- Greg Hackney a écrit :
> Victor Duchovni wrote:the issue is cost/benefit related. At this time, I'm catching almost all
>> 0.7 sec per gif is a lot of CPU (1.4 msgs/sec/CPU just for OCR).
> The good news is that the OCR time is actually less than that, because
> that time included
> a conversion from GIF to PNM format, by a built-in call to the Linux
> "giftopnm" program.
>> A resident GIF-to-OCR daemon may be able to go faster...
> Faster is better. It seems there is sufficient interest in pursuing the
> idea. But
> the initial hurdle as I see it, is first finding a good opensource OCR
> engine. JOCR
> is an underdeveloped project, and has no recent updates to it (although
> they have
> asked for help). Perhaps there's another more mature OCR-text program
of these with bayes and few heuristic rules. and I believe one can
generate images that can't be detected (this is not the case today, but
there's a market for this, so I think this will come). the technology is
already here (and is developped by people who try to fight spam!). look