In this paper, different global and local automatic registration schemes are compared in terms of accuracy and efficiency. The accuracy of different optimization strategies based on a variety of similarity measures (cross-correlation, mutual information coefficient or chamfer distance) is assessed by means of statistical tests. Results from every optimization procedure are quantitatively evaluated with respect to the gold-standard (manual) registration. The comparison has shown that chamfer distance is a robust and fast similarity measure that can be successfully combined with common optimization techniques in retinal image registration applications.