Evaluation of the precision and the robustness of different image registration algorithms with added imperfections. (a) Proposed simulation strategy: two sub-regions with known lateral offset are cropped from an identical image. After applying noise and/or blurring effects on the sub-regions, they are registered using four different algorithms. The calculated amounts of image offset (Δx and Δy) are then compared against the known true values. (b) The performance of different algorithms (from left to right: center of mass, phase correlation, scale invariant feature transform, and intensity-based automatic image registration) as a function of artificially induced noise and image blurring. Registration results with error smaller or equal to 1 pixel are labeled as `excellent'; results with error between 1 and 2 pixels are labeled as `satisfied'; results with error between 2 and 5 pixels are labeled as `fair'; results with error larger than 5 are considered as `failed'.