anyone using a fully linear workflow?

[doug_cooper]

So I've been experimenting with a Photoshop workflow lately based

entirely in the <a

href="http://www.aim-dtp.net/aim/evaluation/gie/index.htm">AIM Linear

colorspace</a>

<p>

As advertised, it definitely provides significant improvements for

operations like sharpening, or other filters which operate on color

data from adjacent pixels and do so assuming linear color spaces. For

example, blurs are affected as well, although artifacts are less

noticable than with sharpening filters.

<p>

In order to get enough shadow detail, despite the claims on the AIM

site, I have needed to use 16-bits per pixel. Even with this, I'm

finding it often tricky to get any sort of good control of shadow

detail. All of the detail is packed into only a few code values -

especially when working with Photoshop's curve editor, which only lets

you edit in 256 values anyway (although it's interploating the extra

values for the purpose of applying the curve).

<p>

I'm now thinking of returning to a Gamma color based workflow simply

for more intuitive color adjustments.

<p>

At that point, I have a couple of options for doing my sharpening:

<p>

Wait until all color work is done, then flatten, convert to

linear/16-bits, and finally do a sharpen.

<p>

Or write a custom plug-in to provide a gamma aware UnSharp Mask

filter. (i.e. check the document's color profile, and convert the

colors into linear internally before sharpening, then convert back to

the document colorspace).

<p>

Has anyone else been working with a linear colorspace at all?

<p>

Is anyone aware of existing sharpening filters out there which have

built-in gamma compensation?

[sean de merchant httpw]

If you want to do things in linear gamma, then write an action that converts to 16 bit color, converts to linear gamma, calls Unsharp Mask but has the call checked so you have to enter the values and click okay, and then converts back to your working space. Or better yet, have it do this on a duplicate image and overlay as a new layer on the original.

 

Myself, I find Timo's sophomoric ranting tiresome and his idea of well sharpened severely overdone. His writings are best avoided if you care about what your images look like.

[marshall]

Timo's widely regarded as a lunatic, though he does seem to have some dedication to his ideas and some knowledge of what he's doing. Then again, I'd thought that Poynton, Fraser, and others had disproven all of his theories effectively a while ago.

 

I have notice that there seems to be linear editing done with raw conversions, and don't fully understand it. I haven't seen anyone doing linear editing after raw conversion.

[kelly_flanigan1]

Give us dell operators, matices, integrals. That linear stuff is for the lit school and arts chaps!

[doug_cooper]

For what it's worth, I use an fully linear color workflow in my profession of computer graphics film production. There is definitely merit to the linear colorspace mathematically -- but for Photoshop it doesn't seem as well integrated as it could be. (Many tools aren't really designed well to work with that kind of color).

 

As for rebuttals to the value of Timo's rants (which yes, can be rather tiresome at times), I'd love to see some links to other points of view. Colorspaces is something that is of definite interest to me, and applicable to both my profession in computer graphics as well as my hobby of photography.

 

Thanks for the input.

[janne]

How about this:

 

Color space transformations are locally linear, because they are smooth. If you have an image with high resolution, the image is locally (for sharpening etc.) compact even in the nonlinear color space. Linear operators commute (are interchangeable), and sharpening is a linear operator, so it does not really matter wheather you do sharpening or color space transformation first.

 

I know things are not that simple in practice. Most of all, images do have edges of objects and other non-smooth features that sharpenings tries to emphasize. But maybe there is a grain of truth in this?

 

(Disclaimer: I haven't read AIM, just the conversation above, so this may be off base.)

[sean de merchant httpw]

Even if a pair of linear operators commute on the real numbers, it in no way implies that they would commute when working on a discrete subset of numbers( i.e., a digital image).

 

With images an order of operations from left to right can mean that

 

x - y + z <> x + z - y

 

i.e.,

 

(255 - 55) + 5 = 200 + 5 = 205 <> 200 = 255 - 55 = (255 + 5) - 55

 

Image processing on discrete data types (8 bit color, ...) is not well defined number systems. The example above illustrates what I would term clipping. Mathematically, this yields that even addition on images lacks associativity. For most operations that are typically commutative this type of example can easily be constructed too.

 

The only real value of such mathematical formalisms is that they teach a person how the well defined interrelationships between elements feels and then they can take and relate those feelings to how the behavior of the numbers in images feel and use that to intuit what is likely to happen.

 

In practical terms, consider using Curves to darken an image and then apply aggressive sharpening in Lighten mode to an image. The reverse could very easily cause clipping and an unrecoverable loss of highlight detail that did not doing it the first way.

 

Nor is sharpening a necessarily linear operation and visually it may be an undesirable property.

 

Nonetheless, if one only considers an 8 bit color image in the midtones and performs the operations in 16 bit color, then the above has some value. But with just 8 bit color it has no real use as differing gammas can cause round off errors.

 

In the end though, it is his ignorance of the underlying structure and a blatant unwillingness to try to grasp them as I learned years ago from his trolling on Usenet that lead to such disrespect for him (well, that and his habit of ruining images in the name of doing it right ;o). It is very annoying to point at that assumption A is wrong and have them give you an overwhelming amount of unrelated details to show why it could not be wrong. It is reminscent of an advanced form of the foolishness involved in calling lens flare aliens/ghosts ( http://www.google.com/search?q=orbs ).

 

some thoughts,

 

Sean