自动获取初始视差是影像匹配自动化的关键问题。提出了分层Fourier-Mellin变换相位相关的影像匹配初始视差的全自动获取方法,无需人工干预和其他辅助信息,能高效、自动地获取影像中心点较精确的初始视差。首先对基准影像与待匹配影像构建金字塔影像,使用Fourier-Mellin变换与相位相关法求解金字塔顶层影像上基准影像与待匹配影像间的旋转、缩放与平移参数,然后利用旋转参数和缩放参数矫正待匹配影像的旋转与缩放变形,再利用平移参数计算待匹配影像中心点与基准影像上对应的同名点的视差近似值,最后在金字塔逐层影像上使用相位相关法对该对同名点的视差近似值进行误差校正,从而得到较精确的初始视差值。
Automatic determination of the initial parallax is a key technology of automatic image matching.The previous methods for determination of the initial parallax are analyzed and summarized in the paper first,and then a novel method,called layered FM phase correlation,is proposed.This method is based on Fourier-Mellin transform theory in combination with layering matching and phase correlation techniques.The method is capable of obtaining high-precision of initial parallax automatically and efficiently without human intervention and other auxiliary data.The validity of the method is conducted by experiments on the several typical image pairs.