Re: [AGG] Gamma, oh my!

[Jim B. <ji...@ch...>]

On 2010-06-22 22:33, Stephan Assmus wrote:
> Before reading that article in particular, I was subject to about the same
> confusion as Vinnie. I also thought that sRGB is perceptually linear, but
> that appears wrong. As can be seen on the page (with a good enough monitor,
> mind you, but the page explains that, too) a grey of value 186 captures the
> same brightness as a black/white mixed pattern. That pretty much proves it
> for me that sRGB is not perceptually uniform, only that the difference
> between any given adjacent gray values is perceived to be about the same
> across the range.

Hi Stephan,

I'm slightly surprised you didn't know this already - Maxim was 
certainly aware of it by the time he wrote his curiously-titled "Texts 
Rasterization Exposures" article three years ago:

http://www.antigrain.com/research/font_rasterization/index.html#toc0008

"On the right there are two pixels and we can credibly say that they 
emit two times more photons pre second than the pixel on the left. 
However they do not look two times brighter. Four pixels will look about 
two times brighter, not two."

However, Maxim clearly still didn't fully understand the issue because 
in the following section he talks about applying gamma to a screenshot, 
then is surprised that it doesn't improve the appearance of the fonts. 
Well, of course it doesn't, because you can't "fix" gamma-incorrect 
rendering after the fact. The only way to get the right result is to 
render with gamma-correct blending operations from the outset.

> As conclusions for my graphics software, I am pretty much convinced that the
> way to go is to use a linear color space as internal bitmap format, versus
> converting to linear space only during blending, as the first of Jim's
> articles suggest. I am wondering though if that's where is third article is
> heading when it goes about pre-multiplied colors? :-) Pre-multiplied, linear
> colors seem to be important enough for all sorts of operations, so that doing
> the conversion on the fly must lead to too much duplicated code.

You'll have to hang on a little longer. There is quite a bit more to 
part three than the first two parts. Hopefully it is worth the wait :)

- Jim