针对数字表面模型(DSM)数据与可见光遥感图像信息融合的实际需求, 提出了一种基于一致点漂移算法(CPD)与相对相位直方图(RPH)的两级配准策略来实现上述数据与图像的自动配准。首先, 利用Canny算子提取图像边缘, 将边缘点作为CPD算法的输入, 实现两幅图像的粗匹配, 从而得到初始对应点集并估算尺度因子; 然后, 定义了一种鲁棒且具有旋转、平移不变性的区域变化信息描述子-RPH, 其在粗匹配结果的保障下还可以实现尺度不变性; 最后, 根据尺度因子在两幅图像中分别定义圆环模板, 并利用RPH测度完成DSM图像与可见光遥感图像精配准。实验结果显示, 使用RPH测度进行精配准后, 基于CPD算法的粗匹配结果得到了有效校正, 在数据自身存在透视失真情况下, 算法配准误差约为2 pixel, 能够满足DSM数据与遥感图像信息融合的需求。

According to the application requirement of information fusion for Digital Surface Model (DSM) data and visual remote sensing images, a two-level registration strategy based on Coherent Point Drift (CPD) algorithm and Relative Phase Histogram(RPH) was proposed to realize automatic registration of the data and images above. Firstly, Canny operator was used to extract edge points,and the extracted points were taken as inputs of CPT algorithm to implement a coarse matching and to obtain the initial matching points. Meanwhile, the scale factor was estimated. Then,the RPH, a robust descriptor characterized by invariance to image rotation and translation, was defined to represent the information of area changes. The RPH was also a scale-invariance under the support of coarse matching results. Finally, ring templates were defined in both images according to the scale factor and the fine registration of DSM data and remote sensing images was achieved by RPH measurement. Experimental results demonstrate that coarse matching results based on CPD are corrected efficiently after fine registration using RPH measurement. The registration error of proposed algorithm is only about two pixels even when images have perspective distortion. These data indicate that the method can satisfy the information fusion requirement of DSM data and visual remote sensing images.