Finding transformation functions, which warps images with nonlinear local deformations, is the most challenging tasks. In transformations like thin plate spline (TPS), multiquadrics (MQ) and weighted means (WM), errors in the correspondence or local image deformations spread over the entire resampled image. These transformation functions may map control points in the sensed image to the corresponding control points in the reference image accurately, but they warp the target image too much, causing large errors in the mapping away from the control points .In this paper a transformation function is presented to deal the local deformations present in the images, which is simple to implement. The results exhibit that the effects of control points are limited to a small neighborhood of the control points, the algorithms keeps inaccuracies in the correspondences local without spreading them to the entire image. Moreover the effect of the deformation decreases as we move away from the control point.
A.M. Siddiqi, M. Saleem, A. Masud
2007 International Conference on Electrical Engineering