Interestingly, early stopping acts as a very efficient regularizer!
See below: instead of choosing to regularize many iterations, it often makes sense to simply stop the restoration at the optimum point. Finding this optimum, however, is the tricky part.