Image recognition using generalized correlation

This paper investigates the use of generalized cross-correlation in pattern matching when the objects may be of one or two dimensions. Generalized correlation can be used to determine the amount of dilatation and rotation between a given template and an object, in addition to determining the relative translation. Two techniques are discussed which break this four-dimensional correlation into two two-dimensional correlations making the problem computationally feasible. The techniques were developed for a specific class of images, however they can be applied to a more general class.