As you probably know, anaglyphs work by adding together two images which are colored in complementary colors (like red and cyan, respectively), one for the left eye and one for the right eye. In order to separate the two cannels again, special colored filters are put directly in front of the eyes (usually the red one for the left eye and the cyan one for the right eye).

Obviously, this works perfect if both channels are projected using separate (non-overlapping) spectra (i.e. light wavelength ranges) and if both filters completely block one of the channels (i.e. the cyan glass should block the red channel and the red glass the cyan channel).

However, as we will see below, in real live, conditions are far from perfect leading to distortions.

There are two main types of perceptual distortions:

"Shadows"

These are sort of "ghost images" viewed by one eye although they are meant to be viewed only by the other eye. The two main reasons are:

• Imperfections of the filters: No filter will block all light at a certain wavelength range; something will always pass through.
• Spectral distribution of the projection device: The red/green/blue colors on the screen are each distributed over a certain wavelength range and these typically overlap slightly.

Brightness imbalance

The left and the right eye see images which have a different brightness. The main reasons are:

• The filter transmission is less than 100% at those wavelengths which are meant to be transmitted. For example, the red glasses should transfer all red light but typically they absorb 10 to 20% of it.
• The projection device outputs the two channels at different brightness. On screens, red is usually less bright than cyan which is made up of both green and blue (pixels).

Fighting distortions: There is not much one can do against image distortions. The best thing is to get glasses which work well for the intended projection device.
Furthermore:

• Brightness imbalance can be compensated by giving the brigher channel less intensity on the projection device (e.g. if cyan appears brigher, one can scale the green and blue channels down from 0..255 to 0..200).
  Brightness imbalance is an inherent problem for colored anaglyphs for obvious reason.
• "Shadows" cannot be suppressed in general. The only thing you can do is fading down the green channel thereby reducing the amount of "cyan" being transmitted through the red filter and hence reducing the ghost images on the "red eye". Note however, that this also has an effect on brightness imbalance.

The quality of a filter can be described by looking at how much light can pass through the filter at all the different wave lengths of interest.

The human eye can see light from 400nm (blue) to above 750nm (red); take the sketch on the right for a rough and inaccurate orientation (figures in nm).

When talking about filters, there are two main properties of interest are transmission (T) and optical density (OD). They both describe essentially the same and the relation is:
T = 10^-OD
OD = -log₁₀(T)

The reason for having both of them is that a transmission graph, while showing nicely the amount of light transmitted at those wavelengths where the filter is (nearly) transparent, will not give away much detail at those parts of the spectrum where the light is absorbed. To put it in other words: In a linear transmission graph, you can very well distinguish 95% transmission from 80% but not 1% suppression from 0.5% or 0.1%. So, the optical density is a logarighmic suppression figure, OD 0 means full transmission, OD 1 means 10% transmission, OD 2 and 3 mean 1% and 0.1%, respectively.

Note: Since human perception has approximately a logarighmic characteristic, percepted brightness scales more like optical density rather than linear transmission percentage.

The glasses were measured using a Varian Cary 3 (UV to visible) Spectrophotometer. The device was allowed to heat up for some time and baseline correction scans for both 0% and 100% transmission were performed prior to measuring the glasses.

The spectrometer can measure up to OD 5, this is why optical density graphs look jerky above 5. However, since OD 5 translates into 0.001% transmission, anything above OD 5 isn't interesting anyway (think of it as "no light at all").

Note that the best glasses won't give you a good image unless the projection device is able to deliver a good image! Screens, for example, normally have overlapping spectra for different color channels (by design) leading to ghost shadows as pointed out above.

Optical density spectrum	Transmission spectrum	Comments
		I got these goggles from Berezin; they offer a great variety of glasses at reasonable cost. The cyan filter has poor suppression of the red channel but the red filter has excellent suppression of the cyan channel. The low transmission of the cyan filter around 500nm helps fighting standard brightness imbalance. The red filter has a nice steep transition around 600nm but the transmission of less than 80% could be better.
		These are plastic-frame glasses from Perspektrum (article 1055b). Good cyan transmission around 500nm (brightness imbalance must be compensated on projector) but poor red suppression. The red filters have reasonable cyan suppression but the red transmission below 700nm is really bad increasing brightness imbalance. Also note OD 5+ below 400nm; these are good sun glasses!
		Simple and cheap paper-frame glasses from Perspektrum. Reasonable red suppression in the cyan filter (OD 2+) and acceptable cyan suppression on the red filter. Also note the nice steep red filter transition around 600nm and a good cyan transmission (less brightness imbalance than the above glasses because the red channel transmits better).