Image registration has a wide application in dynamic target detection, disaster forecast and medical diagnosis. However,traditional feature-based image registration algorithms extract a larg e number of key points with dense distribution, especially for remote sensing image registration.This drawback causes low effic iency and precision in the feature matching procedure as well as a high probability of mismatching.In this paper,an image registration algorithm based on elliptic Fourier descriptor (EFD) is proposed.Since the elliptic Fourier descriptor has a good ability in retaining the characteristics of shape information,it can be applied to describe and match the boundaries of closed re gions.Thus the number of feature points required for image region matching can be dramatically reduced.The proposed alg orithm further uses the matched boundaries to estimate the transformation parameters,and specifies the range of feature searching.Because of the reduction of feature points and the shrinkage of searching range,the efficiency of search ing matched features in the feature domain will be greatly improved.Numerical results also indicate that the proposed algorithm can effectively inhibit the mismatching,and improve the efficiency of feature matching.The efficiency improvement is more o bvious for large size images with rich texture,which assesses the applicability of the proposed algorithm for remote s ensing image registration.