A wavelet-based multisensor image registration algorithm

Multisensor images have different gray level characteristics while contours representing region boundaries are preserved in most cases. A contour-based registration algorithm which uses region boundaries and other well-defined edges as matching primitives is presented. This algorithm integrates the invariant moments with the orientation function of the contours to establish the correspondences of the contours in the two images, e.g. CCD and infrared (IR). The improved wavelet-based fuzzy contour extraction is developed. For the feature matching, an orientation function matching is proposed and integrated to the invariant moments to correspond the extracted contours. The contours with at least two salient points and others with only one salient point are matched separately. The correspondence salient points are used as control-point pairs, and the transformation parameters are estimated based on them. This approach implements automatically and works well for the image pairs in which contour information is well preserved. Experimental results for the CCD and IR images are used to test and verify the theory and algorithm presented in this in this paper.