Another one from Yosemite...

[compilerbitch]

Upwardly Immobile, Yosemite Valley, 2007



This photo has a slight story to it. doseybat and I were walking back from Lower Yosemite Falls, when she spotted this little tree living out a tenous existence on the sheer cliff face. The light was failing at that time, and to make matters worse the part of the valley we were in was in shadow anyway. It was quite difficult to focus, and I didn't get it quite bang on. I used a 150mm lens, probably stopped down a little. I can't remember the exposure I used, but it was probably quite long, at least half a second.

If I had a second chance to shoot this, I'd have (obviously) chosen a time with better light, but the bigger problem was that the tonal values of the leaves of the tree were initially identical to those of the rock. In combination with the lack of sharpness, I had originally consigned the image to the bin, but I used a technique on it that I've been experimenting with a bit recently, and not only did I manage to rescue the sharpness down to the pixel level, it also became possible to make the image really 'pop'.

The first thing I had to do was separate the tones so that you can actually see the tree -- this was done the obvious way, with a levels adjustment layer in Photoshop and a lot of painstaking hand mask painting. I had to use a fairly aggressive curve to render the rock with the tones I wanted, but this had the unfortunate effect of making it look even less sharp than it originally did -- if you ever have part of a curve nearly horizontal, it's (conventionally) really difficult to get a look that is at all acceptable.

So, I had an image with the right tonality, but it looked like a bag of unsharp crapulence. Applying sharpening conventionally really didn't work -- all it managed to do was enhance noise that I didn't want enhancing whilst failing to do anything useful with the compressed midtones in the rock. This is when I decided to use a fancy trick. In essence, it goes like this:

Step 1: Make a copy of the original, unmodified image layer. Drag it right to the top of the layer stack, above any adjustment layers (this is important).

Step 2: Run the high pass filter on it (you can find it in the menus as Filter | Others | High Pass). This will leave you with a mid grey image that shows the finer detail in the original image.

Step 3: Set the layer to Overlay mode (we'll change that later, but it lets you see what you're actually doing). At this point, the original image reappears with enhanced detail, though at this stage it probably looks a bit like it's been sharpened with the size slider too far up. Don't worry about that just yet.

Step 4: Do an unsharp mask on the high pass layer. You can probably crank the amount pretty high, but you'll want to keep the pixel radius down pretty small -- I used 0.3 to best effect.

At this point, you have your original image with all of its levels tweaks, but with greatly enhanced sharpness, including in areas that have been flattened by too-horizontal curves. You can now play around with the layer mode -- I actually used Vivid Light for this image, but different options gave different feels. You can, of course, tweak the amount of adjustment very easily after-the-fact by tweaking the layer's Opacity setting, or even by creating a layer mask.

OK, here's the theory part for why this works. The basic idea is that you're splitting the image into two separate channels, but by spatial frequency rather than by colour, level or any other more familiar attribute. The low-frequency image goes all the way down to DC in signal processing terms, so it's this one that you're most likely to want to tweak if you want to alter tonality, but because the high frequency information -- the detail -- is in a separate channel, you can get away with very substantially bluer murder. This gives me an idea for a 'frequency separation' plugin that uses 2D FFTs to generate 2 or more channels, separated by frequency, that are guaranteed to sum exactly to the original image. However, by tweaking each channel separately with layers curves, it should be possible to have extremely fine control over contrast and sharpness independently of each other. This approach would be a significant generalisation of HDR tone mapping, amongst other things. Hmmm... if and when I get some time... hmm...

Overall, I'm pretty impressed with this technique -- it's the first time I've used it on a for-real image, and as it gave so much better results than the standard approach I think it'll end up in my often-used bag of tricks.

Comments

[lsur.livejournal.com]

Strange coincidence - keirf has also just posted a picture of a tree on a rock. Is it an outbreak or alien invasion, in manner of Triffids???

[dd-b.livejournal.com]

I started trying to read it as involving rocks mostly under water. Then I added the vertical dimension (so maybe the water was coming down a cliff or something). Then I finally got it right :-). For me, in this image, the sequence of interpretations was a good thing, though if you had asked me if an image that initially confused me had a problem I'd probably have said yes off the top of my head.

And the title you've given it is lovely.

Interesting ideas on image processing, too. I'm definitely under-using layers (and I use layers a lot already) and especially blending modes (which are not well explained anywhere; sounds like most people have at best vague ideas what they do and just experiment to see what works, especially for the more exotic ones).

[blue-mai.livejournal.com]

nice one, that sounds rather useful. i'm kind of the opposite in photoshop - having only the most basic wet darkroom skills i use photoshop for what it is, however i don't really understand a lot of how it's doing what it's doing (a combination i think of lack of darkroom and colour knowledge) so it's all a bit hit and miss. everything Adobe i've ever done has been picked up on the go/ intuitively. non-destructive editing with layers is brilliant though.