1. Do a copy merge of the "Mask 1, combine" and paste it into a new raster layer.
2. Create a new raster layer on top of that and start painting white in the face area and black in the background area away from hair strands.
3. Copy-merge/paste to new layer that and did some dodging and burning in between hairs to finish it off. This gives fig 2.
4. Looking at the left side, we see our very grainy and dim extraction of the nearly-invisible hair strands. I just caved in and used the fine hair brush and painted some strokes of somewhat transparent white and then finished up with light smudging to smooth it out a bit. I don't feel too guilty about this part because the hair really is nearly invisible. The only reason we can see anything resembling hair is psychological in nature--we see more than what is actually there. Fig 3 shows how it looks after I brushed in some new hair.
5. So that's it--we've created what I call Mask 1. It's time to apply it to our original image. I like to put the Mask 1 extraction on the worst possible new background. Since the old background was sort of black, I choose the exact opposite--something very bright. I just created a new raster layer (it's called raster 1 in the attached layer depiction) with the new color.
6. Now put Mask 1 into a mask layer. Select the original image and create a mask (show all) on it. Then go back to the Mask 1 group (see layer palette attachment), select all, copy, and then go back to the Mask you just created and do "paste into selection" which transfers Mask 1 into the actual mask layer. (This, BTW, is how you can paste the luminance from any selection into a mask or adjustment layer.)
This is what I did in fig 4.
It looks a bit...well...off. Part of the problem is the original is VERY dark and it tends to look unnatural in front of such a bright background. The other part is the fringe effect folks complain about--but in this case the fringe is dark--ie., the old background color still remains mixed into the hair color near the edges of the mask. In this case, the old background color is nearly black. In part 5, we'll get rid of it...mostly. As I originally said, this image is at the limits of what is possible.
Fig 5 (layer palette) shows the new layers we created in this part.
Great Work on Hair Extraction
Bart: I say it again, this is great work. I have it all captured in an Excel spreadsheet and am working to practice with it. I have numerous real-world images of my grandson which demand its use (including one just like this example, if you can believe it, with dark background right behind the hair area---if nothing else, by learning and applying this technique, you learn to more carefully compose your photos in order to avoid having to apply this technique ).
I am not sure how many PSP users read messages here, but you've given them a gem of a tool/approach. PS users have probably been passing right by, although it seems like most of the principles you describe are applicable across platforms, even if the specific steps and layers, etc are not.
I appreciate your sharing.
Part 5--final episode
Time to de-fringe and re-fringe.
1. Defringing. For most backgrounds, I would remove the fringing at this point by burning or multiplying the fringe areas by the inverse of the background. The inverse of the background in this case is nearly white--ie., when the background is black, you don't have to remove the fringe--it's already gone because black is 0,0,0. That is the one and only thing about this extraction that's "easy" (although the step we saved is a trivial one.)
Fig 1 shows the latest way I've been doing this. The de-fringing mask is simply the inverse of Mask 1--saves a lot of trouble. The de-fringing group is wedged in between the original and Mask 1. I usually use burn blend mode for defringing. However, it's sometimes worthwhile to try multiply as well. Kind of depends on how aggressive you were with masking. For this technique, you want to be slightly non-aggressive in your masking--ie., preserve details at the expense of having more fringing. I think you'll get the hang of it with practice (I'm slowly getting the hang of it myself.) It's nice having this flexibility--you can choose between multiply and burn, and you can adjust the opacity slider as well (although I usually have this one at 100%).
Note that "synth old background" layer--that's the original image that's been cloned to recreate the background in the fringe area. This is how the background generally has to be created in situations where the background has different colors. If it is a plain color, then you can just flood fill with that color. Of course if it's a plain color and there's decent contrast, you shouldn't be using this method--you should be using the background eraser.
As I mentioned, this defringing has negligable effect because I'm essentially burning the fringe with white--which of course is a no-op. I only include it for completeness.
2. RE-fringing. Now we add back in the correct fringing for this background (pale yellow). Fig 2 show the re-fringing group--good news again. Mask 2 is repeated here--it's simply the inverse of Mask 1. Mask 2 is applied to another copy of the new background and the blend mode of the resultant group is set to "screen". Screen (and dodge) tends to mix a new color in (whereas burn and multiply tend to de-mix colors). I haven't found dodge to be appropriate yet, but you never know. I usually slide the opacity slider around until the refringing looks about right. The nice thing about this method is once you've set these sliders for the most difficult background, it's about the correct answer for any other background you will try.
