If I remember correctly, Nathan said that the rendered image exists in
floating point format before it is converted to the final image
format. Would it be possible to gain access to the floating point
values? I'd like to be able to specify my own formula for converting
the floating point value to an 8bit integer.

Why? I want to alter the linearity of the translation from floating
point to integer. If you look at a plot of photgraphic film's response
to light, it isn't linear. There is a certain threshold that must be
achieved before any image is recorded.  Above that, the response is
somewhat linear, until we get to the brightest exposure levels. In the
highlights, increasing the exposure has less and less of an effect.

In other words, the darker picture elements are expanded in dynamic
range (higher contrast), and the brighter elements are compressed in
dynamic range (lower contrast). One of the effects of this would be
the ability to render light sources and not have them "wash out", or
lose detail,  quite so easily. I think that adjusting the response
curve to approximate film would also make the image look more
realistic in general. 

I'm not sure yet if the human eye responds to light intensity values
in a similar non-linear way, but I think it does. I know that human
hearing is non-linear in its perception of dynamic range, but I can't
say for sure about human vision yet.

If anyone needs more of an explanation of what I'm talking about, I'll
be more than happy to scan some charts for you and explain further.

Thanks,
Glen Berry

Post a reply to this message