Abstract: A processed (e.g., captured) video sequence is temporally, spatially, and/or histogram registered to the corresponding original video sequence by generating, for each set of one or more processed frames, a mapping from a selected set of one or more original frames to the processed set, wherein (1) each selected set depends on the selected set corresponding to a previous processed set, (2) each mapping minimizes a local prediction error between the original set and the corresponding processed set, and (3) the accumulated prediction error for the entire processed video sequence is minimized.