Tesseract hOCR to produce xml, not (x)html, or a tool to simulate this, or a tool to collect the div (etc) attributes

[Kim Rönnberg]

Is there a way to make Tesseract produce "real" xml instead of the (x)html hOCR produces, ie. to create xml tags like <ocr_page id='page_1' title='...'> instead of "<div class='ocr_page' id='page_1'...", <ocr_area id='...' title='...'> instead of "<div class='ocr_carea' id='block_1_1'..." etc.?

Or is there somewhere a "ready" something with which the (x)html hOCR produces can be converted to a more "easily" xml parseable format, or, even better, a something that would give me the div's, span's and p's gouped per word, line, area and page readily insertable to a (php) array for inserting into a database, of the data format the hOCR produces now?

Like "file_name", "page_nr", "area_id", "line_nr", "word_nr", "word bbox x1 y1 x2 y2", "the word value", for each word? I realise this means a lot of rows (one per word in a document), but this is something I need.

I have spent some days on this, trying to find something that works on php, but have not managed to find anything.

Regards

Kim Rönnberg

[jsbien]

There are some tools to convert hOCR to XCES (XML Corpus Encoding Format):

 https://bitbucket.org/jwilk/marasca-wbl/

Regards

Janusz

[Tom Morris]

On Saturday, March 26, 2016 at 6:56:17 AM UTC-4, Kim Rönnberg wrote:

Is there a way to make Tesseract produce "real" xml instead of the (x)html hOCR produces, ie. to create xml tags like <ocr_page id='page_1' title='...'> instead of "<div class='ocr_page' id='page_1'...", <ocr_area id='...' title='...'> instead of "<div class='ocr_carea' id='block_1_1'..." etc.? 

Or is there somewhere a "ready" something with which the (x)html hOCR produces can be converted to a more "easily" xml parseable format, or, even better, a something that would give me the div's, span's and p's gouped per word, line, area and page readily insertable to a (php) array for inserting into a database, of the data format the hOCR produces now?

PHP? Ugh! But that aside, what specific problem is it having parsing the output? XHTML *is* "real" XML.

Like "file_name", "page_nr", "area_id", "line_nr", "word_nr", "word bbox x1 y1 x2 y2", "the word value", for each word? I realise this means a lot of rows (one per word in a document), but this is something I need.

You might be interested in the TSV format that was added recently (but it's not available in a release yet):

https://github.com/tesseract-ocr/tesseract/pull/245

Tom

[Helmut Wollmersdorfer]

I usually use Perl for such tasks:

use Mojo::DOM;

open(my $hocr_fh,"<:encoding(UTF-8)",$hocr_file) or die "cannot open $hocr_file: $!";

my $html = '';

while (my $line = <$hocr_fh>) { $html .= $line;} 

my $dom = Mojo::DOM->new($html);

my $ocr_page ={};

# <div class='ocr_page' id='page_1' title='image "isisvonoken00oken_0153.png"; bbox 0 0 2321 2817; ppageno 0'>

for my $e ($dom->find('div.ocr_page')->each) { 

    my $title = $e->{'title'}; 

    print 'page title: ',$title,"\n";

    if ($title =~ m/bbox\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)/) {

      $ocr_page->{x1} = $1;

      $ocr_page->{y1} = $2;

      $ocr_page->{x2} = $3;

      $ocr_page->{y2} = $4;

    }

}

Mojo::DOM is an XML parser allowing to navigate by CSS-selectors (like jQuery).

Of course, there are dozens of other XML parsers available in Perl.

I'm sure, there are similar parsers usable via PHP.

Helmut Wollmersdorfer