RetouchPRO

Go Back   RetouchPRO > Technique > Photo Retouching
Register Blogs FAQ Site Nav Search Today's Posts Mark Forums Read


Photo Retouching "Improving" photos, post-production, correction, etc.

Automatic Skin Smoothing.

Reply
 
Thread Tools
  #121  
Old 11-03-2010, 08:58 AM
bakerser bakerser is offline
Member
 
Join Date: Jan 2008
Posts: 91
Re: Automatic Skin Smoothing.

Quote:
Originally Posted by ShadowLight View Post
the only problem in your "test" is that you didn't make the comparison of what the application does.
Again, please give me a detailed explanation of what was different about the Smart Object I created and one which comes out of your app. I'll adjust the test files and repost.
Quote:
Originally Posted by ShadowLight View Post
I didn't lie about the results, but couldn't attach the PSD directly.
I can host the files for you if you like.
Quote:
Originally Posted by ShadowLight View Post
I didn't test the custom filter (it's hard to use it in dynamic implementation), but if it works as well as contrast -50 (as adj. layer) it should be viable alternative.
Please, explain. It applies just like the other filters you're using, so there shouldn't be anything preventing it from working just as well.
Reply With Quote top
  #122  
Old 11-03-2010, 09:21 AM
ShadowLight ShadowLight is offline
Senior Member
 
Join Date: Oct 2009
Posts: 397
Re: Automatic Skin Smoothing.

Quote:
Originally Posted by bakerser View Post
Again, please give me a detailed explanation of what was different about the Smart Object I created and one which comes out of your app. I'll adjust the test files and repost.
I can host the files for you if you like.

Please, explain. It applies just like the other filters you're using, so there shouldn't be anything preventing it from working just as well.
The slider "Strength" controls the "fill" levels, so whatever it is (alpha, fill, contrast -50, custom filter) it will have to be easily and dynamically controlled.
(personally I find control of the strength quite beneficial.

I'll post you a psd file in a few min.
Reply With Quote top
  #123  
Old 11-03-2010, 09:31 AM
ShadowLight ShadowLight is offline
Senior Member
 
Join Date: Oct 2009
Posts: 397
Re: Automatic Skin Smoothing.

here is the psd:
SkinSmooth.vs.Contrast-50.psd
Attached Images
File Type: jpg SkinSmooth.vs.Contrast.jpg (68.2 KB, 32 views)
Reply With Quote top
  #124  
Old 11-03-2010, 09:32 AM
bakerser bakerser is offline
Member
 
Join Date: Jan 2008
Posts: 91
Re: Automatic Skin Smoothing.

Quote:
Originally Posted by ShadowLight View Post
The slider "Strength" controls the "fill" levels, so whatever it is (alpha, fill, contrast -50, custom filter) it will have to be easily and dynamically controlled.
It just stacks with the other two Smart Filters (GB & HP), so there shouldn't be any interference with your Fill / Opacity adjustments. And (done right) it eliminates any chance of clipping.
Reply With Quote top
  #125  
Old 11-03-2010, 09:46 AM
ShadowLight ShadowLight is offline
Senior Member
 
Join Date: Oct 2009
Posts: 397
Re: Automatic Skin Smoothing.

Quote:
Originally Posted by bakerser View Post
It just stacks with the other two Smart Filters (GB & HP), so there shouldn't be any interference with your Fill / Opacity adjustments. And (done right) it eliminates any chance of clipping.
if this is the case it is implementable
(although I have a hunch that using Fill will be still much faster)

however as you see there is no "clipping" with the fill either
as far as I see in the test image the results are identical

Last edited by ShadowLight; 11-03-2010 at 10:04 AM.
Reply With Quote top
  #126  
Old 11-03-2010, 10:19 AM
bakerser bakerser is offline
Member
 
Join Date: Jan 2008
Posts: 91
Re: Automatic Skin Smoothing.

