> Does the iAppli work by sending the
> image to the server for processing? (I don't have a keitai in Japan to test
> it.) If so, wouldn't that be expensive, in terms of packet charges? Still,
> really useful if you just HAVE to know what those damn kanji are and you forgot
> your WordTank. Best of luck with it.
Yes, it sends a GIF image (about 1500 bytes) to server for each
recognition, i.e. some 16-20 packets all together including HTTP overhead
and reply. It would definitely be nicer to have it all in mobile... and
actually IMO the processor in todays mobiles would be capable of that...
but then comes the dictionary size of 100MBytes (could be probably cut
down to some 30MBs with still sensible recognition rate, but way too big
for 200kB scratchpad anyway).
Overall, though, it is not that expensive if you just don't understand one
sentence... but can be quite costly for anything more.
On the usage note: people, do not send in pictures like
http://kohaku.anime.cz/wrong3.gif - it is just waste of packets :). The
result of preprocessing needs to be like
http://kohaku.anime.cz/correct.gif - really, like that, one line of text
only, up to 10 kanjis, not more. i.e. on this picture there is too big
upper black border that results in inability to trace lines in the text...
and letters are too small to read them.
try it i.e. on some shop sign - that is what it is useful for for now, do
not expect it to translate Basho.
I know, these are just excuses - preprocessing needs to be worked on...
but for now, it is just like this (and is probably not gonna improve
before christmas are finished ^^)
Take care,
Petr
Received on Fri Dec 19 07:25:28 2003
