Abstract:

The core issue of image registration is to increase the speed and accuracy of registration, but the result of the registration much depends on the matching precision of features.In order to increase the matching precision of features, an algorithm based on the twodimensional Gabor wavelet transform for corners matching is proposed in this paper. Firstly,it adopts an improved Harris corner detector to extract corners and obtains the coordinates of corner positions,and utilizes multi masks of twodimensional Gabor wavelet to filter the reference image and unregistered image to extract the duplicate Gabor wavelet coefficient of corner coordinates from filtering the image, and takes this as the feature description of corners .Secondly it introducs two kinds of similarity measurement to match corners.Adopting the longest common subsequence is successful to extract more points of the same name and has a higher match rate ,which has been proved by multiexperiments when parameters are chosen appropriately.

Key words: image registration;corner detected;twodimensional wavelet transform;dynamic programming;longest common subsequence