|
Forums >
Digital Art and Retouching >
HighPass Sucks (+ solution)
Mask Photo wrote: I'll tell you tomorrow / the next day when I dredge it out and use it - offhand I'm not 100% sure. Apr 30 10 05:48 pm Link I just had a thought, while playing around with Photoshop CS5's new blending modes. The loss of detail while separating is easy to understand: the original highpass clips shadows and highlights because it doubles their values and having them clipped, divides them by two (50% opacity), the new version does the division first, meaning that something like 127 gets 63,5 and as we are dealing with integers, 63. So what we need is separation without any division :-) (or a loss free division). My idea now was to create three copies of the background layer. Blur the lowest one with Gaussian Blur (called it "GBlur") or any other blurring filter you want and go on to the next layer (we call that "Add" and the topmost layer "Subtract"). Invert "Add" and set its blending mode to "Linear Dodge (Add)". Merge all visible layers to a new layer using [Shift]+[Ctrl]+[Alt]+[E] and call that layer "Add merged", disable the "Add"-layer as well as the "Add merged". Now go on to the layer we called "Subtract". Change its blending mode to "Subtract" and once again merge visible to a new layer, called "Subtract merged". Disable the "Subtract" layer and put "Add merged" and "Subtract merged" into a group. It's important, that "Add merged" is on the bottom of the group, otherwise, there'll be clipping later! Now invert the "Subtract merged" layer, set the opacity to 50% and change the blending mode to "Subtract". Everything starts to look good, just... inverted. Because of that the next step is inverting the whole group by putting a new "Invert" adjustment layer above both layers. Set the group's blending mode to "Linear Light" and watch what happens: you've got a split. And from my testings so far it's loss-free :-) I uploaded a screenshot of the layer's palette so maybe it's more understandable :-) http://www.abload.de/img/sep1xl9.jpg Just wondering, what do you think of this? Best regards, Jonas :-) Btw. From what I've read Sean did some experiments on getting the "Subtract" mode in CS4 and earlier versions, so I'm pretty sure he'll be able to help out find ways for those as well :-) PS: I tested in 8Bit because the possible loss should be greater. PPS: You could as well merge the group, that way you'll get the normal split layers :-) PPPS: I found it :-). Instead of using "Subtract", just invert the layer and set the mode to "Linear Burn". Edit: Forget, what I just said... doesn't seem to work on tough images :-( May 09 10 04:06 am Link More on this when I finish the article proper, but to dovetail with what Jonas has said above (he is correct that the above does not get 100% results - very close, but not quite): One can get 100% reconstruction in 8bpc or 16bpc by using three layers to do so (vs. 2). I'll explain the math in the article, but here's the procedure: 1.) Duplicate the image to separate 3x (twice if you don't care to retain a copy of the original for comparison). For discussion, we will now call these the bottom, middle, and top layers. 2.) Blur the bottom layer as you see fit - for testing, do something intermediate to make sure you get enough that you should see loss if it's present. 3.) Disable the top layer; select the middle layer. 4.) Apply Image. Select the bottom layer as Source, RGB as channel, check the Invert box, choose 'Add' as your blend mode, 100% opacity, Scale 1, Offset 0. 5.) Disable the middle layer; select and show the top layer. 6.) Apply Image. Select the bottom layer as Source, RGB as channel, do not check the Invert box, choose 'Subtract' as your blend mode, 100% opacity, Scale 1, Offset 0. 7.) Show all layers. 8.) Set the top layer's blend mode to Linear Dodge. 9.) Set the middle layer's blend mode to Linear Burn. 10.) If desired, compare the merged result to the original - you should find 0/0/0 discrepancy. But like I said - more later. May 09 10 05:24 am Link DerW wrote: WoW, tested with 16bits and looks great. May 10 10 08:24 am Link Gabriele M89 wrote: Well, yes :-) May 10 10 08:52 am Link Must not let thread DIE BUMP .  May 18 10 10:29 pm Link Michael Magers wrote: +1 May 19 10 08:53 am Link SRB Photo wrote: niiice May 19 10 10:55 am Link This might be nothing new to some of you. Clip a B&W adjustment layer onto the LF layer and set to Luminosity mode. Play with the sliders. Mask to taste.  May 20 10 07:26 am Link Photons 2 Pixels Images wrote: i'm not near any quality images to test and have no imagination at this time of the morning. what does this effect do? May 20 10 10:26 am Link Photons 2 Pixels Images wrote: Why stop there? Why not convert LF to a smart object, run shadows / highlights, and put an auto-curves / levels on top of everything? May 20 10 10:40 am Link Mask Photo wrote: Not that it's a quality image, but it's something. May 20 10 04:19 pm Link Photons 2 Pixels Images wrote: Awesome - seriously this is blowing my mind and that is the cherry on the cake! May 21 10 07:49 am Link I'm fascinated by this MATLAB stuff, but it's too expensive http://www.mathworks.com/products/image/demos.html May 21 10 12:17 pm Link NothingIsRealButTheGirl wrote: +1 on both counts. I tried to implement 10 lines of MATLAB code in PixelBender and wanted to pull (what little I have) hair out. May 21 10 01:58 pm Link NothingIsRealButTheGirl wrote: Dunno how they do for feature parity in the image processing realm but these two are (seemingly) good open-source equivalents: May 21 10 02:03 pm Link Photons 2 Pixels Images wrote: oh oh yeah, of course. I've done similar things with multiple targeted channel-mixed layers before the b/w adjustment layer