Package: gscan2pdf
Version: 2.11.0-1
Severity: wishlist
Tags: patch upstream
Hi,
please find attached 3 patches improving hOCR support.
- recognize more tags when reading hOCR
In addition this also helps for rotated texts (in a separate report)
- preserve more properties when writing hOCR
- fix-indentation of intermediate non-leaf elements
The patches are against upstream's version 2.11.2
Please consider incorporating them into gscan2pdf's next version
Thanks in advance
Peter
-- System Information:
Debian Release: bullseye/sid
APT prefers testing
APT policy: (990, 'testing'), (500, 'unstable'), (500, 'stable'), (1,
'experimental')
Architecture: amd64 (x86_64)
Kernel: Linux 5.10.0-5-amd64 (SMP w/12 CPU threads)
Kernel taint flags: TAINT_CRAP
Locale: LANG=de_DE.UTF-8, LC_CTYPE=de_DE.UTF-8 (charmap=UTF-8), LANGUAGE=en_GB
Shell: /bin/sh linked to /usr/bin/dash
Init: systemd (via /run/systemd/system)
LSM: AppArmor: enabled
Versions of packages gscan2pdf depends on:
ii imagemagick 8:6.9.11.60+dfsg-1
ii imagemagick-6.q16 [imagemagick] 8:6.9.11.60+dfsg-1
ii libconfig-general-perl 2.63-1
ii libdate-calc-perl 6.4-1.1
ii libfilesys-df-perl 0.92-6+b6
ii libgoocanvas2-perl 0.06-2
ii libgtk3-imageview-perl 6-1
ii libgtk3-perl 0.038-1
ii libgtk3-simplelist-perl 0.21-1
ii libhtml-parser-perl 3.75-1+b1
ii libimage-magick-perl 8:6.9.11.60+dfsg-1
ii libimage-sane-perl 5-1+b1
ii liblist-moreutils-perl 0.430-2
ii liblocale-codes-perl 3.66-1
ii liblocale-gettext-perl 1.07-4+b1
ii liblog-log4perl-perl 1.54-1
ii libossp-uuid-perl [libdata-uuid-perl] 1.6.2-1.5+b9
ii libpdf-builder-perl 3.021-2
ii libproc-processtable-perl 0.59-2+b1
ii libreadonly-perl 2.050-3
ii librsvg2-common 2.50.3+dfsg-1
ii libset-intspan-perl 1.19-1.1
ii libtiff-tools 4.2.0-1
ii libtry-tiny-perl 0.30-1
hi sane-utils 1.0.31-4pm1
Versions of packages gscan2pdf recommends:
ii djvulibre-bin 3.5.28-1
ii gocr 0.52-3
ii pdftk-java [pdftk] 3.2.2-1
ii tesseract-ocr 4.1.1-2.1
ii unpaper 6.1-2+b2
ii xdg-utils 1.1.3-4
gscan2pdf suggests no packages.
-- no debconf information
>From f9d32fbeb11619637ea6263881b6333afc4ebeaf Mon Sep 17 00:00:00 2001
From: Peter Marschall <pe...@adpm.de>
Date: Wed, 14 Apr 2021 17:41:56 +0200
Subject: [PATCH 1/3] Bboxtree: preserve more information in to_hocr()
Keep 'textangle' and 'baseline' properties in to_hocr() method.
