Extracting dates from images with Python and OCR

jjncj.com

Thursday, October 11. 2007

Posted by Joshua Kugler in programming at 11:14 | Comments (0) | Trackbacks (0)

Extracting dates from images with Python and OCR

In my job at EE Internet, we do a lot of remote data collection, including collection from cameras. Now, the makers of these cameras have figured out how to put the date and time as a graphic in the image itself (Umiat Airfield) but for some reason they haven't figured out how to embed that same information in the EXIF Data for that image. Very aggravating.

So, Python to the rescue. I came up with a script that extracts the area given, converts it to black and white, inverts it (because these cameras have light text on dark background, and OCR doesn't seem to like that), and then feeds that file to a command-line OCR program. It then takes that text, parses it for the date, and then prints it out in any date/time format desired. Works quite well.

Here it is in all its glory.

Comments, improvements, critiques, etc., are always welcome. And yes, you can post comments pointing to another program/project/whatever that does what I did and does it way better. :-)


  • It requires Python's PIL image library

  • It defaults to using the gocr command. This is available in the 'gocr' package in Ubuntu. It also seems to work well with ocrad, which is also available in Ubuntu.

  • There should not be any symlink vulnerabilities in this as the exclusive lock is kept on the file while there is data being written to it. Once the data is written, and the file is closed, then it is passed to the OCR program as an input file. Then, it is removed.


UPDATE: Modularized it so you can call it from the command line, or import it into your own Python script and call it that way.
  1.  
  2. #!/usr/bin/python
  3.  
  4. # ##############################################################################
  5. # Filename: image2date.py
  6. # Author: Joshua Kugler, joshua@eeinternet.com
  7. # Copyright Information: (C) 2007, EEInternet, Joshua Kugler
  8. # License Information: Public Domain
  9. # Synopsis: Extracts an area from an image, and does OCR to obtain the date
  10. # Comments: Requires gocr to be installed
  11. # Info: $Id: image2date.py 56 2007-10-11 20:08:01Z jkugler $
  12. # ##############################################################################
  13.  
  14. def err(msg, do_exit=True):
  15.     import sys
  16.     sys.stderr.write(msg + '\n')
  17.     if do_exit:
  18.         sys.stdout.write('')
  19.         sys.exit(1)
  20.  
  21. def do_ocr(filename, left, top, right, bottom, date_format, output_format, strip_chars='', invert=True, debug=False, ocr='gocr'):
  22.     import os
  23.     import subprocess
  24.     import tempfile
  25.     import time
  26.    
  27.     import Image
  28.     import ImageChops
  29.  
  30.  
  31.     try:
  32.         im = Image.open(filename)
  33.     except IOError, e:
  34.         err('Cannot open image file %s: %s' % (filename, e.strerror))
  35.  
  36.     im = im.crop((left, top, right , bottom))
  37.  
  38.     # Converts to black and white because OCR works better that way.
  39.     im = im.convert('L')
  40.  
  41.     # Web cams need seem to like light-on-dark text, so we usually invert
  42.     if invert:
  43.         im = ImageChops.invert(im)
  44.  
  45.     # Get a temporary file so we can save the image.
  46.     (fd, tf) = tempfile.mkstemp(suffix='.ppm', dir='/tmp')
  47.  
  48.     im.save(tf)
  49.  
  50.     os.close(fd)
  51.  
  52.     try:
  53.         ocr_output = subprocess.Popen([ocr, tf], stdout=subprocess.PIPE).communicate()[0]
  54.     except OSError, e:
  55.         os.remove(tf)
  56.         err('Could not execute OCR program "%s": %s' % (ocr, str(e)))
  57.  
  58.     os.remove(tf)
  59.  
  60.     ocr_output = ocr_output.strip()
  61.  
  62.     # Fixes a glitch whereby gocr recognizes two ones as 'll'
  63.     # Seems to do fine with single ones.  Odd.
  64.     ocr_output = ocr_output.replace('ll','11')
  65.  
  66.     if debug:
  67.         err("Raw OCR: '%s'" % ocr_output, False)
  68.  
  69.     if strip_chars:
  70.         ocr_output = ocr_output.replace(strip_chars,'')
  71.  
  72.         if debug:
  73.             err("Stripped OCR: %s" % ocr_output, False)
  74.  
  75.     try:
  76.         time_data = time.strptime(ocr_output, date_format)
  77.     except ValueError, e:
  78.         err('Cannot decode time """%s""": %s' % (ocr_output, str(e)))
  79.  
  80.     if debug:
  81.         err('Time data: %s' % str(time_data), False)
  82.  
  83.     formatted_filename = time.strftime(output_format, time_data)
  84.    
  85.     return(formatted_filename)
  86.  
  87.  
  88. def main():
  89.     import optparse
  90.     import sys
  91.  
  92.     usage = 'image2date --date-format=DATE_FORMAT --left=N --right=N --bottom=N [options] FILENAME'
  93.     _VERSION_ = '1.0'
  94.     parser = optparse.OptionParser(usage=usage, version = '%prog ' + _VERSION_)
  95.     parser.add_option('--date-format', dest='date_format',
  96.                       help='Format of the date being extracted.  See documentation of Python time module for more information', default=None)
  97.     parser.add_option('--output-format', dest='output_format',
  98.                       help='Format of the date printed.  See documenation of Python time module for more information', default='%Y-%m-%d-%H-%M')
  99.     parser.add_option('--strip-chars', dest='strip_chars', help='Characters to strip from the OCRed text', default='')
  100.     parser.add_option('--debug', action='store_true', dest='debug', help='Output the OCRed text and other information', default=False)
  101.     parser.add_option('--invert', dest='invert', type='int', help='Invert the image before performing OCR (1) or not (0). Defaults to 1.', default = 1)
  102.     parser.add_option('--left', dest='left', type='int', help='The left side of the box containing the date/time (in pixels)', default=0)
  103.     parser.add_option('--top', dest='top', type='int', help='The top of the box containing the date/time (in pixels)', default=0)
  104.     parser.add_option('--right', dest='right', type='int', help='The right side of the box containing the date/time (in pixels)', default=0)
  105.     parser.add_option('--bottom', dest='bottom', type='int', help='The bottom of the box containing the date/time (in pixels)', default=0)
  106.     parser.add_option('--ocr', dest='ocr', help='The binary to execute to perform the OCR.  Must accept a filename as its only argument, and print the result to STDOUT. Defaults to gocr.', default='gocr')
  107.  
  108.  
  109.     (options, args) = parser.parse_args()
  110.  
  111.     # Do a bunch of checks to makes sure we have all the options we need
  112.     if not options.date_format:
  113.         err('You must supply the format for the date in the image')
  114.  
  115.     if not (options.left and options.right and options.bottom):
  116.         err('You must supply the coordinates of the left, right, and bottom of the box containing the date/time')
  117.  
  118.     if len(args) < 1:
  119.         err('You must supply the filename of the image on which to perform the OCR operation')
  120.  
  121.     result = do_ocr(args[0], options.left, options.top, options.right , options.bottom,
  122.                     options.date_format, options.output_format, options.strip_chars,
  123.                     options.invert, options.debug, options.ocr)
  124.  
  125.  
  126.     sys.stdout.write(result)
  127.  
  128. if _name_ == "_main_":
  129.     main()
  130.  
  131.  

