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.

Frequency Separation is evil

Reply
 
Thread Tools
  #11  
Old 12-23-2014, 09:16 AM
shift studio's Avatar
shift studio shift studio is offline
Senior Member
 
Join Date: Feb 2009
Location: Calgary, Canada
Posts: 389
Re: Frequency Separation is evil

I want to add that frequency separation may not be appropriate for the cover of Vogue (still debatable), but a lot of work is being done that is not for the cover of Vogue.

Quote:
I don't consider it to be a separation, but a blur and a sharpen.
If your radius is large, you can also think of FS as edges/texture and colour/shape - even more so if you've done an asymmetric split.
I'm not really talking about faces or skin here, but other things. Working image elements broken down this way can be beneficial.

--shift studio.
Reply With Quote top
  #12  
Old 12-23-2014, 09:37 AM
Benny Profane's Avatar
Benny Profane Benny Profane is offline
Senior Member
 
Join Date: Mar 2006
Location: third stone from the sun
Posts: 582
Re: Frequency Separation is evil

Quote:
Originally Posted by shift studio View Post
I want to add that frequency separation may not be appropriate for the cover of Vogue (still debatable), but a lot of work is being done that is not for the cover of Vogue.



--shift studio.


And, of course, the dirty secret of most beauty retouching is that, the better quality of the model (and higher cost), the easier the retouching. There's really not much you have to do to some of those girls unless they were partying for two days before the shoot. Even then......
So, most Vogue covers, after the attention to detail given by the photographer, makeup artist, hair stylist, and others on set, are pretty easy to get right on the back end.
Reply With Quote top
  #13  
Old 12-23-2014, 10:13 AM
Flashtones Flashtones is offline
Senior Member
 
Join Date: Apr 2009
Posts: 956
Re: Frequency Separation is evil

Quote:
Originally Posted by klev View Post
That's not entirely true. Okay so you start out with whatever level of precision stored at 8 or 16 bits per channel. The bits there are essentially a theoretical maximum and do not issue any kind of guarantee, especially with the overall complexity of color engines. You actually lose quite a bit of precision in signal processing alone, given that the values have to be quantized to some digital precision at the camera level, after which you not only have channel interpolation but a series of several mappings between color spaces before you're even in photoshop.

At that point you still lose information with every adjustment. It doesn't necessarily change at 16 bpc, but the step sizes are narrower. Even if the information is not always more accurate, it can be more aesthetically pleasing if banding does not occur.

The idea behind frequency domains is that if the range of a function is in fact bounded, we can measure how often it repeats itself and depict the two dimensional surface mapping as a sum of two or more of these functions. That is probably a horrible explanation, but it's not one of my stronger areas.

Anyway back to what I was saying, you do in fact have finite precision, and from my perspective you do have the potential for an unstable work space (potential for loss of detail over minor actions) in certain circumstances. 8 bits leaves you with too large of a step size, so I wouldn't suggest that. At 16 you do still have two significant adjustments, and you still have the potential for distinct haloing on both passes. If you go to 32 it will take care of the haloing. You could in theory have a more significant loss if recombining areas where the difference in values is very small, but if its eventual destination is 8 bit output, it's insignificant.
Klev, this is a visual field and I stand by my statement, that you will not see a difference in a properly split image at the moment of the split. Sure, one can screw things up afterward, but that's operator error.

It's like converting a 16bit image to 8bit. You don't see a change at the point of conversion, only if you mishandle it thereafter.

ETA: When I say "seeing it" I mean with the ocular nerve, as opposed to knowing there's an imperceptible mathematical difference.

Last edited by Flashtones; 12-23-2014 at 10:32 AM.
Reply With Quote top
  #14  
Old 12-23-2014, 10:41 AM
3pco 3pco is offline
Junior Member
 
Join Date: Jul 2014
Location: Half Moon Bay
Posts: 14
Re: Frequency Separation is evil

Following this thread with interest, I realize I know very little about FS. All the tutorials I've found online limit themselves to a step-by-step approach, so I know how to do it but not why. Can someone point me to some material that explains what's going on und
er the hood here? Thanks!
Reply With Quote top
  #15  
Old 12-23-2014, 11:09 AM
Flashtones Flashtones is offline
Senior Member
 
Join Date: Apr 2009
Posts: 956
Re: Frequency Separation is evil

Quote:
Originally Posted by 3pco View Post
Following this thread with interest, I realize I know very little about FS. All the tutorials I've found online limit themselves to a step-by-step approach, so I know how to do it but not why. Can someone point me to some material that explains what's going on und
er the hood here? Thanks!
http://www.modelmayhem.com/forums/post/439098
Reply With Quote top
  #16  
Old 12-23-2014, 12:40 PM
klev klev is offline
Senior Member
 
Join Date: Jun 2010
Posts: 1,109
Re: Frequency Separation is evil

Quote:
Originally Posted by Flashtones View Post
Klev, this is a visual field and I stand by my statement, that you will not see a difference in a properly split image at the moment of the split. Sure, one can screw things up afterward, but that's operator error.

It's like converting a 16bit image to 8bit. You don't see a change at the point of conversion, only if you mishandle it thereafter.

