基于SURF和矩阵乘法的超大规模遥感图像亚像素配准算法研究

摘要

图像配准是一项基本而又非常关键的图像预处理技术.在很多应用领域,要求配准精度达到亚像素级.现有的相位相关法具有精度高、计算简单等特点,但是随着图像规模的增大,计算付出的时间代价是巨大的.本文提出基于SURF和矩阵乘法相位相关法的超大规模遥感图像亚像素配准算法,采用化整为零的方法,首先把整幅图像划分成不同区域,其次使用改进的Canny算法进行边缘分割,去除无用信息,再次使用SURF算法提取特征,最后在关键点周围使用矩阵乘法相位相关估计图像亚像素偏移量.实验表明本文提出的算法不仅提高了算法运行速度,同时也解决了图像尺寸太大导致一般计算机无法处理的问题.并且由于矩阵乘法相位相关的良好抗噪声特性,因此即使存在噪声,算法仍然可以获得较高的亚像素偏移量估计精度.%Image registration is a basic and important image preprocessing technology.In many applications,it is needed to extend the image registration to sub-pixel level.The existing phase correlation algorithm has advantages of easy calculation and high accuracy;however with the increasing scale of the image,the time cost of calculation is huge.This paper proposes a sub-pixel registration algorithm of very large scale remote sensing image based on SURF and matrix multiplication phase correlation.This algorithm adopts the method "breaking up the whole into p-arts".First,the whole picture is divided into different regions.Secondly,the improved Canny algorithm is used to segment the image,removing the useless information.Thirdly,the feature is extracted by the SURF algorithm again.Finally,the sub-pixel offset is calculated by the matrix multiplication phase correlation around the feature points.According to the experiment,the proposed algorithm can not only improve the operating speed,but also solve the problem caused by general computer cannot handle so large pictures.At the same time,due to the good anti-noise characteristic of matrix multiplication phase correlation,even if there is noise,the algorithm can still get a high-accuracy estimation of sub-pixel offset.