Implementation of generalised cross-correlation with large changes in parameters using genetic algorithms

This paper is concerned with the implementation of a new image processing technique called Generalised Cross-Correlation designed to detect a two-dimensional displacement field between consecutive pictures of some object. The particular problem considered here is the determination of a flow field from experimental tank data. The method consists of maximizing a Generalised Cross-Correlation function in order to compute the parameters in a global mathematical description of the displacement field. This is in complete contrast to traditional approaches such as Cellular Correlation, where an array of constant vectors are estimated using spatially local data. The focus here is on the case where the displacements are too large to employ perturbation techniques and thus numerical optimization is required to calculate the parameters in the flow/displacement model used. A combination of a Genetic Algorithm and a Hillclimbing Method is applied to recover the global maximum of the GC-C function in the presence of many large local maxima. The paper also briefly reviews the basic background of GC-C and develops those GA issues relevant for the implementation of the method.