[comp.compression] Dithered compression

lance@motcsd.csd.mot.com (lance.norskog) (06/04/91)

I've had an idea kicking around.  I'm sure it has a name.

Take a gray-scale image.  Consider each pixel not as a number,
but as the sum of 4 overlapping 2x2 fields.  There is a special
case for edge pixels.  Each field has an arithmetic value ranging
in 0<N.  To transform a picture into this representation,
set up thousands of simultaneous equations, each in 4 variables.
Not a pretty thought.

Consider the fields to be 4x4 instead of 2x2.  Each pixel
is overlapped by 16 fields.

Drop half the fields.  Fold their values into their neighboring
fields in some reasonable manner.

Now, every pixel is formed by eight fields instead of 16.  The
image, when generated from these fields, should be dithered.

If its the background to an animation, add +-1 or 0 to each
field randomly before summing to generate the pixels.  This
would simulate the quantization noise of real video.  If you have
a separate z-buffer channel, use the distance to generate noise
to enhance the distance effect.

Lance