Quote:
Originally Posted by ShadowLight View Post
if this is the case it is implementable
(although I have a hunch that using Fill will be still much faster)
If the Adobe code is done right, they should be equally fast.
Quote:
Originally Posted by ShadowLight View Post
however as you see there is no "clipping" with the fill either
as far as I see in the test image the results are identical
I'll let you know what I see when I get back to a machine with PS.
Reply With Quote top
  #127  
Old 11-03-2010, 01:32 PM
klikibo klikibo is offline
Junior Member
 
Join Date: Oct 2010
Posts: 1
Re: Automatic Skin Smoothing.

thanks for share...
Reply With Quote top
  #128  
Old 11-03-2010, 03:18 PM
bakerser bakerser is offline
Member
 
Join Date: Jan 2008
Posts: 91
Re: Automatic Skin Smoothing.

Quote:
Originally Posted by ShadowLight View Post
however as you see there is no "clipping" with the fill either
as far as I see in the test image the results are identical
You are right in as much that there were no pixels clipped in this image, and at these settings. Which is not to say that there aren't still some errors in the result due to the way the blend takes place (in this image, at these settings these errors were as high as 3/256/channel away from a 'perfect' version). Moreover, as I explained earlier, the timidity of that error level will not be the case for every image and all settings combinations, regardless of limitations placed on the interface(*).

In the grand scheme, it's an algorithm design question. Where the current implementation is easy and has an acceptable error level in best case scenarios (generally - you might disagree given your dislike for the 'noise' of -50C versions), that error level climbs precipitously in the worst case to where it becomes truly interfering with one's work. OTOH, with any of the four slightly less easy (but still pretty easy, all things considered!) implementations offered - your own preference obviously being for the most recent, which should amount to only a single additional line of code - you can get an error reduction at all levels, particularly in those 'worst case' images. From a business perspective, this translates to better user satisfaction, increased product 'buzz', and reduced need for customer / product support.

However implemented, the better solution is and has always been to halve the image contrast before HP + GB are performed so as to reclaim the lost accuracy for all images and to avoid really bad errors in the case of image + setting combinations which have the bulk of their contrast in the band selected for removal. Almost paradoxically, this will give a finer granularity for tuning the 'strength' of the effect, which I agree is a worthwhile design inclusion - it streamlines the workflow!

* -- Seriously, who likes artificially imposed UI constraints? Since PS CS (1!), I've not met one user who liked the 250px GB / HP max. It's arbitrary. It holds people back when they want to do something. In the case of this tool, from using it effectively on high-grade (and high MP) beauty images.
Reply With Quote top
  #129  
Old 11-03-2010, 05:27 PM
ShadowLight ShadowLight is offline
Senior Member
 
Join Date: Oct 2009
Posts: 397
Re: Automatic Skin Smoothing.

Quote:
Originally Posted by klikibo View Post
thanks for share...
you are welcome.

Quote:
Originally Posted by bakerser View Post
You are right in as much that there were no pixels clipped in this image, and at these settings. Which is not to say that there aren't still some errors in the result due to the way the blend takes place (in this image, at these settings these errors were as high as 3/256/channel away from a 'perfect' version).
As I explained earlier, you will never get a "clipped" image. If you did the proper testing you would have notice that the "clipping" actually leaks details, so even if you manage to get a "clipping" it will be in the reverse correction which will let details from the original image in. (as I showed you with an example earlier)



Quote:
Originally Posted by bakerser View Post
In the grand scheme, it's an algorithm design question. Where the current implementation is easy and has an acceptable error level in best case scenarios (generally - you might disagree given your dislike for the 'noise' of -50C versions), that error level climbs precipitously in the worst case to where it becomes truly interfering with one's work. OTOH, with any of the four slightly less easy (but still pretty easy, all things considered!) implementations offered - your own preference obviously being for the most recent, which should amount to only a single additional line of code - you can get an error reduction at all levels, particularly in those 'worst case' images. From a business perspective, this translates to better user satisfaction, increased product 'buzz', and reduced need for customer / product support.
The "noise" is not the big problem with Contrast -50. The fact that if it is not as an "adjustment layer" and will depend on the background luminosity is. The fact that if you change the background, the resulting Contrast -50 is different for the whole image. Even if it were stable, Contrast -50 is not Smart-Object friendly.
(however the "custom filter" might be actually a viable alternative)

To check a "worse case scenario" try to upsize your image first. (or upsize the PSD I supplied, and using SkinSmooth, after the resize adjust the sliders back to "max" positions), and you will see there is still the same result as with the small image. (I tried it with 4400x3300px and it had the same results).

