How to make anaglyphs in Photoshop

Transform

If the stereo base was too big on a landscape (where distant points are high in the picture and near points are at the bottom), slide level 1 until the most distant points are separated by 1/30 of the width of your picture. (If you aim to crop horizontally, 1/40 of the height of your screen, assuming it has the common 3/4 aspect ratio.)
Now distort the left bottom corner of level 1 in and the right bottom corner of back-ground in with equal distortion, until the closest forground objects are at zero parallax. If distortion is done equally on the two layers, vertical buildings stay stereoscopically vertical - magic!

The distortion function changes the perspective, as if you are looking down from a height on the landscape. This reduces the parallax difference and is a subtle way to overcome the stereoscopic inverse square law of parallax. (A trick invented by the author, but possibly some unknown 3D image processor has thought it up before?)

Set zero parallax

Advanced: The window itself can be moved forward (or even back) so it sits stereoscopically above the screen, which usually involves having a separate window frame image. Once you have a window frame, you can arrange for an object (e.g. horse's head) to emerge in front of the frame by setting the frame inside the anaglyph, with a white background. Then you can have a "wv" which is not a wv, an "in joke" for stereoscopists.

Some anaglyph workers prefer to have the main object at zero parallax to reduce ghosts and are not obsessed with wv. Despite the logic of that, they will never win prizes!

Strict workers will not allow any object to be in front of the window even if it does not cut an edge, but they probably have anal retention?

Check alignment

Crop

A Red/Cyan anaglyph ends up in the three RGB channels on your computer screen, which can be separately processed in Photoshop. The red filter sees the red channel as black and white. The cyan filter shows green and blue channels with colour, but red is mostly grey.

Ghosts: The filters are not perfect and there is leak of green through the red filter and red through the cyan filter. This leak is less obvious if the luminosity of the ghost areas is pretty much the same in red and green. Deghosting strives to get the luminosities balanced. The eye is more sensitive to contrast differences than to colour differences, which is why the "Analux" deghosting method is so powerful, especially if you use the Wattie modification depending on stepped curves (see later).

Colour: Colour is never perfect but can be remarkably good in green and blue. Low intensity red is only seen if the cyan filter leaks red. If you strive for red, the anaglyph must be balanced for a leaking cyan filter, such as Mirachrome.

Retinal rivalry: Cyan colours are more intense where red is dim. A blue sky looks more vivid where the red channel is dark. But difference in luminosity can produce retinal rivalry or "flicker" as first one eye and then the other is suppressed in an attempt to make sense of the confusing information. The trade-off between colour, retinal rivalry and ghosts means you should set up the anaglyph while wearing the goggles your audience will use. On the internet there is no control of that and you should balance for a "standard" or average goggle, like Proview.

Reduce RGB colours of Layer 1 to only red:

Reduce RGB colours of Background layer to Green and Blue

You now have an anaglyph

Now go to Final Adjustments

Red channel preparation

The left image becomes red, but because red often shows only part of the image detail, it is often better to incorporate some green and blue. During anaglyph viewing, the red channel becomes monochrome (lightness only) and because the cyan filter cuts it out, red does not contribute a great deal to the final colour impression. Just how well red is filtered by cyan filters depends on the particular glasses you use. Anachrome purposely does not cut red particularly hard.

Or you can be more sophisticated by reducing green and blue in the left image before desaturating. Take advantage of this step to reduce the contrast of the left image. It is best done in curves by raising the bottom and dropping the top of the green and blue curves, while optionally increasing mid-tone contrast in the red curve.
This ameliorates the problem of bright red areas causing ghosts and dark parts of the anaglyph showing cyan ghosts because the left image ends up without major dark or light areas.

Imre Zsolnai-Nagy favours using the magenta channel from CMYK mode for the left image.

Ctrl mAs you change the RGB curves, keep the middle of each curve in the centre of the graph or you will end up with too dark (or light) in the final red image.

OR

Use the channel mixer and bring some green and blue luminosity into the red channel. Since this is done numerically, you can develop a formula for doing this with some precision.

Right image preparation

Paste left desaturated red into right red channel

Portrait Framing

In the example, the anaglyph is in a forward-projecting frame. I first saw this done by Guglielmo Menegatti and thought it was pretty cool, especially for portraits. His frame method has its own section in the soft-ware he has written, but this simple version is to put a grey frame around the image by increasing the canvas size. Select the frame, feather the selection, paint with grey. Now choose red channel and move the selection a few pixels right. Now choose green channel and move left. A good way to tell if an object is sitting above the screen surface is to run your cursor over it. If the cursor drops below the surface of the object, it is sitting proud of the screen.

Floating windows

A clever way to remove ghosts and have the window frame in front of the picture, by working in Ph0toshop (rather than Guglielmo's software), is described by Shahrokh Dabiri. This is a very powerful technique, which has now also been implemented in StereoPhotoMaker.

Roger Harris has a large variety of window effects.

