Re: Color Image Formation
Pierre Weiss comments my suggestion that the presence of colourful
sources scatters the colour gradients:
But I think it doesn't occur in natural scenes (maybe in some discos).
I've plotted the histogram of
abs(<grad(u1)/|grad(u1)|,ortho(grad(u2)/|grad(u2))|>)
(u1 corresponds to the red channel, and u2 corresponds to the green
channel), or lambda-/lambda+ (Di Zenzo's geometry), for some good images
(I mean non noisy).
And effectively, less than 10% of the pixels have a value higher than 0.1.
If you had an example, with natural light, it would be great. I'm still
not convinced this approach is bad.
Quantitatively it is as good as your samples are. No point in arguing.
But in processing of photos there is always something "wrong".
You go skiing, say, JFLA in French Jura.
Your suit is white.
Sun light is white.
Snow is white.
You make a photograph.
Snow is blue. Your suit, because of the importance of indirect lighting
is bluish in shadows, and yellow under the sun. It might be harmful or
not...
This is one of the reasons why Dan Margulis devotes so many place in his
books to the 'search for neutral colours' on images undergoing
correction.
Toutes mes amitiés.
Jerzy Karczmarczuk
.
