多特征匹配的高光谱图像配准方法

Hyper-spectral image registration method based on multi-feature matching

基于区域的配准方法从处于离散性图像坐标点中,检索配准特征点完成高光谱图像配准,而离散性图像坐标点导致图像特征点提取存在一定误差,导致配准后高光谱图像结构特征缺失。为此,提出多特征匹配的高光谱图像配准方法,首先使用基于Forstner算子的高光谱图像特征提取方法,提取两幅待配准高光谱图像中目标特征,然后基于提取的特征,采用基于多特征匹配的高光谱图像配准方法实现配准。经实验结果验证,该方法提取的高光谱图像特征完整度系数最大值为0.9899,配准后的同类型、不同类型的多特征高光谱图像结构相似度指数始终大于0.9800,且与同类配准方法相比,配准速度提高了6倍与8倍左右,应用性能较为显著。

The region-based registration method retrieves the registration feature points from the coordinate points of the discrete image to complete hyper-spectral image registration.However,the coordinate points of the discrete image lead to certain errors in the extraction of image feature points,resulting in the loss of structural features of hyperspectral image after registration.For this reason,a multi-feature matching hyper-spectral image registration method is proposed.Firstly,the feature extraction method of hyper-spectral image based on Forstner operator is used to extract target features of two hyper-spectral images to be registered.Then,based on extracted features,the multi-feature matching based hyper-spectral image registration method is adopted to realize the registration.Verified by the experimental results,the maximum value of the hyper-spectral image feature integrity coefficient extracted by this method is 0.9899.The structure similarity index of multi-feature hyper-spectral image of the same type and the different types after registration is always greater than 0.9800,and compared with the similar registration method,the registration speed is improved by about 6 times and 8 times,which has significant application performance.