Convert hOCR to PDF

Jonathan Brinley — Thu, 02 Apr 2009 20:05:46 +0000

As I mentioned recently, OCRopus OCR software output an hOCR file. What is hOCR? hOCR is an open standard for representing OCR results in an HTML document (not to be confused with HOCR). It is basically a microformat using div and span tags’ class and title attributes to convey information from the OCR process, as you can see from this example of a basic hOCR document:


  PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
  "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
 xmlns="http://www.w3.org/1999/xhtml">
  >
     content="ocr_line ocr_page" name="ocr-capabilities"/>
     content="en" name="ocr-langs"/>
     content="Latn" name="ocr-scripts"/>
     content="" name="ocr-microformats"/>
    >OCR Output>
  >
  >
     class="ocr_page" title="bbox 0 0 2548 3300; image /path/to/scanned/image.png">
       class="ocr_line" title="bbox 659 143 863 177">Some Text>
       class="ocr_line" title="bbox 723 275 916 324">More Text>
    >
  >
>

For the more details, see Thomas Breuel’s complete hOCR specification draft. The example, though, shows all that I need to know. From the div[@class='ocr_page'], I can learn the dimensions (in pixels) of the image that was OCRed (as well as the path to that image on my machine). From a span[@class='ocr_line'], I can learn the location and dimensions of the bounding box around a line of recognized text, as well as the content of that line.

That information, along with a copy of the original image (or a sufficiently similar image) is enough to create a PDF of the image with selectable text.

Creating a PDF

A while back, Florian Hackenberger created a basic hOCR to PDF converter in Java. It does its job reasonably well, but it’s somewhat rough around the edges. Since I’m more familiar with Python than Java, it seemed like a good idea to rewrite the code in Python, so I could more comfortably hack it.

And so I present to you, dear reader, the hOCR Converter Python script (note: this script depends on the ReportLab PDF Library, which, in turn, depends on the FreeType 2 Font Engine). Right now, it has two primary functions:

Given an hOCR file, create a text-only (i.e., no HTML) document
Given an hOCR file and an image, create a PDF

Usage is pretty simple:

from HocrConverter import HocrConverter
hocr = HocrConverter("myHocrFile.html")
hocr.to_text("output.txt")
hocr.to_pdf("myImageFile.png", "output.pdf")

You end up with a text file at output.txt containing the contents of the body of the hOCR document and a PDF file at output.pdf containing your image superimposed on top of the text.

Note that the image you use to create the PDF need not be the same image used for the OCR. If, for example, you used a 300 or 400 DPI image for the OCR, but you want a smaller file for the PDF, you can create a 72 DPI version of the image and feed that through the script instead of the original.

With a simple script that iterates through a directory of images, calling OCRopus and this script for each image (maybe with some ImageMagick and pdftk thrown in for good measure), you can quickly OCR a large batch of images and convert them to searchable PDF documents.

Future Development

I’d like to add the capability to convert the output of various OCR programs into hOCR format. For example, OmniPage, which is much better than OCRopus when it comes to layout analysis for complex images, can output documents in its proprietary XML schema. I should be able to transform that into an hOCR document and then use this script to create a PDF from there.

I would welcome any other suggestions or contributions.

Download the HocrConverter script from github

Command Line PDF Editing

Jonathan Brinley — Thu, 09 Oct 2008 14:17:35 +0000

As I’ve mentioned before, Acrobat’s JavaScript API lags far behind other Adobe applications. Its limitations turned a seemingly simple project I was working on into an exercise in futility.

Overview

I have a collection of a little over 5,000 PDF files, the output of an OCR job. Each file contains one page of a newspaper. Four pages put together would make one issue. My goal, then, is to take four PDF files (e.g., 1896-09-24_001.pdf, 1896-09-24_002.pdf, 1896-09-24_003.pdf, and 1896-09-24_004.pdf) and merge them together into one file (1896-09-24.pdf). Then repeat 1,300 times or so to get the rest of the issues.

Sounds like an ideal job for a small script. Unfortunately, Acrobat only gives JavaScript access to the file system for opening and saving files. It has no way to read a directory for a list of files, which is rather fundamental to the task at hand.

PyPdf Almost Works

At Jay Luker‘s suggestion, I tried out PyPdf. It seems to do everything I needed. And indeed it would, except it can’t read my PDF files. It does its job just fine with other files, but not these that OmniPage created. It seems the files are missing an attribute that PyPdf looks for, so I end up with a KeyError.

pdftk to the Rescue

So, after much gnashing of teeth, Jason Ronallo suggested I try pdftk, a simple command line tool that can merge and split PDFs, among other capabilities. To merge the issue noted above takes just one line:

pdftk 1896-09-24_001.pdf 1896-09-24_002.pdf 1896-09-24_003.pdf 1896-09-24_004.pdf cat output 1896-09-24.pdf

A simple Python script can just call pdftk repeatedly to take care of the whole collection.

Many thanks, pdftk and #code4lib.