Looks pretty nice. I think there'd be a way of computing the blurriness without the iterative approach and without the aliasing produced in the process.