[Tig] color vision

[Richard Kirk]

Hi.

A bit more on black & white .vs. color...

We can see color differences even when the two objects we are comparing 
are separated by a large angle. We are unable to  do the same with 
luminance information. If we plot our sensitivity to frequencies of 
chroma and luminance contrast, then the luminance  sensitivity peaks at 
about 10 cycles ber degree, and falls off slowly for higher angles.

I used to work with printed images, which typically had a contrast range 
of 100:1. It was necessary to flatten out the contrast range if you were 
to represent a typical transparency on print. The simplest non-spatial 
way of doing this is to compress the luminance in some approximately 
perceptual space. However, this often made the images lack edge 
contrast, so, I attempted to filter the images, removing the long-period 
luminance contrasts where possible to preserve the short-range 
contrasts. This is very similar to the high contrast composite images 
people produce today. Given the right subject, you get an image with 
something with what looks like an unnaturally high tonal range on a 
color reflection picture. This, in many ways, is what I had expected my 
original approach of enlarging the luminance contrast to look like. 
However, on unsuitable subjects, like black and white text against a 
grey background, this can look absurd.

We prefer pictures with a boosted contrast, but if we simply boost the 
luminance contrast on a color image, then we are not getting the right 
color and luminance contrasts at high and low spatial frequencies. If, 
however, we drop the boost on the luminance low frequencies, then things 
seem to work again. This suggests that there is something in the low 
spatial frequencies of the luminance component of color contrast 
detection that limits how much tone boost we can accept.

As for Rob's suggestion of reducing saturation - there is an easy way of 
doing this. We can reduce the luminance. Not all the way to scotopic 
vision (though that is a solution of a different sort) but so the Hunt 
effect limits our perception of colorfulness as the light levels go down 
and our color-difference signals start to get swamped by noise. Some of 
you may not have met 'colorfulness' - it isn't an easy concept, so 
here's an example. I used to make 8mm stop-motion animations as a child. 
I used to project these on a silver screen in the sitting-room when I 
could get it. When this was in use, I used a pad of paper on the floor 
of my bedroom as the screen. I noticed that the image was apparently 
much more colorful and somehow 'crisp' looking than on the large screen. 
Yet, as you move the projector towards the screen and the whites get 
brighter, our senses tell us that all the individual colors are staying 
the same. So, we need separate numbers for the saturation with respect 
to white (relative colorfulness) and the amount of color we are seeing 
(absolute colorfulness).

You can do experiments to estimate how colorfulness varies with light 
level by trying to match saturation in your two eyes when presented with 
a target with different light levels - you will want a lot more 
saturating in the dimmer image to get a match. Okay - this is not really 
a realistic viewing model, but colorfulness is a hard thing to measure.

PS:
This is also a good argument for not making digital projectors as 
saturated as you can. If you filter the light really hard to get 
saturated primaries, you will lose brightness, and your sensation of 
colorfulness will actually go down. More light can be as good as more 
color, particularly down at 50 nits and below.

Cheers.
Richard Kirk

-- 
FilmLight Ltd.  	Tel: +44 (0)20 7292 0400 or 0409 224 (direct)
Artists House,		Fax: +44 (0)20 7292 0401
14-15 Manette Street    
London W1D 4AP