Judging from the fact that it is a docomo phone, I'd say the biggest
limitations will be in the sandbox itself. As with my camera-based demo,
it will most probably be possible to do a sample appli that will
recognize strokes and... print them to screen, maybe. But you are
limited by three main factors (in case API allows pixel-precise position
reading):
1. Sandbox - output of text recognition will not be usable in other applis
2. Storage memory - to get decent recognition, you need a vocabulary. In
case of online recognition this will probably be possible to put into
few megabytes, maybe even around one MB, but if you want hinting from
offline recognizer (improves recognition rates by a big margin), you are
looking at database sizes of magnitudes of tens of megabytes. Definitely
not possible under current architecture of sandbox.
3. usability of pen-based interface on screen this small.
I am sure these things will improve, as technically it is definitely
possible. Just at current state of things, I think Java applis have too
limited API, which unfortunately means you have to wait for the
manufacturer himself to implement different IME inputs (i.e. camera
recognition on new NEC handset)
Petr
> So i wonder if you can actually read pixels, or just the button AWT
> components have hooks to respond to the touch screen as well as the
> keyboard.
>
> Something like a kanji recognizer would need a pretty low level API, a
> bit different from a remote control with buttons.
Received on Wed Jul 7 18:48:20 2004