pdftogt.rb

#!/usr/bin/env ruby

Encoding.default_external = Encoding::UTF_8
Encoding.default_internal = Encoding::UTF_8

require 'nokogiri'

pdf_filename = ARGV[0]
xml_filename = File.basename(pdf_filename,'.pdf') + '.xml'

$stderr.puts "Converting to XML..."
`pdftohtml -c -zoom 3 -xml #{pdf_filename} #{xml_filename}`

doc = File.open(xml_filename) { |f| Nokogiri::XML(f) }

width = doc.xpath('//page').first.attr('width').to_i
height = doc.xpath('//page').first.attr('height').to_i

$stderr.puts "Converting to #{width}x#{height} PNG files..."
`mkdir -p pdfimages`
`convert -density 300 #{pdf_filename} -type Grayscale -background white +matte -resize #{width}x#{height} -depth 8 pdfimages/page_%05d.png`

$stderr.puts "Cropping into groundtruth..."
`mkdir -p groundtruth`
i = 0
j = 0
doc.xpath('//page').each do |page|
  page_filename = sprintf("pdfimages/page_%05d.png", i)
  $stderr.puts page_filename
  page.children.select{|node| (node.name == 'text') && (!node.content.strip.empty?) && (!node.content.include?('Copyright'))}.each do |text|
    # width x height + left + top
    png_filename = sprintf("groundtruth/%06d.png", j)
    txt_filename = sprintf("groundtruth/%06d.gt.txt", j)
    `convert #{page_filename} -crop #{text['width']}x#{text['height']}+#{text['left']}+#{text['top']} +repage #{png_filename}`
    File.open(txt_filename, 'w') {|file| file.write(text.content)}
    $stderr.puts "#{png_filename}: #{text.content}"
    j += 1
  end
  i += 1
end