Very Computer

Check this out :

http://afterstep.sf.net/afterimage/

This is something that I wrote for AfterStep window manager,
but it has grown to be rather powerfull raster image
manipulation library. The accent has been made on high quality
image processing - instead of 8bit per channel with saturation,
it does 32 bit per channel with subsequent translation into 8bit
  with error diffusion to compensate for bandwidth loss. Image
data could then be compressed using RLE on per channel basis. It
actually performs very well, despite perceived complexity.

Some other things to mention :

Supports about 10 file formats.
Could work with and without X windows.
XML based image manipulation.
High quality 128bit raster engine.
15 image blending methods.
Fast image scaling with interpolation.
HSV adjustment.
Gradients and TrueType antialiased text rendering.
In memory image compression.

Have fun.
Sasha.

Is that 32-bit integer or 32-bit float?

It is both :)
I shift 8bit data leftwards by 8, then do all the calculations
as if it was 32bit integer, then at the end I normalize it into
16 bit, and then carry over half of the lower 8 bit onto the
next pixel, and then shift rightwards by 8 to get back to 8 bit.
So you may consider it to be 24.8 float.

Similar thing is done when transfering image onto X Drawable on
16bpp displays - I carry over half of the lost bits onto the
next pixel. Surprisingly such a simple technique yields very
noticeable improvement in image quality.

i think you mean 24.8 fixed point, and thats not floating point by any
means...

Quote:

> >>Check this out :

> >>http://afterstep.sf.net/afterimage/

> >>This is something that I wrote for AfterStep window manager,
> >>but it has grown to be rather powerfull raster image
> >>manipulation library. The accent has been made on high quality
> >>image processing - instead of 8bit per channel with saturation,
> >>it does 32 bit per channel with subsequent translation into 8bit
> >>  with error diffusion to compensate for bandwidth loss. Image
> >>data could then be compressed using RLE on per channel basis. It
> >>actually performs very well, despite perceived complexity.

> > Is that 32-bit integer or 32-bit float?

> It is both :)
> I shift 8bit data leftwards by 8, then do all the calculations
> as if it was 32bit integer, then at the end I normalize it into
> 16 bit, and then carry over half of the lower 8 bit onto the
> next pixel, and then shift rightwards by 8 to get back to 8 bit.
> So you may consider it to be 24.8 float.

> Similar thing is done when transfering image onto X Drawable on
> 16bpp displays - I carry over half of the lost bits onto the
> next pixel. Surprisingly such a simple technique yields very
> noticeable improvement in image quality.

is there a win32 version of it ?