Possible issue with ColorSpace

[RussellMcOrmond]

Dev Effort

2D

Description

We have some CMYK which aren't differentiated in the JHOVE 1.16.5 reports. I'm wondering if this is a problem in the software, or that I'm not using the tool correctly.

Using ImageMagick's identify it notices the differences between a lone CMYK image and Greyscale images.

Jhove indicated YCbCr for all of them.

cihm@quark:/opt/wip/Temp/export/oop_issues/oop.debates_HOC0104_01-sip/data/files$ identify -verbose -format "%f %[colorspace]\n" 000?.jpg 
0001.jpg CMYK
0002.jpg Gray
0003.jpg Gray
0004.jpg Gray
0005.jpg Gray
0006.jpg Gray
0007.jpg Gray
0008.jpg Gray
0009.jpg Gray

cihm@quark:/opt/wip/Temp/export/oop_issues/oop.debates_HOC0104_01-sip/data/files$ /opt/jhove/jhove 000?.jpg | grep ColorSpace
      ColorSpace: YCbCr
      ColorSpace: YCbCr
      ColorSpace: YCbCr
      ColorSpace: YCbCr
      ColorSpace: YCbCr
      ColorSpace: YCbCr
      ColorSpace: YCbCr
      ColorSpace: YCbCr
      ColorSpace: YCbCr
cihm@quark:/opt/wip/Temp/export/oop_issues/oop.debates_HOC0104_01-sip/data/files$ /opt/jhove/jhove -v
Jhove (Rel. 1.16.5, 2017-03-20)
 Date: 2017-11-27 16:38:57 EST
 App:
  API: 1.16.5, 2017-03-20
  Configuration: /opt/jhove/conf/jhove.conf
  JhoveHome: /opt/jhove
  Encoding: utf-8
  TempDirectory: /tmp
  BufferSize: 131072
  Module: AIFF-hul 1.3
  Module: ASCII-hul 1.3
  Module: BYTESTREAM 1.3
  Module: GIF-hul 1.3
  Module: GZIP-kb 0.1
  Module: HTML-hul 1.3
  Module: JPEG-hul 1.2
  Module: JPEG2000-hul 1.3
  Module: PDF-hul 1.8
  Module: PNG-gdm 1.0
  Module: TIFF-hul 1.7
  Module: UTF8-hul 1.6
  Module: WARC-kb 1.0
  Module: WAVE-hul 1.4
  Module: XML-hul 1.4
  OutputHandler: Audit 1.1
  OutputHandler: TEXT 1.5
  OutputHandler: XML 1.7
  Usage: java JHOVE [-c config] [-m module] [-h handler] [-e encoding] [-H handler] [-o output] [-x saxclass] [-t tempdir] [-b bufsize] [-l loglevel] [[-krs] dir-file-or-uri [...]]
  Rights: Derived from software Copyright 2004-2011 by the President and Fellows of Harvard College. Version 1.7 to 1.11 independently released. Version 1.12 onwards released by Open Preservation Foundation. Released under the GNU Lesser General Public License.
cihm@quark:/opt/wip/Temp/export/oop_issues/oop.debates_HOC0104_01-sip/data/files$ 

I noticed this issue when trying to confirm image differences in the context of cantaloupe-project/cantaloupe#177

[carlwilson]

Hi @RussellMcOrmond, I've taken an initial look and there's not a straightforward fix leaps out at me so I won't be addressing this for 1.18 which goes out today. I'm going to assign @tledoux to this as he implemented the colour space code and might be able to give you a fuller answer.

[tledoux]

Hi, I implement the colorProfile code for JPEG, not the ColorSpace.

In fact, the problem seems to come from the line 1009 in JpegModule where the presence of a JFIF header wrongly implies a YCbCr color space.

So this seems like a bug (at least, a shortcut) that needs to be corrected.

@RussellMcOrmond, can we add your files to the corpora so that we can build tests for these cases ?

[carlwilson]

Thanks @tledoux

[RussellMcOrmond]

Yes, @tledoux , you can use those files as test cases.

You can download using Cantaloupe from our repository, of I can attach them if that is easier. Turns out we only had two CMYK images in that collection which is the title pages of the English and French versions of the same document.

http://parl.canadiana.ca/view/oop.debates_CDC0104_01/1?r=0&s=1
http://parl.canadiana.ca/view/oop.debates_HOC0104_01/1?r=0&s=1

From cantaloupe-project/cantaloupe#177 , to download you:

• Enable "developer tools" in Chrome, Firebug for Firefox or equivalent that will show network traffic
• Go to one of the two above URL's
• The first page will have "There was an error retrieving this page." where on other pages there is an image.
• Open the URL for the first image in a new tab (It has the 0001.jpg just before /full. Cantaloupe is also only returning 586 bytes, which represent the headers alone). This URL has an authorisation token, which is why I needed to describe it this way. You will note that the page is blank.
• If you change the /!800,800/ in the URL to /full/ you will see the image directly streamed to you (unmodified by Cantaloupe). Save this file.

[carlwilson]

Hi @RussellMcOrmond I had a look at this for the 1.20 release but I haven't been able to fix it at a first pass as the research left me a little confused. Reading the APP0 material I've Googled does suggest that the use of a JFIF header implies CMYK. It's also obvious that tools like ImageMagick do more than this but the "rules" aren't straightforward. I agree with @tledoux that the current shortcut should be replace, it's just not currently clear by what.

[thorsted]

I wanted to add our use case. We have an image with an ICC Profile "Display P3" which is throwing an ErrorMessage "Invalid ICCProfile in APP2 segment, No description in ICC profile v4" Offset 596. Jpeg is Well-Formed and valid.