Motion Estimation in Image Sequences

A Frequency Domain Technique Based on Energy Radial Projections for Robust Estimation of Global 2D Affine Transformations

“…An important characteristic of this class of algorithms is that the use of whole image areas, rather than of sets of features, makes them more insensitive to additive noise and generally more accurate than the feature-based ones. The other side of the picture is that, unless one uses direct noniterative methods [45] or efficient implementations [42], the featureless methods are usually more computationally intensive because huge amounts of data enter iterative minimization algorithms; besides, these algorithms need to be properly initialized in order to converge to the global solution and not to fall into local minima.…”

“…Since our algorithm utilizes the whole image information content through the Fourier transform, a natural comparison can be established with featureless techniques for estimating global affine transformations [1,41] or projective transformations [42][43][44][45][46][47][48], which subsume the first ones as a particular case. An important characteristic of this class of algorithms is that the use of whole image areas, rather than of sets of features, makes them more insensitive to additive noise and generally more accurate than the feature-based ones.…”