Wavelet-based image registration by using boundary features
Gao Cheng-min · Computer Engineering and Applications Journal · 2008
Putting pieces of images together into the original image is a valuable application, one key step is to search the matched edges.First,we transform all the edges between two corners into curve L1,which is rotationally invariant,then decomposite and reconstruct L1 by using wavelet packet analysis to get the edge curve L2 in different resolution.Then L2 are transformed by DCT(Discrete Cosine Transformation) into feature vectors.The similar pairs of edges on different pieces are selected by calculating the minimal Euclidean distance between feature vectors,then the boundary registration can be done and the original image can be restored from the pieces.This algorithm can find alike edges accurately,is rotationally invariant.