基于高斯拟合的高光谱影像配准算法

Registration algorithm for hyperspectral image based on Gaussian fitting

传统的基于区域的配准方法是搜索配准控制点在离散的图像坐标点上进行的,从而限制了配准控制点定位精度这一问题,所以文中提出了一种基于高斯拟合的高光谱影像配准算法。与传统基于区域的配准方法类似,该方法是利用图像灰度信息,建立两幅图像之间的相似性度量,搜索使相似性度量值最大或最小的点作为配准控制点,但与传统方法不同之处在于,在搜索过程中,并不是直接寻找极值点作为配准控制点,而是通过在搜索过程中,首先生成相似度矩阵,利用极值点附近的值求出高斯拟合函数系数,利用高斯函数的极值点作为配准控制点。在对多组Hyperion高光谱影像进行配准的实验中,精度均优于传统方法,达到了亚像素级,满足后续的融合、变化检测等需要。

The traditional registration method is based on the search area registration and it is carried out at control points of the image coordinates of discrete points, but this method will limit the positioning accuracy of the registration control point. Aiming at this problem, a high spectral registration method which is based on Gaussian fitting was preseuted. Similar to the traditional registration method based on region, this method also used the gray information of images to build the similarity measure between two images and searched the point at which the similarity measure can reach its maximum or minimum to be the registration control points. Different with the traditional methods, it did not go straight for the extreme point and used it as the registration control point during the process of search, instead, the similarity-matrix was produced at first during the process of search and coefficients of Gaussian fitting function could be obtained from the value near the extreme points, the extreme points of Gaussian fitting function were used as the registration control points to complete the registration. The multiple sets of experimental results of hyperion high spectral registration all show that the method presented in the paper is more accurate than the traditional methods, and the registration accuracy reaches sub-pixel successfully, the method can meet the follow-up demands such as fusion, change detection and so on.