Actual applications require specifi c methods to reduce the infl uence of the

correlation to distortions that appear when input images differ from their

reference model. When the distortion model can be described analytically

(such as in-plane rotation, scaling, etc.), specifi c methods give invariant

outputs to those particular distortions but reduces signifi cantly the fi eld of

practical applications. To overcome those limitations, we have fi rst proposed

the ASFS algorithm that allows to include a set of distorted images

in order to decrease the sensitivity of the correlation to nonanalytically

described distortions. Based on the ASFS algorithm, we have developed a

nonlinear two-step architecture