W. Jiang and M. Zhou (PRC)
Image Processing and Analysis, Linecorrespondence, Image registration, Hough transform
This paper presents a new image registration approach based on multiple line correspondences, which is especially efficient for the registration with rotation. A modified local Hough transform (MLHT) is proposed to look for line correspondences. All line correspondences are analyzed to get the image rotation angle and the rotation center, then the translation. And the most remarkable characteristic of this approach is that the rotation and translation of image registration are uncoupled and can be determined independently. This method overcomes the uncertainty caused by some other ways such as point-point, point-line correspondence and improves the speed of the computation rapidly. The comparison of results of different edge detectors shows that this method is insensitive to them. It can be used in practice based on PC with pure software support, especially for image sequence analysis, such as background registration, panoramic image creation, image stabilization etc.
Important Links:
Go Back