Signed-off-by: Peter Marschall <pe...@adpm.de>
---
lib/Gscan2pdf/Bboxtree.pm | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/lib/Gscan2pdf/Bboxtree.pm b/lib/Gscan2pdf/Bboxtree.pm
index 59edaf61..2ee47834 100644
--- a/lib/Gscan2pdf/Bboxtree.pm
+++ b/lib/Gscan2pdf/Bboxtree.pm
@@ -514,6 +514,12 @@ EOS
$string .= $SPACE x ( 2 + $bbox->{depth} ) . "<$tag class='$type'";
if ( defined $bbox->{id} ) { $string .= " id='$bbox->{id}'" }
$string .= " title='bbox $x1 $y1 $x2 $y2";
+ if ( defined $bbox->{baseline} ) {
+ $string .= '; baseline ' . join( $SPACE, @{ $bbox->{baseline} } );
+ }
+ if ( defined $bbox->{textangle} ) {
+ $string .= "; textangle $bbox->{textangle}";
+ }
if ( defined $bbox->{confidence} ) {
$string .= "; x_wconf $bbox->{confidence}";
}
--
2.30.2
>From 11ed93483c800082525cd6a3afcfb08f0e15be9c Mon Sep 17 00:00:00 2001
From: Peter Marschall <pe...@adpm.de>
Date: Wed, 14 Apr 2021 18:02:16 +0200
Subject: [PATCH 2/3] Bboxtree: recognize more hOCR elements in _hocr2boxes()
Recognize additional elements 'ocr_header', 'ocr_footer', 'ocr_caption'
as well as their 'ocrx_...' counterparts when parsing hOCR into a Bboxtree
Recent versions of tesseract seem to generate some of these elements
instead of 'ocrx_line'.
As the line-like elements contain impoartant information, gscan2pdf needs
to recognize them, in order to
* properly diplay the OCR'ed text
* preserve as much information as possible when storing hOCR files
Signed-off-by: Peter Marschall <pe...@adpm.de>
---
lib/Gscan2pdf/Bboxtree.pm | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/lib/Gscan2pdf/Bboxtree.pm b/lib/Gscan2pdf/Bboxtree.pm
index 2ee47834..e8fa83d3 100644
--- a/lib/Gscan2pdf/Bboxtree.pm
+++ b/lib/Gscan2pdf/Bboxtree.pm
@@ -138,6 +138,15 @@ sub _hocr2boxes {
when (/_par$/xsm) {
$data->{type} = 'para';
}
+ when (/_header$/xsm) {
+ $data->{type} = 'header';
+ }
+ when (/_footer$/xsm) {
+ $data->{type} = 'footer';
+ }
+ when (/_caption$/xsm) {
+ $data->{type} = 'caption';
+ }
when (/_line$/xsm) {
$data->{type} = 'line';
}
--
2.30.2
>From 32b921376cb61ab8750ecf11c2152ec30517a579 Mon Sep 17 00:00:00 2001
From: Peter Marschall <pe...@adpm.de>
Date: Wed, 14 Apr 2021 18:41:47 +0200
Subject: [PATCH 3/3] Bboxtree: fix indentation of intermediate closing tags in
to_hocr()
When writing hOCR files using to_hocr(), make sure closing tags
of intermedate non-leaf elements are correctly indented, even
if sibling elements follow.
This makes sure to also keep the "visual" structure of hOCR files
generated by the OCR engines.
Signed-off-by: Peter Marschall <pe...@adpm.de>
---
lib/Gscan2pdf/Bboxtree.pm | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/lib/Gscan2pdf/Bboxtree.pm b/lib/Gscan2pdf/Bboxtree.pm
index e8fa83d3..4c267e02 100644
--- a/lib/Gscan2pdf/Bboxtree.pm
+++ b/lib/Gscan2pdf/Bboxtree.pm
@@ -495,8 +495,10 @@ EOS
while ( my $bbox = $iter->() ) {
if ( defined $prev_depth ) {
if ( $prev_depth >= $bbox->{depth} ) {
+ if (@tags) { $string .= '</' . pop(@tags) . ">\n" }
+ $prev_depth--;
while ( $prev_depth-- >= $bbox->{depth} ) {
- $string .= '</' . pop(@tags) . ">\n";
+ $string .= $SPACE x ( 2 + $prev_depth + 1 ) . '</' .
pop(@tags) . ">\n";
}
}
else {
--
2.30.2
