|
The transformation relating two different views of a two- or three-dimensional object scene is derived as a function of the displacement of the camera, relative to the scene, between the taking of the two views. It is shown how this transformation can be used to place the two views into exact area registration, as required for interframe processing operations such as frame addition, frame subtraction, frame comparison, etc. It is also shown, for the case of a two-dimensional object scene, how the transformation relating the two views can be derived from the images themselves, without knowledge of the camera displacement. A practical example of this action is given, illustrating automatic exact area registration of the different views. The algorithm used is based upon a generalization of classical cross correlation in which the dimensionality of the correlation surface is extended from two to eight, corresponding to the eight degrees of freedom existing between any two views of a common object scene.
|
