Jignesh N. Sarvaiya, Suprava Patnaik, and Alpesh Dobariya
Image Registration, Active Contour, Radon Transform, Control Point
This paper describes an interactive feature extraction approach, used for image registration. Remotely sensed aerial images are often featured with compositions which remain unchanged over a period of time. Structures like tall buildings, bridges, water reservoirs occupy the same relative position with almost same shape and can easily be indexed. In two images, with different orientation and depth of view one can easily index rough boundary of at least three such features. The center of these features, with sub-pixel accuracy can be utilized as control points for deriving the transformation matrix, essential for the registration of the two images. Image registration accuracy, to a great extent depends on precision control point selection. Though mining control points manually is very tedious and does not guaranty required accuracy, smearing rough boundary is an easy and almost effortless job. In this work we have proposed active contour deformation approach to be implemented on rough feature shapes, followed by center extraction to acquire precised control points. Radon Transform is then used to estimate rotation and scaling. Using this method, we have achieved accurate registration of image pairs captured with wide affine deformation. The result is compared with a state-of art algorithm to justify the perfection achieved in registration, with reduced complexity.
Important Links:
Go Back
