the current version sometimes ends up with pointmatches having a high…

[StephanPreibisch]

…er error than epsilon if one re-fits the model using the inliers. It has to iterate until the number of inliers didn't change once