LinearSharpen and Conversion Engine

[Roger_Cavanagh]

Following Pekka's remarks about the ACE vs Microsoft conversion engine, I tracked down the link on Timo's site. (http://www.aim-dtp.net/aim/photoshop/v6/slope-limiting/index.htm, if anyone wants to read it).

Timo raises a possible issue with image quality and shadows when ACE is used in profile conversion rather than the Microsoft Engine. So far, however, I haven't found anyone else who recommends using the MIS engine instead of ACE, but I'm still googling. :)

I did a few experiments with D30 images and there is, indeed, a difference between the two. How noticeable seems to depend on the subject matter. Certain shades of red seem to be more affected than other colours. I didn't see the additional noise and loss of detail that Timo's example has.

As LS D60 has several profile conversions compared to a maximum of one in LS342 or two, if sRGB for web is counted, I looked at the effect here. Here is the LS D60 no sharpening output:

http://www.pixelpixel.org/images/linked/cdp/lsd60/craig1-lsd60.jpg

Note: this D60 image was kindly provided to me by Craig Zendel.

I then processed a second copy of the image changing the conversion engine to ACE. Both images were converted to 8-bit and merged as separate layers in a new document. Layer blending was changed to difference. This gives a black image when pixels have the same RGB values in each layers, and is progressively brighter the greater the difference.

I then created a selection using the magic wand on almost black (1, 1, 1) with a tolerance of 1, i.e., where RGB was the same or almost the same. Inverting this selection gives the parts of the images that are most different:

http://www.pixelpixel.org/images/linked/cdp/lsd60/craig1-lsd60-diff.jpg

The differences are biggest in the G and B channels, sometimes as much as 32 points on the histogram.

Frankly, I'm not sure how much practical significance this is, except it does show that an apparently small change can have a big impact.

I offer the topic up for further comment, and I will continue to research.

Regards,

[T a z]

Thanks for doing such an involved test, Roger. I have long wondered about Timo's assertion that the Microsoft engine was superior. However, I never could quite see the problems he writes about. They very well could be there...but I've just not been able to confirm them for myself (not yet, anyway).

Your test does indeed show there's a difference in the way the image is converted, but like you, I'm not sure what this really tells us.

BTW, in case you aren't familiar with Timo, you must realize that some of his theories and assertions often go quite a bit against the grain of the prevailing mainstream thought in the color management field. His "battles" are legendary over on the Adobe forum. Mind you, I'm not at all saying his work and his theories are quackery or anything like that...only that he does find himself almost always on the opposite side of the theoretical fence.

Actually, for awhile I used his "Native PC" profile instead of sRGB for images that were meant for the web. The basic idea behind his profile is that the vast majority of computer monitors are not calibrated. He insists that sRGB images will not look right when viewed with a non-calibrated display due to the Gamma level of sRGB.

Interesting theory, but one that I must say I did see for myself. In my tests, on a non-calibrated monitor, his profile produced a more pleasing image than one that was converted to sRGB. However, on a calibrated display, I found sRGB to be more pleasing.

So...what does all this have to do with your experiment? Nothing! :) It's just that I find Timo's theories interesting and it's nice to see someone attempt to test them out.

-Taz

[Roger_Cavanagh]

Hi Taz,

Yes, I was aware that Timo often takes contrary views. But he does produce some worthwhile stuff, although some of what he says does go over my head. :)

I spent a couple of hours going through google links and still couldn't find anyone who was not recommending ACE. And related to that there were several votes for using perceptual intent rather than relative colorimetric for RGB to RGB conversions.

Cheers,

[Roger_Cavanagh]

I posted a query about conversion engine differences over at Rob Galbraith's: http://www.robgalbraith.com/cgi-bin/ultimatebb.cgi?ubb=get_topic&f=30&t=000013

Yer man Andrew Rodney responded. He didn't exactly give chapter and verse, but said:

They shouldn't unless you hit a bug. The differences between CMM's should be no more than about 1-3% per color channel. I'd stick with ACE in all cases UNLESS you are planning to do colorspace conversions outside of an Adobe product that supports ACE and an Adobe that supports it as you'll end up with differences that could be a problem. IOW, stick with one CMM for all conversions. Since ACE isn't available outside of Adobe applications, this could be a factor if you're doing in RIP conversions or using other methods to convert a file (like a camera capture product).