改进相位相关算法的小基高比影像亚像素匹配  

Sub-pixel matching for small baseline and height ratio stereo image pairs based on enhanced phase correlation algorithm

作　　者：李禄 范大昭[1] 耿则勋[1] 申二华[1] 

机构地区：[1]信息工程大学地理空间信息学院,郑州450052

出　　处：《测绘科学》2014年第7期118-121,共4页Science of Surveying and Mapping

基　　金：国家教育部博士后科研基金项目(20080440234)

摘　　要：相位相关技术在视频图像监测、遥感影像匹配和计算机视觉领域有着广泛的运用,本文在简要介绍相位相关基础上,提出了一种针对小基高比影像改进的相位相关算法。该算法先利用传统相位相关算法计算立体像对的整体相对视差,然后在参考影像和匹配影像中逐点分别做开窗处理,对每个窗口采用Hanning窗进行变换,再进行相位相关得到互功率谱函数,对该函数进行SVD分解和相位解缠得到行和列两个方向上的一维相位信息,对行和列相位进行处理并做最小二乘估计,即可得到逐点的亚像素级视差。当立体像对上所有点的视差计算完后,采用中值填充的方法,对匹配结果差的进行剔除和代换,得到所有点的亚像素级的视差。实验证明,本文提出的改进相位相关算法具有较高的匹配精度和成功率。Phase correlation is very popular for optical flow surveillance, remote sensing image matching and computational vision. This paper proposed an enhanced phase correlation algorithm for small baseline and height ratio stereo image pairs based on traditional algorithm. Integer of disparity from image pairs was computed by traditional algorithm, and each pixel both on reference image and matching image was made windows for Hanning transformation. When PC matrix was decomposed by SVD and unwrapped, both vector of line and column were identified and then estimated the disparity on sub-pixel level by least square fit. At last, a median shift propagation algorithm was used to refine all the matching results. The result of the experiment was proved that the algorithm would have a high ratio of success with high precision.

关 键 词：相位相关 亚像素 小基高比 HANNING窗 SVD分解 相位解缠 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]