was released. May 23 10 12:36 am Link Just a quick tip to share. If you are like me and don't like to clone/heal onto the original layers but instead onto a blank layer, this could present a problem when cloning/healing with the High Frequency layer. The only way to do this is to select "Current layer" for your sample. This would cause you to have to go back and forth from the HF layer to the blank layer each time you want to resample. Or, if you want to use "Current & below" for your sample, you can hide all layers below the HF layer but then you will only see the grey-ish HF layer which is not always the easiest to judge how the changes you are making will look in the final. Cheer up, there's a way to change this to use "Current & below" for your sample and still see what the overall will look like so you can see what your changes are doing to the final. And it's very simple. 1. Hide all layers below the HF layer. 2. Make a copy of your LF layer and place it above the HF and HF heal/clone layers. 3. Change this copy to blend mode "Overlay". 4. Unhide this copy. You should now see something that looks very close to the original. Close enough that you can determine what your cloning/healing on the HF is doing to the overall image. You can now also keep the HF heal/clone layer active and for sample use "Current & below" since the only layer visible below that is the HF layer. You can also create an action that will hide all the layers below HF and unhide the LF copy and another action to do the opposite. Or, if you're like me, use a script that will do this for you as soon as you select the HF heal/clone layer and change it back when you select another layer. I hope this is useful for some of you. There may be other uses for this, also. I haven't explored the possibilities. Jun 05 10 05:04 am Link I'v red a lot of this great stuff here and wow, it's great. But really sorry, can't make it work on a fabric wrinkles removal, can someone please please show me the way? I really need some help. Ok, seems i'm on it, i'm using the script provided and i'm working on the low freq layer. Dark areas are going out, wow this is grat. What about removing the high freq stuff? Just clone and heal inside the high freq layer? Thanks for answers and sorry for all this questions. Thanks a lot Jun 30 10 05:29 am Link Martin Benes wrote: I hope my PM helped you figure it out. Jun 30 10 04:26 pm Link 900th post Jul 01 10 04:54 am Link Koray wrote: Congratulations, Sean! Jul 01 10 06:59 am Link Pixels 2 Pixels wrote: Yep thanks! Jul 01 10 07:45 am Link  Thanks guys - who knew this would hang on so long? The credit is the forum's, though - I'm just a nerd who blurts what he's thinking a little too often. Without a community that buys into sharing knowledge, this wouldn't have gone anywhere. So props to all of you. Jul 02 10 03:31 am Link wtf happened to robert randall? havn't seen him on forums in ages, and his acct is closed? Jul 02 10 01:44 pm Link syd47421 wrote: PM sent. Jul 02 10 04:15 pm Link SRB Photo wrote: I think one of the reasons it did hang on so long is that there are so many uses and different ways to apply a separation and for me, it's fun trying things out and discovering those different uses. Jul 02 10 04:17 pm Link Simple frequency separation to make photos of textures with light falloff into nicely tile-able textures. http://www.neilblevins.com/cg_education … exture.htm via this thread Jul 04 10 03:01 am Link SRB Photo wrote: Hey there, I'd love to understand this process, I'm a newbie to MM but have been retouching for years and was rather fond of the High Pass method for sharpening. So when I saw this thread I had to see what this commotion was all about Jul 05 10 12:16 am Link haydenwoodphotography wrote: do what first post says and follow links to the other posts and you shall be fine Jul 05 10 01:04 am Link haydenwoodphotography wrote: As Koray points out, please review the "sticky'd" posts referenced in the first message and then let us know if / where you're still stuck. Jul 05 10 07:11 am Link Wow, what a read! You guy's are bloody geniuses. Many thanks to those of you that created these actions. Stunning work! My head is hurting after trying to absorb what was said, however beer is helping LOL Have a good weekend. Jul 09 10 02:06 am Link Koray wrote: FWIW, I think that this article is the basis for how NiK are able to make TC operate so quickly. Not surprisingly, it was written by one of Kai's original gurus. The technique is patented, so direct duplication is out for the time being. I've emailed the author asking whether he's licensed any direction implementations which are available for download / purchase - I'll let you know if / when I hear something. Jul 18 10 07:45 am Link I'm sorry for being an idiot, but how do I determine what blur radius I need? I haven't wrapped my head around that unfortunately. Aug 04 10 05:28 pm Link Peter Claver wrote: Aug 04 10 05:31 pm Link Kantorka wrote: Aug 04 10 06:06 pm Link Pixels 2 Pixels wrote: Here's another video which might be helpful to shedding light on radius: Aug 05 10 08:23 am Link syd47421 wrote: Pixels 2 Pixels wrote: Can you PM me as well? I'd really like to know too. Pixels 2 Pixels wrote: This link doesn't seem to be working for me. Did something happen to it or is my computer giving me a hard time? Aug 08 10 09:44 am Link Ruben Vasquez wrote: syd47421 wrote: Pixels 2 Pixels wrote: Can you PM me as well? I'd really like to know too. You may get a couple errors. It should still load, though. But it is a windows media video. I also made a flash video of it. You can find that one here.. Aug 09 10 07:14 pm Link Pixels 2 Pixels wrote: Thanks! Can you send that pm to me as well? Aug 09 10 09:02 pm Link |
.
.
