R. Montoliu and F. Pla (Spain)
Motion Estimation, Image Registration, Generalized Least Squares Minimization.
In the literature of computer vision and image processing, motion estimation and image registration problems are usu ally formulated as parametric fitting problems, solved us ing least squares-based techniques. The assumption that the grey level of all the pixels of a region remains constant between two consecutive images (brightness constancy as sumption) can not be used directly using an ordinary least squares technique because its lack of linearity. The well known solution of this problem derives the optic flow equa tion as linearized function to be minimized. Nevertheless, it is possible to directly use the brightness constancy as sumption using a non-linear least squares-based estimator. The generalized least squares technique can be used in this context. In this paper two hierarchical least squares-based mo tion estimation algorithms are compared in order to demon strate that the use of a generalized least squares estima tor, and therefore the brightness constancy assumption, can produce more accurate results than the use of ordinary least squares-based estimator and the optic flow equation.
Important Links:
Go Back
