--- Austin Franklin <darkroom@ix.netcom.com> wrote:
> There are not any other dimensions than I stated (positional and
> value) to
> the data you get from a digital camera.
Well, you did get off-topic so I took the right to go off-topic as
well. For instance for an interlaced digital video camera time sure can
be important for interpolation.
> > > Interpolation requires the addition of new data points, like when
> a
> > > scanner
> > > that has an optical resolution of 1200 DPI gives you 2400 DPI.
> >
> > Interpolation does not require new data points, it can produces
> them.
>
> Of course interpolation produces them how is that different than "the
> addition of new data points"? The definition of interpolation
> REQUIRES that
> additional data points be "produced" or, more accurately, "created".
A requirement is something that is necessary in order to do the thing
with the requirement. But creating additional data points is the result
of an interpolation. Nit-picking, as you do. Maybe I should better let
that be as my English is not really the best.
> > Exactly, you insert the blue, green, and red data points where they
> are
> > missing on the 6Mpixel grid.
>
> You are NOT inserting any new data points. The 6M points are already
> there.
> You are only changing their value. Simple as that.
Yes, there are 6M pixels but these are incomplete (at least what
photography is concerned) as each pixel is missing two color channels.
As you mentioned before one important 'dimension' is the position of
each such value. The green, blue, and red values are all positioned on
different positions on the 6M pixel sensor. So for all positions where
there is a green value there is no blue and red value. You have to
interpolate(or use any other method) the values for the two missing
channels to get a true color pixel. That is what you refer to by
'changing values'. You are inserting/interpolating the values in the
two missing color channels.
> Interpolation is an algorithm, and as such, the derivation of color
> information does not have to be done with interpolation. Good, bad,
> or
> indifferent.
Sure, you don't even need to 'change their (pixel) values' as you
suggest. Just use an RGB pattern, look at it from FAR away and it looks
ok (well you might have to adjust the gain for each channel, etc). But
my original question was about resolution. I doubt that this or your
nearest neighbor solution would give any reasonable price/performance
ratio, resolution, etc. Especially considering that the sensor costs
quite a bit more then the minimal additional computation power to use a
simple interpolation that already can increase performance quite a bit.
But I stop it here as it starts escalating.
Robert
__________________________________________________
Do You Yahoo!?
Make international calls for as low as $.04/minute with Yahoo! Messenger
