Prosumer DSC's Colour Management Issue and the Challenge for Print Makers

Dr Tony Kaye, Kodak Ltd

At the Royal Photographic Society's Imaging Science Group "Good Picture - 2006" Management and Manipulation symposium, held on the 18 December 2006, a presentation entitled "Prosumer DSC's Colour Management Issues and The challenge for Print Makers" was made. The following article summarises some of the key issues raised in that presentation

The digital camera market is evolving fast and cameras, particular digital SLRs and "high end" or "prosumer" advanced amateur cameras now have facilities that were previously confined to professional cameras. Many have the option to store images in more than one colour space; often a choice of sRGB or Adobe 1998 is available. Additionally many cameras can save images as "raw" files, allowing the decision over which colour space a JPEG or TIFF image file should be encoded with, to be deferred till time of conversion. Images encoded in Adobe 1998 colour space should be capable of producing images with a wider colour gamut than those in sRGB colour space, provided the output device/medium has a colour gamut greater than sRGB, and that the entire "flow" is properly colour managed. Often, however, photographers see little or no advantage of Adobe 1998 compared to sRGB, and some actually see a lower colour gamut from images captured and stored in Adobe 1998 than sRGB. This was investigated as these observations were contrary to expectations.

To obtain the desired colour reproduction, image manipulation and printing applications need to be colour managed. To do this, image files contain metadata which is non-image data within a file that allows applications to know what the individual red, green and blue code values in an image represent. If two files have the same pixel code values the images are only identical if the colour space is the same. If you have two images with identical code values, but one is in sRGB and the other in Adobe 1998, then the images are different. There are two ways in which the colour space metadata can be embedded in an image file

  1. EXIF data
  2. ICC profile

EXIF data is a method specified by the Japanese Electronics and Information Technology Industries Association in JEITA CP-3451 "Exchangeable Image File Format for Digital Cameras". ICC profile is a method specified by the International Color Consortium in ICC.1:2004-10 (Profile Version 4.2.0.0). While they both serve to convey colour space information, they do it in entirely different ways. An ICC embedded profile provides the necessary "maths" to allow colour managed applications to convert from the image colour space to the profile connection space, which is the backbone of colour management. EXIF data is a simpler approach where a "flag" is raised saying whether an image is in sRGB colour space or "unclassified". When the "unclassified" flag is raised and the colour space is Adobe 1998, additional metadata is provided that indicates to the software, information about the colour matching primaries used and the white point. Depending upon the camera and software used, the colour space information may be communicated via an ICC profile, EXIF data or both. Poor colour can result if colour space information is not communicated correctly. Three areas were studied closely to determine where colour management could fail and give rise to the observation that images from Adobe 1998 files are less colourful (i.e. lower colour gamut) than those from sRGB.

  1. Cameras
  2. Image manipulation software
  3. Print making

Cameras

By the use of EXIF readers, image files may be interrogated to determine what colour space the image is in, and whether this is communicated via a "flag" or an ICC profile. A small sampling of cameras revealed that where a "flag" and an ICC profile were present, they were not always consistent. One camera was found that would embed an Adobe 1998 ICC profile for images captured in Adobe colour space, but would raise a contradictory sRGB "flag". Clearly if this image file was used with equipment/software that only reads "flags" in EXIF data or equipment/software that gives priority of EXIF "flags" over ICC profiles, false assumptions about the colour space of the image would be made. It is not appropriate to mention the make and model of the camera, as the manufacturer has been advised of the findings, and subsequent production may no longer exhibit this effect.

Image Manipulation Software

Software from a number of camera manufacturers was examined. By use of specially "engineered" files with conflicting EXIF "flags" and ICC profiles, it was possible to determine if the software was both "flag" aware and ICC profile aware, and which has precedence. The software was used to "develop" the camera raw files into both sRGB and Adobe 1998 colour spaces. Additionally the software was used to open both sRGB and Adobe JPEG files, and to convert sRGB to Adobe and vice versa. When the resulting files were examined with the EXIF readers, it was discovered that software from one manufacturer would create files with conflicting ICC profiles and EXIF "flags". It is not appropriate to mention the manufacturer of the software as they have been advised of the findings.

