问题
I want to be able to detect a pattern in a PDF and somehow flag it.
For instance, in this PDF, there's the string *2. I want to be able to parse the PDF, detect all instances of *[integer], and do something to call attention to the matches (like highlight them yellow or add a symbol in the margin).
I would prefer to do this in Python, but I'm open to other languages. So far, I've been able to use pyPdf to read the PDF's text. I can use a regex to detect the pattern. But I haven't been able to figure out how to flag the match and re-save the PDF.
回答1:
Either people are not interested, or Python's not capable, so here's solution in Perl :-). Seriously, as noted above, you don't need to "alter strings". PDF annotations are the solution for you. I had small project with annotations not long ago, some code's from there. But, my content parser was not universal, and you don't need full-blown parsing -- meaning being able to alter content and write it back. Therefore I resorted to external tool. PDF Library I use is somewhat low-level, but I don't mind. It also means, one's expected to have proper knowledge of PDF internals to understand what's going on. Otherwise, just use the tool.
Here's a shot of marking e.g. all gerunds in OP's file with a command
perl pdf_hl.pl -f westlaw.pdf -p '\S*ing'
The code (comment inside worth reading, too):
use strict;
use warnings;
use XML::Simple;
use CAM::PDF;
use Getopt::Long;
use Regexp::Assemble;
#####################################################################
#
# This is PDF highlight mark-up tool.
# Though fully functional, it's still a prototype proof-of-concept.
# Please don't feed it with non-pdf files or patterns like '\d*'
# (because you probably want '\d+', don't you?).
#
# Requires muPDF-tools installed and in the PATH, plus some CPAN modules.
#
# ToDo:
# - error handling is primitive if any.
# - cropped files (CropBox) are processed incorrectly. Fix it.
# - of course there can be other useful parameters.
# - allow loading them from file.
# - allow searching across lines (e.g. for multi-word patterns)
# and certainly across "spans" within a line (see mudraw output).
# - multi-color mark-up, not just yellow.
# - control over output file name.
# - compress output (use cleanoutput method instead of output,
# plus more robust (think compressed object streams) compressors
# may be useful).
# - file list processing.
# - annotations are not just colorful marks on the page, their
# dictionaries can contain all sorts of useful information, which may
# be extracted automatically further up the food chain i.e. by
# whoever consumes these files (date, time, author, comments, actual
# text below, etc., etc., plus think of customized appearence streams,
# placing them on layers, etc..
# - ???
#
# Most complexity in the code comes from adding appearance
# dictionary (AP). You can safely delete it, because most viewers don't
# need AP for standard annotations. Ironically, muPDF-viewer wants it
# (otherwise highlight placement is not 100% correct), and since I relied
# on muPDF-tools, I thought it be proper to create PDFs consumable by
# their viewer... Firefox wants AP too, btw.
#
#####################################################################
my ($file, $csv);
my ($c_flag, $w_flag) = (0, 1);
GetOptions('-f=s' => \$file, '-p=s' => \$csv,
'-c!' => \$c_flag, '-w!' => \$w_flag)
and defined($file)
and defined($csv)
or die "\nUsage: perl $0 -f FILE -p LIST -c -w\n\n",
"\t-f\t\tFILE\t PDF file to annotate\n",
"\t-p\t\tLIST\t comma-separated patterns\n",
