"Over" alpha compositing doesn't work in any sensible way for depth (z) values,
and will be wrong for even the color channels if the images are not specified
in strict depth order.
At some point later, we may wish to add a "depth compositing" function, in
which the order of the images (on the command line) is unimportant, and the
layer ordering determined pixel-by-pixel based on the z values (in which case
we should fail if any of the input images lack a depth channel).
So for now, please have over() fail if the input images have z_channel >= 0.
-- lg
On Jun 21, 2012, at 5:16 PM, Stefan Stavrev wrote:
> I am going for one function that will classify an image as:
>
> {G, GA, GZ, GAZ, RGB, RGBA, RGBZ, RGBAZ, INVALID}
>
> where G stands for grayscale. This is work in progress, I am just
> brainstorming. But let me convince you right away of one big positive thing.
> We can have more such input validation and image classification schemes. For
> example one for the command line user and one for the API user. We can have
> more or less strict schemes. Imagine if we don't do this and after a year we
> change our mind about what is an RGBA image, you would have to change all the
> code in all the algorithms I will do. But if we do this, you would just
> change the "scheme" I talk about. You feel like the current rules are too
> strict? Fine, just change the scheme, don't change the code all over the
> place.
>
> Larry told me to ignore "channelnames", so for this classification I am
> taking
> into consideration only nchannels, alpha_channel and z_channel.
>
> I provide definitions plus questions for each to provoke ideas.
>
> 1. G: (nchannels == 1 && alpha_channel == -1 && z_channel == -1).
>
> Is an image with 1 channel and that specified as alpha channel, a grayscale
> image?
>
> 2. GA: (nchannels == 2 && alpha_channel >= 0 && z_channel == -1).
>
> What if the alpha channel is the first one, AG instead of GA? Is AG invalid?
>
> 3. GZ: similar reasoning as for GA.
>
> 4. GAZ: (nchannels == 3 && alpha_channel >= 0 && z_channel >= 0 &&
> alpha_channel != z_channel).
>
> What about GZA or AGZ, does order matter?
>
> 5. RGB: (nchannels == 3 && alpha_channel < 0 && z_channel < 0).
>
> This is where having more schemes is good. You could have a not so strict
> scheme and say RGB for an image with 3 channels no matter if alpha or z
> channels
> are specified. Or you could have a stricter scheme and say an RGB image has 3
> channels and alpha and z channels are not specified.
>
> 6,7,8. Similar ideas.
>
> 9. INVALID: everything else.
>
> Let me know what you think and let me emphasize again, I think this is very
> important to solve now.
> It will apply later for all the algorithms I will work on.
>
>
>
>
>
>
>
>
>
>
>
> _______________________________________________
> Oiio-dev mailing list
> [email protected]
> http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
--
Larry Gritz
[email protected]
_______________________________________________
Oiio-dev mailing list
[email protected]
http://lists.openimageio.org/listinfo.cgi/oiio-dev-openimageio.org