Defined tags for this entry: , , , ,
Related entries by tags:
Blogs are like e-mail addreses...
The Futility of Comments
My feelings on PHP, exactly.
Objects have failed...maybe...kind of
A New Job!
Trackbacks
Trackback specific URI for this entry
No Trackbacks
Comments
Display comments as (Linear | Threaded)
No comments
Add Comment

Enclosing asterisks marks text as bold (*word*), underscore are made via _word_.
Standard emoticons like :-) and ;-) are converted to images.
E-Mail addresses will not be displayed and will only be used for E-Mail notifications
 
 



Our Links

Calendar

Back November '08 Forward
Mon Tue Wed Thu Fri Sat Sun
          1 2
3 4 5 6 7 8 9
10 11 12 13 14 15 16
17 18 19 20 21 22 23
24 25 26 27 28 29 30

The Good Blogs

I'm giving this a trial run. Right now, I'm only promoting Tech, Business, Venture Capital, and Auto categories, so hopefully the content will be pretty clean. If you see something that would not be in line with my standards of "family friendly," please contact me at once, and send along the URL of the offending article.

Technorati

Blog Information Profile for Pedahzur
Add to Technorati Favorites

Archives

November 2008 (3)
October 2008 (4)
September 2008 (7)
August 2008 (4)
July 2008 (4)
June 2008 (4)
Recent...
Older...

Syndicate This Blog

Powered by

Serendipity PHP Weblog

Categories



All categories