$N is a natural extension of the $1 approach to multi-stroke symbols. It lifts restrictions on (or removes the ability to specify) stroke order and direction. This moves it closer to a free-sketch symbol recognizer. It has the possibility to also lift the stroke number restriction, though that can be kept on if desired (for speed, or extra constraint).
Very much in the spirit of $1, $N is very simple. The templates are very similar to $1 templates, all 1 "stroke" and the same PPS points per stroke for every template. Matching two templates is also the same. Instead of figuring out how to put together N strokes into 1, $N just tries all possible combinations of stroke order and direction. Simple, thorough.
I think anyone who makes a multi-stroke version of $1 is automatically a pretty cool person. I mean, it really was just asking for it. $N is a good effort. Really, really in the spirit of $1. Super duper simple.
I disagree with some of the design choices. Including the "in the air" portions of the symbol introduces ambiguity (x vs alpha (the alpha glyph for this font looks terrible, sorry)) and dilutes the data (longer strokes means more space between points when resampled). Having complexity based on the number of strokes is a tough battle. Having O(N!*2^N) complexity in the number of strokes is an even tougher battle. 5 strokes means >3000 templates, 6 strokes goes up to >46000, 7 strokes is >645000. As stated, this limits the expressivity of the template set.
That being said, no one is claiming $N is some sort of silver bullet. In fact, the authors are very upfront about the limitations of $N.
No comments:
Post a Comment