Table 2.

Summary of the methods used to evaluate image-sequence registration.

Methods	Result
Rough Evaluation
movie	image sequence appears to be stable, but the test is not conclusive, because small error growth cannot be observed
spatio-temporal image	moving objects appear as slanted lines and fixed objects as vertical lines
difference image	darker and lighter areas represent non-registered areas; good for visually comparing consecutive images
Precise Evaluation
errors in estimated parameters	not possible to assess, because there is no ground truth available
radiometric errors	lower errors considered as a better match; suited for consecutive images and images with negligible brightness variation; we estimated homography parameters between consecutive images by minimizing this error (Section 4.1, our paper [26])
geometric errors	flipping images and randomly checking coordinates; this method gives the final result