Solution to the Registration Problem Using Differential Evolution and SSD-ARC Function

The problem of image registration is to find the best set of parameters of an affine transformation, which applied to a given image yields the closest match to a target image (possibly with noise). We present a method to perform parametric image registration based on Differential Evolution. Besides using Differential Evolution, we propose to use an error function robust enough to discard misleading information contained in outliers. The results are compared to those obtained using Genetic Algorithms. It is clear that Differential Evolution outperforms Genetic Algorithms in terms of speed (number of evaluations), and quality of the solutions (accuracy). The quality of the solutions provided by Differential Evolution is so good that they do not need to be refined by gradient methods. At the end we present a general analysis and discussion about why DE converges in a better way than GA.