3. Fig 3 shows how the extraction looks now. The blotchy messiness is mostly gone now, but the fringing still looks sort of dark. That's because the foreground object is dark. You could screen some more background onto it, but the real solution is to boost the effective EV of the foreground object. Fig 2 shows where to add the boosting curve. By putting it here, it boosts everything associated with the extracted object and has no impact on the new background. Consequently you can vary the amount of boost quite a lot and it still looks natural. Fig 4 shows how it looks after EV boosting.
4. If you want further cleanness (ie., web-publishing), you can turn up the opacity of the re-fringe group and/or use a low opacity, very soft white brush and paint around the edges of the hair on Mask 2 in the re-fringing group--this lets more of the new background show through the hair. What's nice here is if you screw up, you can just go grab another copy of Mask 1 and invert it to start over again.
If you will eventually be using a "realistic" background anyway, that sort of cleanup might not be needed. On top of that most images with any decent amount of contrast will come out looking almost perfect at this point. But even this image looks pretty good on real backgrounds (Fig 5 shows some examples).
Link to fig 5:
PS7 Layers and Mask Help - Hair Selection
I am resurrecting this thread because I need some basic help on PS7 masks and layers... (no question is too dumb here, only the unasked ones---right?).
I have been trying to increase my PS7 skills in terms of extractions and selections and have been trying to eliminate "fringing". Bart Hickman, who has contributed a lot to my learning in PSP10, worked up an example of fringe removal, and he put it here (http://tinyurl.com/brwt8). He did this in terms of PSP layer stack/palette, and I pretty much understand it that way. In trying to accomplish a similar (same?) result using PS7 and its layer pallete, I am frustrated and cannot.
The PSP layer system works pretty intutively and masked layers can be used with ones above to create composite or blended results. The key is that the mask is on its own layer.
What is throwing me off is that the PS7 layer system puts the mask ON THE SAME LAYER, and I do not know how to set it up to work equivalently as the PSP layer/mask combination.
Here is a word example to go with the figures below (or in Barts tutorial shown above).
I mask out the hair and put it in front of a black background; the hair "whispy" areas have a blue fringe.
I apply a channel mixer mask and adjust so that all that is showing is the fringed hair. I want this to be a mask.
I sample the color of the fringed color and fill a layer with this color and invert it.
I want to multiply blend this inverted color LAYER, masked by the channel mixed fringe area with the original fringed image LAYER, which will then eliminate the blue fringe.
I cannot do it. I have studied (and tried) clipping groups and layer sets (remember, PS7 and not PSCS), and can not find the combination of mask, layer, group, set, or ??? to accomplish the removal of the blue fringe from the hair.
This is complicated to explain and I don't blame anyone for passing by, but if anyone can set me straight, I would be grateful.
>>>I sample the color of the fringed color and fill a layer with this color and invert it.
>>>>I want to multiply blend this inverted color LAYER, masked by the channel mixed fringe area with the original fringed image LAYER, which will then eliminate the blue fringe.
Try loading the mask of the fringe area (Option click on the Layer's mask). With that selection still active, select the inverted blue filled layer and click on the "apply mask" button at the bottom of the layer's palette.
This should load the same mask selection into the blue layer.
I'm glad I came in late on this one. Instead of putting in a lot of time trying to get it right, now I can just try the great techniques posted here. Thanks for doing all the hard work!
The mask is the biggest difference between PSP and PS. Once you get that figured out, the rest is cake.
The first attachment shows the generic translation from PSP mask to PS mask. To apply a mask to the active layer, PSP creates a mask layer on above the active layer, then puts the two into a group. So while I agree with you the PSP method is more intuitive, it's much less compact.
Figure 4 from that tutorial will then simply look like the second attachment with the "Remove old fringe" set to multiply.
I did this already and discovered it doesn't quite work as well in PS as it does in PSP. There's something different about the way the two programs are doing multiply with masks.
No matter. PS has the linear light blend mode--change the blend mode of the "Remove old fringe" to linear light. Then adjust the opacity up or down to make it look good--50% seems about right. This is kind of nice because it means you have +/-50% of fudge range to work with.
I finished up the method and the layer palette is shown in the 3rd attachment (corresponds to figure 7 in the tutorial.) Still use the screen blend mode for that.
One more important thing: Layer groups default to "pass through" blend mode in Photoshop. Be sure to change it to "normal".
Hey Bart: I was hoping you would respond.
