken as the similarity decision measure, the initial matching feature points and the initial model parameter values of the transformation were determined. Secondly, making the mutual information as similarity measure, more established correspondence feature points were obtained based on the position control of the search strategy. Thirdly, the deeply optimized affine transformation model parameters were obtained by using the control points and the weighted least squares optimization algorithm together. In this way, a more sophisticated image registration was completed, and a root mean square error was used to evaluate the result of the registration. Finally joint histogram was taken as the registration precision evaluation standard to test registration effect. Experimental results demonstrate the effectiveness and accuracy of the proposed method.%针对仿射变换的光学图像自动配准精度不高的问题,提出了一种基于特征的由粗到细的遥感图像自动配准算法.首先采用SIFT特征进行了特征点的粗匹配,将输入图像映射为一个具有平移、缩放、旋转不变性的局部特征向量集,采用特征向量的欧氏距离作为相似性判定度量,通过两两比较找出匹配的若干对特征点对作为初始配准点对,以完成输入图像的粗匹配；其次,以互信息作为相似性测度,基于位置控制的搜索策略,确定了更多的特征点的对应关系；然后,利用控制点结合加权最小二乘优化仿射变换的模型参数,完成了图像间的精细配准；最后引入了联合直方图,以其作为配准精度的评价标准来检验配准效果.研究结果表明,该算法对于高光谱遥感图像具有较高的配准精度,速度快、可靠性高.
展开▼
