cuneiform team mailing list archive

[Marcin Miłkowski <milek_pl@xxxxx>]

Hi,

W dniu 2009-11-19 21:53, Alexey Kryukov pisze:
> Hi,
>
> I am new to this list and I am interesting in using hOCR in order to
> generate a hidden text layer in DjVu books.
>
> I see hOCR support has been greatly improved after the latest release.
> However still there are some glitches. First of all, the format
> currently used for x_bboxes data looks a bit strange: cuneiform first
> writes a text line and then an empty<span>  element with character bbox
> info, i.e.:
>
> <span class='ocr_line'...>Some text<span class='ocr_cinfo'...></span></span>
>
> I may be wrong here, but, according to my understanding of the spec this
> <span>  should rather enclose the corresponding text, i. e.:
>
> <span class='ocr_line'...><span class='ocr_cinfo'...>Some text</span></span>
>
> I am not sure writing a parser for the currently produced hOCR would make any
> sense, as such a parser probably would be incompatible with the output of
> other hOCR-capable engines. Can anybody comment on this issue?

Well, I can see that ocrodjvu that includes djvu2hocr creates the tags 
the way you describe. It's probably a slight glitch in cuneiform code.

Anyway, I compiled the current version a couple of days ago and found 
that the image value for title is always "none.txt", which is plainly 
incorrect:

<div class='ocr_page' id='page_1' title='image "none.txt"; bbox 0 0 2816 
2112'>

I'm using a sed script to correct it for my purposes (I run it in a 
loop, so I always know what image has been passed to cuneiform) but it 
should be fixed in the code as well, I guess. Or maybe I'm missing 
something?

Regards
Marcin