Aim at automatic registration of temporal images

multispectral images and multisensor optical images

a feature-based automatic image registration algorithm is described in this paper.On the basis of point mapping techniques

it deals with the registration of images with affine geometric distortion.First

with edge detection and a postprocess

the closed boundary regions in both images are extracted.Next

the correspondence of closed boundary regions is developed by chain-code correlation and invariant moments.And the center of gravity in the correspondent regions is used as matching points.After a global consistency check is applied to the matching points

the primitive control points are acquired.Finally

initial transformation parameters are estimated based on the primitive control points.The number of control points is added by salient point chip correlation matching in the reference image and the transformed image.An iterative refined parameter process is devised to improve the accuracy of registration.Experimental results with various kinds of remote sensing images have verified high accuracy and reliability this algorithm.