Gently adjust the red channel in curves
watch very carefully what happens to the ghosts - both red and cyan ghosts can be influenced. You may need lots of control points and end up with a rather wavy curve before the ghosts are optimised. You do not have to remove all ghosts - the Analux step is still to come.

Increase saturation, watching what happens to the ghosts. Often you will find a level where the ghosts are at a minimum. If you have done well with the red channel curves, saturation is often less effective.

Adjust brightness. This is done in Lab mode using the lightness channel and adjusting in curves. Lab mode does not disturb the colour balance. Curves gives you tight control over highlights, shadows and mid-tones, but levels only works on the middle of the scale and the author finds that less satisfactory. Working in Lab mode is probably much the same as the Analux method for removing ghosts, except the Analux inventors use levels. The aim is to reduce contrast in the brightness levels which are causing the worst ghosts, what I call "balancing the ghost lightness." You can do this over the whole picture, or in a widely feathered selection containing the ghosts.

To reduce local contrast in "curves," set a control point above the ghost brightness level and another below it, then bring these to almost the same level. You get a nearly horizontal portion of the curve at the ghost lightness. This is a very potent way to remove ghosts. Once that is done, return the rest of the curve to around 45 degree slope. You end up with a step in the 45 degree curve, right at the ghost lightness level. (It is impossible to have this tight local contrast control in "levels.") The result is some over-all loss of contrast. However, ghost lightness stays "balanced" and contrast can be increased again later, in RGB mode, without bringing the ghost back too much.

Now return to RGB mode. Bad residual ghosts can be locally burned in or blurred with a small brush size and about 30% exposure level.
Red ghost: work on the red channel.
Cyan ghost: work on the green channel.
Cyan ghosts on computer monitors are not usually a problem in blue because the red filter usually cancels blue better than green. However, burning just green does result in an object that looks a bit too blue. The advantage of keeping that blue is the retained detail allows much better 3D impression.

If you still have ghosts, many workers advize changing to CMYK mode. Adjustments in the cyan channel are good for removing red ghosts.

Noisy sky: blur the blue channel, often after selecting the sky. You can also adjust contrast for individual channels in curves, watching through the red/cyan goggles for the effect on ghosts. Image / Adjustments / Channel mixer can be used for advanced correction of ghosts. Be careful, heavy use of the channel mixer can reduce ghosts but also reduce the 3D effect. Dmytro Bezsmertnyy has popularised this method.
(I prefer to channel mix on the image pair before making the anaglyph, as described above. Mixing after the anaglyph is created means you are combining right eye colours with left eye colours, which explains how the 3D effect can be dstroyed.)

Ctrl u

Sharpening

Sharpening should be done with care, because a sharpened image does not save well in jpg, except at high quality settings. Sharpening can also bring back ghosts. "Smart sharpen" is best as unsharp mask introduces halos, which can spoil an anaglyph. However, judicious sharpening makes the 3D jump out much better, since it gives each eye "something to bite on," by improving texture detail. More advanced sharpening involves masking the edges, to avoid edge haloes (which are especially bad in anaglyphs) and only sharpening the low contrast areas, to improve texture.

Colour correction

Resize and save

It is better to resize late (assuming you have plenty of RAM) so that small errors and halos become less obvious than when you do tweaks on the final size image. Bicubic sharper or smoother can be used for resizing (Photoshop CS) and each may have a different effect on ghosts, so try both. You can always undo if something goes wrong.

When saving as a jpg, do not compress too much or ghosts with a noisy pattern will show up. When saving for the web in Image Ready, always choose more than 50 (E.g. 51) to avoid chroma sub-sampling, which is very severe on anaglyphs. Chroma sub-sampling is especially damaging to sharpened images.

Ctrl z

Incredibly useful key sequence to undo mistakes. Multiple mistakes can be corrected in the History palette.

Prepare the right image for Blue Colour:

Lighten the right image to overcome the very dark blue filter.

Filter / Other / High pass.

Final adjustments

Wearing the ColorCode goggles you can fine tune the stereo window by moving the blue channel, using horizontal arrow keys only. Adjust the blue channel brightness to optimum using curves (or levels). Too bright and the colours will shift to blue. Red becomes less saturated as blue becomes brighter. Adjust the red and green channels individually with curves (or levels) to: suit your eyes, optimise colour, optimise contrast (not too much) and reduce any ghosts. (Ochre/Blue anaglyphs can be almost ghost-free and very close to true colour). The CC filter has a low transmission for green and in forest scenes a bit of green boost seems to help, especially with a classic S shape curve on the green channel. Adjust the colour saturation, just as described above for red/cyan, to reduce ghosts and optimise colour.

A standard method to brighten dark anaglyphs using levels is given by Mathew Reeves:

Adjust the midpoint gamma of the overall image after rendering the anaglyph. In Photoshop, you can do this simply by using the Levels adjustment with the following settings:
Input Levels: 0, 1.2, 255
Output Levels: 0, 255
The second Input Level determines the midpoint brightness value of the image, and can be used to increase overall image brightness without supersaturating the image or losing image definition.