About that "clipping" you refer to... from the test it looks like it is not lum-related but also freq. related. It does not affect the light/dark areas but the higher frequency ones only.

Strictly speaking, if you adjust and lose half of your information (what you do with C -5o first) it will be less precise if you have a wider dynamic range to apply the corrections. Using Opacity 50% will lower the signal strength in half and should be actually more precise than losing DR for corrections first (using C -50).
(unless you apply the C-50 after you have applied gb/hp)

Here from what I can tell, the HP algorithm fails. The so called "clipping" depends on the image-size and the used HP:r , not on luminosity. (ie HP:r to image size -> defines the error)
So the largest problem will show when you use a thumbnail with HP:r=250 (those are your settings).
Otherwise as long as this computational error is not triggered, the fill/opacity method should be more precise as it uses larder DR for computing the gb and hp.

Quote:
Originally Posted by bakerser View Post

* -- Seriously, who likes artificially imposed UI constraints? Since PS CS (1!), I've not met one user who liked the 250px GB / HP max. It's arbitrary. It holds people back when they want to do something. In the case of this tool, from using it effectively on high-grade (and high MP) beauty images.
well... it is a limitation, but not quite in this case.
You know that SkinSmooth automatically corrects the values for the image size in order to always get the same effect with a given slider positions. And I just tested it with 8000x12000 px upsized image, and it is still in-bounds at maximum setting (although at the edge of hp:250).
Anything bigger will get cut by the r:250 limitation, but still this is quite large for a closeup portrait.
(and the max values are quite hard so in reality I doubt anyone will need to use them)... so imho r:250 is still quite sufficient.

Last edited by ShadowLight; 11-03-2010 at 05:35 PM.
Reply With Quote top
  #130  
Old 11-03-2010, 07:17 PM
bakerser bakerser is offline
Member
 
Join Date: Jan 2008
Posts: 91
Re: Automatic Skin Smoothing.

Quote:
Originally Posted by ShadowLight View Post
As I explained earlier, you will never get a "clipped" image. If you did the proper testing you would have notice that the "clipping" actually leaks details, so even if you manage to get a "clipping" it will be in the reverse correction which will let details from the original image in. (as I showed you with an example earlier)
And I've repeatedly explained that this is not the case. Your 'leaked' detail is a symptom, not a success. This is evidenced by having errors in excess of 3/255 with even so easy an image to work with - were it appropriately separating the data and only retaining what it really should, it would be identical to an 'Apply Image' version of the same thing. I've included below a PSD showing everyone watching the difference - as before with a contrast boost for quick visibility; use the measurement tools for real knowledge. And of course, as a rather 'best case' example, the error here is minimal relative to what may occur in regular usage.
Quote:
Originally Posted by ShadowLight View Post
The "noise" is not the big problem with Contrast -50. The fact that if it is not as an "adjustment layer" and will depend on the background luminosity is. The fact that if you change the background, the resulting Contrast -50 is different for the whole image. Even if it were stable, Contrast -50 is not Smart-Object friendly.
(however the "custom filter" might be actually a viable alternative)
Are you still stuck on the difference in means? You should check out the previous PSD if so - it's irrelevant to this process.