ETA: When I say "seeing it" I mean with the ocular nerve, as opposed to knowing there's an imperceptible mathematical difference.
Ah at least I know you have a sense of humor. There are extremely perceptible errors. When you try to separate these things, your top layer in essence contains the edge data for the image. This can be useful for things such as faking texture on a flat area. My issue in that area is partly with the algorithm used. It causes significant haloing just like unsharp masking. I think that places some limits on the practical application there. The issue with applying it at 8 bits or just on data that doesn't have adequate precision is that you can end up with banding, at which point you would have to add noise to mask it.
Reply With Quote top
  #17  
Old 12-23-2014, 02:31 PM
Flashtones Flashtones is offline
Senior Member
 
Join Date: Apr 2009
Posts: 956
Re: Frequency Separation is evil

Quote:
Originally Posted by klev View Post
Ah at least I know you have a sense of humor. There are extremely perceptible errors. When you try to separate these things, your top layer in essence contains the edge data for the image. This can be useful for things such as faking texture on a flat area. My issue in that area is partly with the algorithm used. It causes significant haloing just like unsharp masking. I think that places some limits on the practical application there. The issue with applying it at 8 bits or just on data that doesn't have adequate precision is that you can end up with banding, at which point you would have to add noise to mask it.
Show me. Show me an image that visually deteriorated by doing a proper apply image separation.

As an alternative, I could take an image, do a separation on it, flatten it, and challenge you to tell which was the original and which the separated image, but then you could say I didn't choose a good example. So you show me your best proof.

And FWIW, this is yet another example of people throwing the term FS around without defining what they mean by it. From hanging around I know your distaste for PS's gaussian blur, and I concur. But you can do a FS around a surface blur, or any other variant of your choice.

That said, show me your worst.
Reply With Quote top
  #18  
Old 12-23-2014, 02:34 PM
AKMac's Avatar
AKMac AKMac is offline
Senior Member
 
Join Date: Dec 2010
Location: London and Argyll, Scotland.
Posts: 309
Re: Frequency Separation is evil

FS can be used in many ways and situations, and like any other tool it can be used well or badly. The end result is what matters, so if there is no evidence of it's use then surely it can't be "evil" to use it. The important thing about shortcuts is that they free up time to spend on the high-quality stuff. I often use FS in unimportant areas of an image if I need to meet deadlines and focus on the aspects that really need to be A1. Also it can be a very effective tool for things like fabrics where there is a strong texture/pattern with underlying larger creases that need removed. Doug asked for specific examples, so...in the series linked below I used FS extensively in the clothing, but there was no FS used anywhere else.
I also often use the technique to do the "heavy lifting" with certain problem areas, and then back it off and finish by hand.
http://www.akmacretouch.com/portfolio-book.php?id=238
Reply With Quote top
  #19  
Old 12-23-2014, 03:11 PM
Doug Nelson's Avatar
Doug Nelson Doug Nelson is offline
Janitor
 
Join Date: Aug 2001
Posts: 7,068
Blog Entries: 21
Re: Frequency Separation is evil

Quote:
Originally Posted by AKMac View Post
I also often use the technique to do the "heavy lifting" with certain problem areas, and then back it off and finish by hand.
A reasonable approach. Truly you have learned to dance with the devil

Thank you for posting an example.
Reply With Quote top
  #20  
Old 12-23-2014, 03:17 PM
klev klev is offline
Senior Member
 
Join Date: Jun 2010
Posts: 1,109
Re: Frequency Separation is evil

Quote:
Originally Posted by Flashtones View Post
Show me. Show me an image that visually deteriorated by doing a proper apply image separation.

As an alternative, I could take an image, do a separation on it, flatten it, and challenge you to tell which was the original and which the separated image, but then you could say I didn't choose a good example. So you show me your best proof.

And FWIW, this is yet another example of people throwing the term FS around without defining what they mean by it. From hanging around I know your distaste for PS's gaussian blur, and I concur. But you can do a FS around a surface blur, or any other variant of your choice.

That said, show me your worst.
I wasn't entirely discounting the idea of frequency domain operations, and google scholar has failed me on turning up a specific reference (I remember the authors, just not the article name).

I would have to create a synthetic example to show that, because I never preserved any results after loosely testing it. I haven't really tried surface blurring it, but you're right the crux of my issue here involves PS's gaussian blur as a basis for almost anything. I think it has some extreme shortcomings and would not be written that way today. Apart from that more stable versions have been tested in other things, such as grading HDR imagery. There's a really neat abstract on that here if you would like to see it.

Anyway with PS's gaussian blur the two passes still add up correctly or with a small enough error (most of the time with the possible exception of skies and seamless backgrounds) in 16 bpc, but my concern is that your detail layer does in fact contain halos along edges where the low frequency layer contains inverted halos. Unadjusted these add up correctly. When you start to modify one, you can start to see edge artifacts on some subsets of images due to the alteration of their offset. I found the headache of that + increased file size to be a burden.

My other issue is just that it generally irks me whenever something is talked about so generally as a tactic for achieving something visual rather than in a more contained sense of how a specific technical problem can be rectified.
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
Frequency separation: Really non-destructive? baryon Photo Retouching 27 10-09-2013 05:19 PM
question about frequency separation method Caravaggio Photo Retouching 12 07-17-2013 12:21 AM
"asymmetric" frequency separation? drode Photo Retouching 12 12-25-2011 02:55 AM
Action for Frequency Separation jonathan_k Photo Retouching 13 04-15-2011 01:17 PM
Extended frequency separation fraiseap Photo Retouching 16 12-31-2010 01:33 PM


All times are GMT -6. The time now is 08:09 AM.


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