"\t-c or -noc\t\t be case sensitive (default = no)\n",
"\t-w or -now\t\t whole words only (default = yes)\n";
my $re = Regexp::Assemble->new
->add(split(',', $csv))
->anchor_word($w_flag)
->flags($c_flag ? '' : 'i')
->re;
my $xml = qx/mudraw -ttt $file/;
my $tree = XMLin($xml, ForceArray => [qw/page block line span char/]);
my $pdf = CAM::PDF->new($file);
sub __num_nodes_list {
my $precision = shift;
[ map {CAM::PDF::Node->new('number', sprintf("%.${precision}f", $_))} @_ ]
}
sub add_highlight {
my ($idx, $x1, $y1, $x2, $y2) = @_;
my $p = $pdf->getPage($idx);
# mirror vertically to get to normal cartesian plane
my ($X1, $Y1, $X2, $Y2) = $pdf->getPageDimensions($idx);
($x1, $y1, $x2, $y2) = ($X1 + $x1, $Y2 - $y2, $X1 + $x2, $Y2 - $y1);
# corner radius
my $r = 2;
# AP appearance stream
my $s = "/GS0 gs 1 1 0 rg 1 1 0 RG\n";
$s .= "1 j @{[sprintf '%.0f', $r * 2]} w\n";
$s .= "0 0 @{[sprintf '%.1f', $x2 - $x1]} ";
$s .= "@{[sprintf '%.1f',$y2 - $y1]} re B\n";
my $highlight = CAM::PDF::Node->new('dictionary', {
Subtype => CAM::PDF::Node->new('label', 'Highlight'),
Rect => CAM::PDF::Node->new('array',
__num_nodes_list(1, $x1 - $r, $y1 - $r, $x2 + $r * 2, $y2 + $r * 2)),
QuadPoints => CAM::PDF::Node->new('array',
__num_nodes_list(1, $x1, $y2, $x2, $y2, $x1, $y1, $x2, $y1)),
BS => CAM::PDF::Node->new('dictionary', {
S => CAM::PDF::Node->new('label', 'S'),
W => CAM::PDF::Node->new('number', 0),
}),
Border => CAM::PDF::Node->new('array',
__num_nodes_list(0, 0, 0, 0)),
C => CAM::PDF::Node->new('array',
__num_nodes_list(0, 1, 1, 0)),
AP => CAM::PDF::Node->new('dictionary', {
N => CAM::PDF::Node->new('reference',
$pdf->appendObject(undef,
CAM::PDF::Node->new('object',
CAM::PDF::Node->new('dictionary', {
Subtype => CAM::PDF::Node->new('label', 'Form'),
BBox => CAM::PDF::Node->new('array',
__num_nodes_list(1, -$r, -$r, $x2 - $x1 + $r * 2,
$y2 - $y1 + $r * 2)),
Resources => CAM::PDF::Node->new('dictionary', {
ExtGState => CAM::PDF::Node->new('dictionary', {
GS0 => CAM::PDF::Node->new('dictionary', {
BM => CAM::PDF::Node->new('label',
'Multiply'),
}),
}),
}),
StreamData => CAM::PDF::Node->new('stream', $s),
Length => CAM::PDF::Node->new('number', length $s),
}),
),
,0),
),
}),
});
$p->{Annots} ||= CAM::PDF::Node->new('array', []);
push @{$pdf->getValue($p->{Annots})}, $highlight;
$pdf->{changes}->{$p->{Type}->{objnum}} = 1
}
my $page_index = 1;
for my $page (@{$tree->{page}}) {
for my $block (@{$page->{block}}) {
for my $line (@{$block->{line}}) {
for my $span (@{$line->{span}}) {
my $string = join '', map {$_->{c}} @{$span->{char}};
while ($string =~ /$re/g) {
my ($x1, $y1) =
split ' ', $span->{char}->[$-[0]]->{bbox};
my (undef, undef, $x2, $y2) =
split ' ', $span->{char}->[$+[0] - 1]->{bbox};
add_highlight($page_index, $x1, $y1, $x2, $y2)
}
}
}
}
$page_index ++
}
$pdf->output($file =~ s/(.{4}$)/++$1/r);
__END__
P.s. I tagged the Question with 'Perl', to maybe have some feedback (code corrections, etc.) from community.
回答2:
This is non-trivial. The problem is that PDF files are not meant to be "updated" on anything less than a page. You basically have to parse the page, adjust the PostScript rendering, and then write it back out. I don't think PyPDF has the support for doing what you want.
If "all" you want to do is to add highlighting you can probably just use the annotation dictionary. See the PDF specification for more information.
You might be able to do this using pyPDF2 but I haven't looked into it closely.
来源：https://stackoverflow.com/questions/19414763/detect-and-alter-strings-in-pdfs