ACR 5.5 RC shifts Sony colours – a little

Adobe’s Camera Raw Release Candidate 5.5, which introduces global corrections for all Bayer sensor cameras requiring a differential value treatment between the two Green channels, has a visible effect on the colour of Sony Alpha file conversions.
To see this you need the right image, very good colour vision and a good monitor. To measure it – or at least view the differences made – you need any program which allows a Difference or Subtraction calculation to be made between two identical file exports, one with ACR 5.4 and the other with ACR 5.5. You must make the 5.4 (or earlier) test export before installing ACR 5.5, as 5.5 is now updated via an installer program which removes the earlier version. The old method (download the plug-in, replace earlier copy manually by finding the right directory) was scrapped with the change from 4.x to 5.x; it also makes it difficult to install ACR 5.x on any system where CS4 or an equivalent up to date Elements installation is not present.
The DNG converter 5.5RC also automatically replaces earlier DNG converters resident in your Applications directory. If you want to keep these, move them before running the installer. These comments apply to Mac OSX, Windows may differ but Adobe tends to be consistent.
The differences are small, affecting all three channels (RGB) and can be seen in the two filetypes I have tested – Alpha 900 and Alpha 380. The Alpha 380 file was tested using a MacBeth Color Checker studio image, and the 900 file was an outdoor scene with lots of subtle green in grass which tended to render slightly yellowish in ACR 5.4, requiring more than the expected reduction in WB setting (relative to Auto or As Set) to neutralise. While the trend is not totally removed, for the same camera profile and default calibration there is a visible improvement in the grass colour.

To see the change in pixel values, the identical exports using 5.4 and 5.5 were opened in separate Photoshop windows, and Calculations was used to create a New Document using Difference as the operator. This appears black to the eye, but moving the Levels sliders (shown inset) quickly reveals the evidence. This is not an effect created by any other cause – doing a calculation using two copies of the same file also produces a black image, but Levels shows that there are no value differences detected:

With the A900 file, the extent of the differences detected on the histogram of the new document was roughly up to level 14 G, level21 R and level 47 B (as expected, changes to the way the green channels are processed has most effect on the R and B output). With the A380, the value range on the calculation document was to around level 36 G, 42 R and 47 B. This indicates that the values of some pixels are changed by as much as 47 (steps on 0-255 scale) (blue channel) when converting using ACR 5.5 compared to 5.4. However, the incidence of values changed by this amount is tiny. Most changes are in the order of 1 to 3 on the 0-255 scale, and the Difference calculation does not distinguish between insignificant shifts (in the high bit range) and important ones (anything happening in the darker tones).
You can download the Adobe Camera Raw and DNG Converter Release Candidate software free from:
http://labs.adobe.com/

You can download sample files (a reduced size 6 megapixel export from A900, and a full size 14 megapixel colour checker test from A380 – above, 5.5 conversion) from my pBase pages, and compare them yourself:
http://www.pbase.com/davidkilpatrick/image/116295636
This will take you to the first of four images. It is an AdobeRGB profiled JPEG and will appear yellowish and flat as viewed through sRGB default browsers. The second pair with the colour checker have been changed to sRGB and will look correct. They are much larger 14 megapixel images and also give a useful idea of the very accurate colour rendering associated with the Sony 14.2 megapixel CCD as used in the A350 and A380.
While I have not yet had a chance to assess ACR 5.5 and its changed Bayer conversion values on a wide range of files, it is possible that noise characteristics will be improved by the fine-tuning Adobe has done, and that ‘fixing’ the de-Bayer process brings with it other benefits.
– DK