ezpkns retouching wrote:
I was in a Victoria's Secret the other day and this 4 foot tall poster had a fake blue sky multiplied lazily all over the edges of the model. Probably looks fine at catalog-size.

Yeah, it pisses me off to go in there and see the 4 and 8 foot banners with obvious gaussian blur as the sole manner of skin work, poorly masked, and not an iota of concern given to the lack of texture.  I completely understand that they have a high optempo and aren't hiring folks like Bob to do their retouch work, but when you can learn as quick and better techniques on a forum like this, it's embarrassing to walk around the store as a photographer.  It's not like the models have bad skin, don't have great makeup, etc.

Edit to admit bias:  I'd probably give up parts of my anatomy to shoot Alessandra Ambrosio, so maybe I'm just jealous.

Snap2 wrote:
or you can apply a Curve to the High Pass layer using different blending modes-its still the same image

This is part of what I meant. This curve when applied to a High pass layer-




Precise output clipping values are 62 and 193 on an 8 bit image. The HP is then set to Linear light, same as with your Apply Image. The shadow and highlight clipping that occurs when HP is taken to 50% is reduced, this must be a biproduct of the way photoshop determines opacity. So High Pass is not so crap after all?

Sean Baker wrote:
As a smoothing operator, the sporadic artifactual divets…

!! what?

Snap2 wrote:
Ill make a contribution to the great subject - When I said in post 81


This is part of what I meant. This curve when applied to a High pass layer-

Precise output clipping values are 62 and 193 on an 8 bit image. The HP is then set to Linear light, same as with your Apply Image. The shadow and highlight clipping that occurs when HP is taken to 50% is reduced, this must be a biproduct of the way photoshop determines opacity. So High Pass is not so crap after all?

!! what?

My beef with high pass isn't to do with its contrast level - indeed as you note there are plenty of ways of dealing with its 'doubled over' contrast values.  Where I've laid out my arguments against it is in that you can't reliably reconstruct your image from it and GB, while they are purported inverse operators.  Try it yourself if you need the verification - three copies, one original on top set to Difference, one copy High Passed at 30px or whatever you choose - contrast dropped by your choice of techniques and in Linear Light, bottom layer GB'd at the same radius as the HP layer.  You'll see that the image isn't a perfect reconstruction, as we've been discussing likely because of the speed optimizations which Adobe included sometime before CS was developed / released.

As to your other question, HP results in sporadic pixels which are not of an accurate value - most often well off the 50% gray they should be as they aren't really in areas of contrast.  Were HP a true inverse of GB, with GB being the culprit of inaccuracy as was suggested, we would see the inverse of these pixels in the smoothed data which it outputs.  As we do not see it, I don't believe that the problem lies in GB itself, but rather the shortcuts taken in the HP filter's code to keep its speed up.

toan thai photography wrote:

this one took me over 16hrs. d&b and cloning. no blur and high pass were used. most other beauty images took equally as long.

http://www.redlanternstudio.com/gallery/17.html

http://www.redlanternstudio.com/closer_look/8.html (detail)

Whatever the argument, whatever the process..I think that you are doing some great work here. The before and after shots prove that. This is some expertise I wish I had..

Sean Baker wrote:
likely because of the speed optimizations which Adobe included sometime before CS was developed / released.

Now that makes sense. You are suggesting that more 'accurate' math, would slow the filter down? High Pass was in Photoshop 5 first version I used, and was probably brought in versions 3 or 4

I have Analysed both the different method, both with difference and their histograms - and you are completely right with a difference blend the Apply does look to be more accurate. HP loses it only on very severe highlights and shadows. Both filters destroy the image borders, but differently - which suggest a  different mathematical method.

Sean Baker wrote:
Where I've laid out my arguments against it is in that you can't reliably reconstruct your image from it and GB, while they are purported inverse operators.

To the best of my knowledge, Adobe never claimed they were inverse operators. They perform opposing functions granted--low pass and high pass--but there are numerous ways to approach both, and there's no obvious obligation for Adobe to have coded them symmetrically. (Obvious, given that until the product had been out for over 10 years, HP was rarely used other than for image analysis, and even now it's not in widespread use.)

I'm not 100% positive how whether these apply to all high/low/bandpass filters, but differences in corner frequencies and order will have a huge impact on how reciprocal a pair of filters will behave in electronic designs, even for the same filter design. (A 2nd order (6db) highpass isn't reciprocal to a 1st order (3db) lowpass, for example.) And different designs (butterworth/bessel/etc) will have significantly different side-effects, even though they're intended to perform the same primary tasks.

Snap2 wrote:

Now that makes sense. You are suggesting that more 'accurate' math, would slow the filter down? High Pass was in Photoshop 5 first version I used, and was probably brought in versions 3 or 4

I have Analysed both the different method, both with difference and their histograms - and you are completely right with a difference blend the Apply does look to be more accurate. HP loses it only on very sharp highlights and shadows. Both filters destroy the image borders, but differently - which suggest a  different mathematical method.

If need be, you can retain your borders properly by expanding the image size (canvas size, to PS doctrine) by the radius you employ for the operator.  This will give it enough data to 'bite' on and retain the original image.  Using the Apply Image techniques, however, this shouldn't be necessary for the manner in which it approaches the problem (if technically inaccurate at the edges).

Kevin_Connery wrote:
To the best of my knowledge, Adobe never claimed they were inverse operators. They perform opposing functions granted--low pass and high pass--but there are numerous ways to approach both, and there's no obvious obligation for Adobe to have coded them symmetrically. (Obvious, given that until the product had been out for over 10 years, HP was rarely used other than for image analysis, and even now it's not in widespread use.)

I'm not 100% positive how whether these apply to all high/low/bandpass filters, but differences in corner frequencies and order will have a huge impact on how reciprocal a pair of filters will behave in electronic designs, even for the same filter design. (A 2nd order (6db) highpass isn't reciprocal to a 1st order (3db) lowpass, for example.) And different designs (butterworth/bessel/etc) will have significantly different side-effects, even though they're intended to perform the same primary tasks.

Adobe LiveDocs wrote:
High Pass - Retains edge details in the specified radius where sharp color transitions occur and suppresses the rest of the image. (A radius of 0.1 pixel keeps only edge pixels.) The filter removes low-frequency detail from an image and has an effect opposite to that of the Gaussian Blur filter.

I grant that they state it is an 'effect' and consequently has multiple interpretations, but it's closer than my comfort level.  All the same, your point about different techniques having different side-effects and accuracies is a good one; I'm simply advocating that more accurate techniques could be included and I'd prefer having the option - largely out of geekiness but also out of marketability for the product line.  Clearly I am going to have to spend more time in my copy of Russ, though .