Re: [ft-devel] Downscaling of rendered bitmaps?

[Nikolaus Waxweiler]

Alright, I modified ft_smooth_render_generic() like this: I abuse "vmul" 
to inflate the outline times 3 on the y-axis, then average these 1x3 
pixels into a downsampled bitmap:

  for ( y = 0; y < downsampled_bitmap_height; y++ )
  {
    for ( x = 0; x < pitch; x++ )
    {
       downsampled_bitmap.buffer[y * pitch + x] =
         ( ( bitmap->buffer[3 * pitch * y + x] +
             bitmap->buffer[3 * pitch * y + pitch + x] +
             bitmap->buffer[3 * pitch * y + 2 * pitch + x] ) / 3 );
    }
  }

This alone makes no difference; as discussed in 
http://www.beatstamm.com/typography/RTRCh4.htm#Sec11, rounding to sample 
boundaries is also needed. This is trivial on the x-axis, since all 
rounding instructions there can be replaced by Round_None, as my v38 
patch does. But what to do about the y-axis? Rounding to none there 
makes a difference, but stems are no longer vertically aligned to the 
pixel grid obviously, so the hinting improves little on no hinting.

Anybody know how Microsoft does 6x5 supersampling?

I'm currently thinking about rigging the rendering pipeline somehow to 
double glyph size before hinting (10pt -> 20pt) to get around the 
rounding to sample boundaries issue and downsample it after 
rasterization. Hm.