I'm following this approach. One of the most interesting things so far has been observing just how separate recall and production are. There are kanji that I can recognise instantly, and recall meanings and pronunciations, but I can't visualise them at all.