There are a number of technologies for large scale stereoscopic projection, each with their own advantages and disadvantages. Except for the recent Infitec systems (which have other issues), they all introduce some level of what is normally referred to as ghosting. That is, some of the image that is destined for the left eye leaks over to the right eye and visa-versa. In CRT and active glasses systems this is caused by such things as finite phosphor rise and decay times and limitations in the LCD shutter glasses. In polaroid based systems it is caused by both non perfect polarising materials (two linear polarisers at 90 degrees to each other do not completely block the light) and limitations in the preservation of polarised light of either direct or rear projection screen. It is this ghosting that is a main source of eye strain and it places limits on the degree of parallax a viewer can sustain.
The light arriving at the left eye say is then made up of two components, most of it is the light intended for the left eye but there is also some fraction (hopefully small) of the light intended for the right eye. The same applies for the right eye, the light reaching it is the sum of the intended right eye image plus some leakage from the left eye image. So a novel way of reducing this ghosting and the annoying visual effects that result is to subtract the ghost image contribution from the source image such that the ghosting reconstructs the complete (intended) image.
In the following consider a simplified situation where we ignore the loss of the transmitted light. The ghosting factor will be denoted "A", typically under 0.1. Note that typically this will be a slightly different factor for each r,g,b component since there is normally a wavelength dependence in the imperfections of the projection system.
If the input images are modified, that is, some fraction of each image is subtracted from the other image then the situation is shown below. It can be seen that the resulting image at the eye is now the intended image plus an error term that is multiplied by the square of the ghosting factor (much smaller).
There are many simplifying assumptions made above but the basic idea is valid and using this technique it is straightforward to reduce the effects of ghosting.
One particularly important consideration is the gamma of the projection system, that determines the amount by which the pixel values of the input image should be adjusted .... it is the light intensity that needs to be compensated for not the pixel values.
It is also obvious that one cannot subtract from black, the solution to this is to live with the ghosting in the black areas of the image, or one can adjust the pixel values of the image so that there are no pixels below the correction value. This can be done by scaling all the pixels or raising just the offending ones.