Novel image registration method using edge correlation

By considering edge similarity of images sensed in different wave bands of the same landform, a novel image registration method using edge correlation as the similarity measure is proposed. Edge correlation involves spatial information of images, which reflects well their structures. We employ significant edge curves with long length, instead of isolated edge points, to calculate the correlation which makes the proposed method more robust to noise. Experimental results prove that, for images corresponding to the same landform, edge correlation nearly always yields a single sharp peak at the correct alignment parameters, and the proposed method has more accurate registration results than mutual-information-based methods, especially under noisy conditions.