Printmaking

To determine the impact of an Adobe 1998 colour space image being treated as if it were sRGB a little "Photoshop Work" was done. A digital representation of the Macbeth Color Checker Chart in Lab space was duplicated and converted into two new images, one in Adobe 1998 colour space, and one in sRGB. The colour profile of the Adobe 1998 image was then reassigned to sRGB. Both images were then converted back to Lab space and the Lab values of the coloured patches were determined from the "info panel". The Lab values of the patches were compared to the values of the original Colour Checker Lab image. Ideally the values should have matched the original. For the sRGB image the values matched perfectly for all colour patches, except the cyan patch. This was to be expected as the Lab values for the cyan patch lies outside the sRGB gamut. However for the Adobe 1998 image that had its profile reassigned to sRGB no patches matched the original. Some colours were notably lower in colour saturation and some exhibited significant changes in hue. On average the colour patches were 15% low in colour saturation. This is shown in the following charts.

Thus for those cameras and software that can produce Adobe 1998 colour space images with a sRGB EXIF "flag", we would expect lower saturation results when used in Adobe 1998 colour space than sRGB, i.e. a lower colour gamut and not the higher one "promised".

To confirm whether this occurs in practice, a survey of web based online photofinishers and mini-labs offering prints from digital cameras was conducted during 2005. The survey was confined to laboratories offering services for amateur photographers. A customer suite of images were used in the survey such that for each lab the following could be determined:-

  1. Was the lab "colour aware"?
  2. If the lab was "colour aware" was this based upon ICC profile, or EXIF "flags"?
  3. If the lab was "colour aware" which had priority embedded ICC profiles or EXIF flags?
  4. Could the lab handle TIFF as well as JPEG images?

Of the ten laboratories surveyed only one of them was "colour aware", and it was based upon the EXIF flag not the embedded ICC profile. Thus 90% of the labs surveyed would fail to treat Adobe 1998 images differently to sRGB images. Additionally the one lab that was "colour aware" would only treat Adobe 1998 images correctly if the image has the correct EXIF "flag" something which we have seen earlier is not guaranteed. The prints received back from the laboratories were measured with a spectrodensitometer and the L* a* b* values of coloured patches within the images determined. The results from the survey were in broad agreement with the predictions from the "Photoshop Work". On average across the non "colour aware" laboratories, the colour saturation from the Adobe 1998 images were 10% lower in saturation than the sRGB images. "Greens" in particular exhibit noticeable changes due to combination of lowered saturation and hue change. The "Photoshop Work" predicted a drop of 18% in saturation and a hue shift of -9.6º. The average result from the survey was a loss of 13% in saturation and a hue shift of -12.3º.

All labs could print RGB JPEG images, but out of the ten labs surveyed, only three could handle 8 bit TIFF images, and only one could handle 16 bit TIFF images.

CONCLUSIONS AND RECOMMENDATIONS

The consumer photofinishing industry is lagging behind the camera industry in its ability to handle images in multiple colour spaces and file formats. While more recent releases of software used in photofinishing printers are likely to become "colour aware" this still doesn't guarantee that Adobe 1998 images will be reproduced well, as not all cameras and their associated software use the correct EXIF "flags" to designate the colour space of images. Until this situation changes significantly, photographers may well be disappointed with prints from their digital cameras if they use the Adobe 1998 colour space, as compared to sRGB, lower gamut results not higher are likely to be obtained. Additionally not all laboratories can handle TIFF images. Unless a photographer knows that the entire imaging flow including print making is colour managed correctly, as is common in professional and commercial laboratories, the use of sRGB is a "safer" choice than Adobe 1998. Additionally as not all labs can handle TIFF images, minimally compressed JPEG images are a "safer" choice than TIFF images. These recommendations are not to be seen as opposition to the use of non compressed file formats or wider gamut colour spaces, but pragmatic recommendations based upon "real world" consumer/amateur photofinishing. A photo enthusiast may decide that working in Adobe 1998 colour space with an uncompressed file format and conversion to sRGB and a minimally compressed JPEG for printing by a consumer laboratory may give them the "best of both worlds".