CS3 Math Errors

RG
Posted By
Rags Gardner
Jun 2, 2007
Views
386
Replies
5
Status
Closed
I have previously documented some CS2 Adobe math errors, particularly in Lab 16-bit mode. The correspondence included a formal defect report, Adobe forum posts, and discussions with developers at trade shows. I have also posted the details here: Adobe Math < http://www.rags-int-inc.com/PhotoTechStuff/ColorCalculator/A dobeMath.html>

The errors were acknowledged by Adobe and I was hoping for a correction in CS3. This did not happen in the CS3 Beta. I provided feedback, but never got any response. I can now verify that it did not get fixed in the final CS3 product.

This week I was doing some extensive testing of my ACR color calibration scripts for a future update. I encountered a new problem. If I paint in an image with Lab values in 16-bit, the resulting colors are not correct. These are minor differences, usually when the a,b values are near Β–1 to +1. They are not only wrong in the Info Panels, they are wrong in the image file itself. Exploring this, I employed both the eyedropper and the color sampler toot. Both gave different answers, both wrong. These are filled color patches, so there is no color noise involved.

So I went back to CS2 and ran the same tests with the same image. The Info Panels were wrong as always, but consistent. When I re-painted the patches with CS2, all the image file color values were correct, exactly as they should be. When I read this corrected image in CS3, the color values are still correct. The CS3 Info Panels are still incorrect and still inconsistent.

The bottom line and reason for this posting is that CS3 not only did not correct the known errors, it has introduce some new ones, painting with Lab colors. Today, there was an update available that included something called Adobe Color Common STI. I donΒ’t know what this is, but I did the update and re-tested. No change.

The problem is with conversions between true 16-bit and Adobe 15-bit. The unambiguous 15-bit RGB midpoint becomes ambiguous with Lab values. That would be a trivial factor if the conversions were consistent. Apparently they are not. This needs to be addressed at the root cause. The Info Panels and histogram are not the cause, just symptoms. Random patches in the wrong places will only exaggerate the problems, not fix them.

Let me also say these errors are not big enough to create issues in a normal workflow. But they do compound quickly with multiple color conversions. Of course we all already know we should avoid this if possible.

While weΒ’re talking about CS3 problems, there are two others I have seen.

If the Bridge is opened and I do a cut and paste with some unrelated application it frequently hangs. If I then close the Bridge the application recovers nicely. BTW, I have seen this with more than three unrelated Windows applications.

Sometimes Photoshop will simply close and disappear. No message other than the offer to send a problem report to uncle Bill. At least twice, this happened when Photoshop was busy running a script and I simply opened the ExtendScript tool to work on a different script.

All in all, CS3 seems more stable than CS2 at the same time in the product life cycle. Good work folks.

Cheers, Rags πŸ™‚

Must-have mockup pack for every graphic designer πŸ”₯πŸ”₯πŸ”₯

Easy-to-use drag-n-drop Photoshop scene creator with more than 2800 items.

RB
Robert_Barnett
Jun 2, 2007
I know the transform tools in Photoshop have rounding errors. Back in the Photoshop CS1 days Russell Brown even made a script that used this "defect" to create some cool effects. It was called Shake, Rattle and Roll. Unfortunately, he never bothered to update that script for CS2 or CS3. I still have it if anyone is interested.

Robert
RG
Rags Gardner
Jun 2, 2007
Robert,

These are not rounding errors. Rounding errors with 16-bit integers are about 0.004. For 15-bit they are about 0.008. Floating point math is good to about 14 decimal places (depending on the hardware). These errors are in the range of 0.5 to 1.2.

Just FYI.

Cheers, Rags πŸ™‚
RB
Robert_Barnett
Jun 4, 2007
Never said they were rounding errors. Just commenting that the "faulty" math you were speaking off is no surprise since Photoshop also has rounding errors.

Robert
JT
Joel_T_Goodman
Jun 28, 2007
Rags, are you saying that you actually got the Color COmmon STI update to install? I’ve tried over 4 or 5 times and it never finishes!!! (I even let it go over night and it still didn’t get half way)
RG
Rags Gardner
Jun 29, 2007
Joel,

The update took less than two minutes. No problems.
Windows XP/Media.

Cheers, Rags πŸ™‚

Master Retouching Hair

Learn how to rescue details, remove flyaways, add volume, and enhance the definition of hair in any photo. We break down every tool and technique in Photoshop to get picture-perfect hair, every time.

Related Discussion Topics

Nice and short text about related topics in discussion sections