Sorry for my denseness, but it seems like a key step is missing.
In figure 4 from your tutorial, you create Mask2 by using a Channel mixer to isolate the blue fringies. Where is that in what you posted above? Your attachment of the PS set shows the Mask on the inverted blue layer as if it just popped out of nowhere. It has to be created from a channel mix adj layer, yes? Its even shown in your Figure 3. I even have it on a PS adj layer but ..... so, I just created a channel mask by using the magic wand on it, inverting it, and saving it as a channel (called Mask2). Now I'm stuck on getting this combined with a filled layer of the browish inverted blue color (your "remove old fringe" layer).
What am I missing??? (or is a step or two missing from your comments?)
Thanks for helping.
Edit: I reread Swampy's note and tried that and got something directionally correct it seems, but the quality of the Mask2 I created with the magic wand, and saved as a channel which I load back, really sucks....so the resulting image is a combination of some blue some brown and some hair....so, it seems like I still need to understand what you are showing me.
Last edited by Ken45140; 04-03-2006 at 09:40 PM.
Mask 2 itself is created up in fig 2 "The making of Mask 2". In PSP, I would have simply raster-copied the "Mask 2 construction" group and pasted it into the "mask 2" mask layer. In PS, I alt-select the mask, then paste which puts the clipboard into the mask.
To copy a raster image into a mask, do the following (in this case, you actually want to do a copy-merge of a group and paste that into a mask)
1. Alt-click the eyeball for the Mask 2 construction group (you should now see Mask 2)
2. Ctrl-A, then Ctrl-shift-c. (you've now got Mask 2 in the clipboard)
3. Alt-click the eyeball again to return layer visibilities to previous states.
4. Alt-click the mask thumbnail that will become Mask 2.
5. Ctrl-V (puts the clipboard into the Mask).
Here's a cheat sheet for all the basic mask operations in Photoshop (at least the one's I use.)
I think one of your sticking points is you're trying to combine the creation of Mask 2 into the defringing group and that's not what I was doing (I don't think it's even possible). The group where I create Mask 2 is completely separate from the group that does that actual defringe/refringe operation.
I've attached a layer palette showing *almost* the entire layer stackup. I say "almost" because the "fringy object on black background" also requires a couple of layers to create. I didn't include this for PSP because the layer stackup was way too big. Note how the Mask 1 and Mask 2 construction groups are down at the bottom underneath the new background and thus not contributing directly to the final image. They are simply a scratch-pad area used to calculate the masks. Once I'm done with that, they aren't needed anymore.
Two final notes:
-I don't think you want to use fill layers here. Instead use a raster layer. Just sample the original background color, then fill the raster layer, then invert it. In many cases, you won't have a uniform background color anyway.
-The mask computation is the hard part and I usually iterate two or three times for both Mask 1 and Mask 2.
Last edited by bart_hickman; 04-04-2006 at 01:09 AM.
Probably, the KEY key (main key) was the cheat sheet. I own at least four PS books, and have on my desk two more CS books from the library, and while I "confess" to not having read any of them cover to cover, I have done a lot of reading. I cannot recall a single mention of Alt-clicking on the eyeball to "see" that layer, nor any of the Ctrl-Clicks or Alt-Clicks on the various mask and image icons. I will go looking for such mentions now and will probably find them many places now.
In any case, I have my version of the layer stack and the result--see attachment. Now some more practice.
I guess the PSP approach, while less compact (more layers) has a more intuitive "stack buildup" process. If you are making the original mask from color channel breakout and associated channel contrast setting, the PSP is tons more awkward (due acknowledgment to your color channel separation script). But now that I see the PS approach, and practice with it, it will become the more "handy".
Again, many thanks.
PS: I hope all you lurkers are taking notes as this seems like a super way of isolating complex subjects from the background.
|Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)|
|Thread||Thread Starter||Forum||Replies||Last Post|
|Selection Tool||T Paul||Scratch Pad||2||04-15-2009 10:46 AM|
|Help removing complex hair from crazy background||renato17||Image Help||16||11-24-2006 09:30 PM|
|Retouching Hair Strands - Does CS2 have a photo "TUBE" mode?||ray12||Image Help||3||09-06-2006 08:39 PM|
|How to make a perfect selection of hair ?||loornard||Image Help||14||03-02-2005 02:30 PM|
|13 Photoshop Selection Tips||DJ Dubovsky||Photo Compositing||5||10-20-2004 07:49 AM|