Of course, if you prefer the ease of implementing the Custom filter option, there's certainly nothing wrong with that. I among many would be excited to see the change made.
Quote:
Originally Posted by ShadowLight View Post
To check a "worse case scenario" try to upsize your image first. (or upsize the PSD I supplied, and using SkinSmooth, after the resize adjust the sliders back to "max" positions), and you will see there is still the same result as with the small image. (I tried it with 4400x3300px and it had the same results).
Well now that wouldn't make much sense. The problem that I'm describing is the relative frequency+contrast distribution of the original image. Scaling the image up doesn't change that relationship - it just changes the absolute values which get used in the separation; nothing more. So, while you do see the same result, it only proves that you didn't change the actual test conditions. A different image, or one which has had its local contrast enhanced in advance of testing would actually change the frequency+amplitude relationship; scaling does nothing to that.
Quote:
Originally Posted by ShadowLight View Post
About that "clipping" you refer to... from the test it looks like it is not lum-related but also freq. related.
That's what I've been saying - contrast over area (2D spatial freq + amplitude if you will).
Quote:
Originally Posted by ShadowLight View Post
Strictly speaking, if you adjust and lose half of your information (what you do with C -5o first) it will be less precise if you have a wider dynamic range to apply the corrections. Using Opacity 50% will lower the signal strength in half and should be actually more precise than losing DR for corrections first (using C -50).
This is a necessary tradeoff and one which gives superior results to any alternative, short of using two layers to accomplish one bandstop (which would be easy to code btw ). What you propose will always run into accuracy and clipping issues, and a broadly distributed (tonally and spatially) precision error is far more aesthetically pleasing than an absolute loss of tone + detail at either end of that tonal range.
Quote:
Originally Posted by ShadowLight View Post
Here from what I can tell, the HP algorithm fails. The so called "clipping" depends on the image-size and the used HP:r , not on luminosity. (ie HP:r to image size -> defines the error)
Please read the HP Sucks thread or my blog for a longer explanation of this. As I've been telling you for the last few pages, this is a matter of contrast (so, 'lum' in most of your own writing) distributed across spatial range defining the local contrast. A local contrast value in excess of 50% of the total image contrast (tonal) range will result in clipping. This isn't a failure of the HP filter, it's algorithm, or anything else - it is nothing more than our exceeding of the working image's bit depth. When I said "High Pass Sucks," I was wrong - how we were using High Pass sucked.

If you want to see that HP can retain all detail, go to my blog and read about the 3-layer frequency separation (related to my statement earlier when discussing the ability to create a 'perfect' bandstop if we used a second layer).
Quote:
Originally Posted by ShadowLight View Post
well... it is a limitation, but not quite in this case.
You know that SkinSmooth automatically corrects the values for the image size in order to always get the same effect with a given slider positions. And I just tested it with 8000x12000 px upsized image, and it is still in-bounds at maximum setting (although at the edge of hp:250).
Right, but while you're correct to think that an image-size independent settings guide is good for novice users (at least, 'novice' WRT FS), you're failing to account for the amount of image area covered by the skin in all images which might be used. We spent some time talking about this same thing in that same HPS thread attempting to derive a good formula which could apply to any image in a semi-automated manner and find just such a range of values. Ultimately, though, we came to the conclusion that a very, very advanced UI would be needed in order to do it right (something beyond what can be done without writing a full plugin, and still complex at that). As it is, the current restrictions lock individuals retouching wedding shots (in which skin may account for only 1% of the image area) into the same available sizes range as a beauty and detail photographer for whom the face may be well larger than the image area.
Quote:
Originally Posted by ShadowLight View Post
so imho r:250 is still quite sufficient.
That's because you're still fairly new to the FS thing and haven't gotten outside the box with it yet. All things in time .

PSD
Reply With Quote top
Reply

  RetouchPRO > Technique > Photo Retouching


Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is Off
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off


Similar Threads
Thread Thread Starter Forum Replies Last Post
Glamour retouch challenge here!!!! superkoax Photo Retouching 332 02-08-2011 09:52 PM
Automatic Skin Tone Retouching - Seeking feedback Pixarra Photo Retouching 15 11-09-2010 01:38 AM
Skin Smoothing Technique (Again) Novi Photo Retouching 10 10-16-2008 04:25 PM
need help doing pageant eyelashes and skin smoothing w/ photoshop cs sandylavallie Photo Retouching 3 05-06-2005 08:53 AM
skin smoothing action john_opitz Photo Retouching 10 08-21-2002 10:48 PM


All times are GMT -6. The time now is 05:46 PM.


Powered by vBulletin® Version 3.8.4
Copyright ©2000 - 2017, Jelsoft Enterprises Ltd.
SEO by vBSEO ©2011, Crawlability, Inc.
Copyright © 2016 Doug Nelson. All Rights Reserved