Neural computation methods for the point correspondence problem
Hiroshi Sakou, Hadar I. Avi-Itzhak · 1991
Summary form only given. Three neural computation methods for helping to overcome the point correspondence problem in the computer vision field are discussed. The first is for two-dimensional correspondence between a model's points and the input points assumed to have been transformed from the model's points by an unknown affine transformation. The second is for correspondence between the model points on a three-dimensional object and the input points perspectively projected on a two-dimensional plane from the model points after an unknown motion of the object. The third includes a Boltzmann machine.>