Topic: Edges (Page 1 of 1)

Just... Edges!

Yeah, looking like I'm finally going that way again. Some things have been on my mind for a long time. Going on about a year for this. Maybe a bit longer.

I really detest the bulk of the edge mask tutorials out there. As a matter of fact, I have yet to see one that I even come close to liking.

One of my beefs is that there is never any preping the photograph. A few hit on this aspect, but most don't. Sure, there is plenty of post, but never any pre.

But my biggest beef is the desaturing. OMG this makes a big difference. When desaturing anything, the method makes a difference. RGB is 3d - take advantage damnit.

----

Desaturating

Several different methods to try after running Find Edges:

-Finding Minimum
-Finding Maximum
-Extracting Luminosity
- Channel Mixer
- Image > Adjustments > Desaturate

(Min Max Shop for those Win/PC folks. Man, this bauble plug comes in handy. Or you can easily record yourself an action.)

----

Channels

In the places were an edge mask is used, greyscale is required. I have yet to see someone use an edge mask in full RGB.

What we usually see is RGB > Find Edges > greyscale.

But it doesn't have to be like that.

Try RGB > greyscale > Find Edges.

Now, the greyscale step can be any channel you want - it does not have to be RGB or any of them taken individually. One of the better all-purpose methods is to extract Lum and run Find Edges on that.

Or even try this:

RGB > Raw Hue > Find Edges > greyscale

For even more savviness:

RGB > Colour (hue + sat) > Find Edges > greyscale

And don't forget that you can use Layer ChOps or regular ChOps to compliment the different methods. Find Edges on Lum, Find Edges on Sat, then Multiply them together.

I suppose some of that would fall under pre tricks.

----

Pre Tricks

Rather than wait to try to tease out the data after Find Edges, try to tease it out before Find Edges.

Here is a trick I'm fond of when dealing with low contrast:

- Channels > R + Image > Adjustments > Equalize
- Channels > G + Image > Adjustments > Equalize
- Channels > B + Image > Adjustments > Equalize
- RGB + Filter > Stylize Find Edges

Or you can try Equalize on full RGB, but I prefer on the seperate channels because of how Equalize works. (Equalize on seperate RGB channels can lead to some interesting colour balance tricks.)

Or you can use other adjustment like Auto Levels or whatever. They all work differently and I highly recommend getting to know them.

----

Seems like I am always harping on contrast.

To set in opposition in order to show or emphasize differences.

That's what Find Edges is all about. Do some reading on Sobel and you'll see what I mean.

Where do you want the edges? What differences already exist in those places? How do you tease it out or capitalize on existing differences?

Good old Biker Chic is back like an old friend popping in to say hello.
Great stuff warjournal, I can't play with edges yet as I don't have PS on this laptop but I will do as soon as I get back to my main machine.


Those who look for monsters should look to it that
they do not become monsters. For when you gaze
long into the abyss, the abyss also gazes into you.

Tao, get to playing because here come more.

----

Find Edges and Noise

Find Edges is pretty noisey. As Public Enemy said many years ago, "Bring the noise!" A few years later, Anthrax gave them a hand. Find Edges will do that in spades. Ugh.

Before

You can blur before the fact to get rid of very high noise. Blur, Blur More, or Gauss with a low radius. Personally, I prefer Blur More. If you opt for Gauss, a slightly higher (relative) radius can cause some 'rounding' of the 2d data. This rounding goes back to Kai's well known Gauss + Levels combo.

Another option is Median. Median will largely respect edges, but can cause blotchiness or banding. I use Median when dealing with raw hues because no new values are introduced within the radius. If you extract raw hues and Gauss, you will get intermediate values all over the place (Snowy Mask on crack) and this can be bad. Median avoids that all together.

After

Back to desaturating for a bit.

For finding Min/Max, Apply Image with Darken/Lighten is used. That is a very specific calculation. Who says you have to use Darken/Lighten? Try Screen/Multiply for a little extra oomph. For even more oomph, try Add/Subtract.

For Apply Image and Add/Subtract as we are using them for this, use Subract and check the Invert option.

R > Apply Image: G Subtract Invert
R > Apply Image: B Subtract Invert

After all that, whichever way you go, there is a good chance you will be left with artifacts and some noise. Even with pre-blurring, you will probably end up with noise.

Lots of ways to deal with it, but I want to show you a bit of a power trick. Seriously, this is a power function.

Once you have your noisey edges, give is some Gauss, like around 4 or whatever to start. Then Edit > Fade Gauss using Mode: Colour Dodge and bring Opacity down to taste.

- Filter > Blur > Gaussian Blur using 4-ish
- Edit > Fade Gaussian Blur
-- Mode: Colour Dodge
-- Opacity to taste

That should go a long way to getting ride of post noise. Will even adjust the choke a bit as a bonus. Yeah, it's a bonus. It's not a bug!

Another thing I'm big on is fixing little things and letting them add up to a bigger and better whole. Fix a little bit of the noise before Find Edges, fix a little bit of the noise after Find Edges - OMG uber squeaky clean.

Notice that I didn't mention using Levels/Curves/Contrast et al to clean the post noise? Hehehe.

Now that most of the brain ticklers are out, I'm thinking I'll give LaPlacian a quick whirl. This should be pretty quick-n-sleazy.

Until then
play.fiddle.learn

LaPlace

Find Edges is Sobel with a few small steps thrown in. In particular, Pythagoras and invert. No big deal. However, Find Edges has a radius of 1. That's fine, but sometimes you might want a little extra width in certain places in your edges. This is where LaPlace and High Pass come in. Difference at a distance. Frequency. Yippie.

I've heard folks talk about it before, even myself in that one blur thread by Klaatu. But I don't think anyone has really given explicit instructions. If someone has, I missed it. Fairly simple to figure out with a few hints, but I'm gonna give it up in case you've never heard or fiddled with it.

- Random photograph
- copy to new layer
- Filter > Other > High Pass to taste
- Edit > Fill
--- Using: 50% Grey
--- Mode: Difference

That's about it. Now all you have to do is use your prefered method to desaturate and whatever. Use some of the ChOps talked about above. Invert it and junk.

Two things to point out about High Pass.

The first is that things are done from 128. The difference from there should never yield any values greater than 128. After Fill > Difference, check the Histogram and you'll see what I mean. Easy enough to tweak with Levels or Curves, but getting too extreme will probably bring out compression artifacts.

The other thing is what happens at the center of the difference. You'll get like a locus of nothing where the difference crosses the event horizon. Maybe someone with a superior venacular and classical mathematitional edumacation can explains much betterer.

That event horizon or crossing over is why I recommend complimenting High Pass with Find Edges. It's usually a good idea to fill in the Valley of Zero Point. Or not. I don't know for sure for you when you are doing your thing.

A lot of the tips that I talked about with Find Edges also apply here.

With some practice, you should be able to see what you need and go straight for the jugular. Or maybe come up with a general all-purpose technique that quenches your thirst most of the time. Or maybe pick up a little nugget and apply to something entirely different from edging.

Or maybe you already have an all-purpose edging technique and don't wish to learn anything from this. Egads!

Hahaha.
Too funny.

I'm not sure if I'm going to go any further from here. I might drop a few more notes as they strike me. Maybe answer some questions if they come. Refine a few things. Maybe show a few case studies. Just don't know right now.

Until then, be the guru.

play.fiddle.learn