Thanks a lot. I think I'll try the word level training.

Raj

On Jul 18, 11:34 am, pkt <[email protected]> wrote:
> On 15 Éïýë, 13:58, Raj Julha <[email protected]> wrote:
>
> > Hi
>
> > I notice that due to the complexity of extracting characters from
> > handwritten historical texts some research has been done in
> > recognising words by using word images. I'm planning to train the
> > engine for a specific handwriting so I'd like to use a set of word
> > images and their corresponding transcription. Can I do that with
> > OCROPUS? If yes any hints on how I could do that?
>
> I think you are refering to "segmentation-free" recognition.
> AFAIK Ocropus doesn't have a "segmentation-free" recogniser,
> so if you insist on taking this route you would have to implement
> the recognition step (line images -> character hypotheses FST)
> yourself I think.
>
> On the other hand, it seems Ocropus' recognizer approach
> (oversegmentation + MLP neural net) was initially developed
> for handwriting recognition so if the handwriting you care about
> is not cursive / very different from type, you can give it a try.
>
> Wrt training, from what it seems from the latest posted screencasts,
> Ocropus can be trained fine with word-level, line-level or even
> page-level ground truth and it will try to do the alignment itself.

-- 
You received this message because you are subscribed to the Google Groups 
"ocropus" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to 
[email protected].
For more options, visit this group at 
http://groups.google.com/group/ocropus?hl=en.

Reply via email to