基于特征点的高分辨率卫星遥感影像自动配准方法

Method of feature points based automatic registration high resolution satellite remote-sensing image

在高分辨率卫星遥感影像配准中,由于数据量的迅猛增长以及分辨率的提升,使配准精度逐渐下降,配置工作越堆越多。为此,文中设计一种基于特征点的高分辨率卫星遥感影像自动配准方法。对高分辨率卫星遥感影像数据实施云层二值化、灰度拉伸这两种预处理。云层二值化的处理主要通过OSTU算法实现,灰度拉伸处理使用的方法是直方图均衡。基于影像特征点,通过SURF算法实施高分辨率卫星遥感影像的特征点提取。通过提取的特征点在遥感影像上构建多个小面元区域,利用小面元微分纠正的方式对2幅影像实施精确配准。为实验配置硬件环境与软件环境,使用2种影像数据测试设计方法的配准性能。测试结果表明:所设计方法对于白天非雪天拍摄的图像与白天雪天拍摄的图像的配准精度都很高,二者差异较小;对于相同场景、不同传感器拍摄的影像以及不同场景、同一传感器拍摄的影像,设计方法的配准精度均较高。说明设计方法能够克服不同因素的影响,实现较为精准的自动配准。

In the registration of high-resolution satellite remote-sensing images,due to the rapid growth of data volume and the improvement of resolution,the registration accuracy gradually decreases,and the configuration work becomes more and more. On this basis,a feature points based automatic registration method of high-resolution satellite remote-sensing images is designed. The two preprocessing methods of cloud binarization and gray stretching are conducted for high-resolution satellite remote-sensing image data. The cloud binarization is mainly realized by means of OSTU algorithm, and the histogram equalization is used for gray stretching. Based on the image feature points,the feature points of high-resolution satellite remotesensing images are extracted by means of SURF algorithm. The multiple small bin regions are constructed on the remote-sensing image by means of the extracted feature points,and the small bin differential correction can be used to accurate registration two images. The hardware environment and software environment are configured for the experiment,and two kinds of image data are used to test the registration performance of the design method. The testing results show that the design method has high registration accuracy for the images taken on non-snowy days and snowy days,and the difference between them is small. For the images taken by different sensors and the same scene and images taken by different scenes and the same sensor,the registration accuracy of the design method is high. It shows that the design method can overcome the influence of different factors and realize more accurate automatic registration.