[fpc-pascal] fpImage resample via interpolation

[Mattias Gaertner]

On Fri, 28 Nov 2008 15:47:35 +0100 (CET)
marcov at stack.nl (Marco van de Voort) wrote:

> In our previous episode, Graeme Geldenhuys said:
> > >
> > > I'm curious: who needs more than 13 different image
> > > interpolations?
> > 
> > Someone that writes image editing/manipulation software like Gimp,
> > Pixel32, Photoshop etc...  ;-)

Where do you see 13 different resample algorithms in these programs?


> > Someone must have though it's important, otherwise there wouldn't
> > exist that many interpolations around.

lol :)

 
> And in general sb without any speed constraints :-)
> 
> 64-bit per pixel. brrr.

The nice thing about fpimage is that 64bit is only the processing size.
You are free to use any lower depth.
And this is the big problem of TFPBaseInterpolation. It creates a 64bit
temporary image. Imagine a 65mb 8bit grayscale image and try to
StretchDraw it. TFPBaseInterpolation first creates a temporary image of
8-times the size - more than 500mb.
This is one of the reasons I wrote another TFPCustomInterpolation
without temporary image. And by precomputing the horizontal
and vertical interpolation factors it is faster too. Especially if
you use non linear interpolations. And it does not have the shrinking
bug that ignores pixels when shrinking more than factor 3.
And it does not have the enlarge bug that the last rows/columns
remain empty.

Now there are three choices:

a) replace TFPBaseInterpolation with my version

b.1) keep TFPBaseInterpolation and add my version
(TFPBase2Interpolation). Add for each fpimage interpolation class a new
one descending from TFPBase2Interpolation.

b.2)  keep TFPBaseInterpolation and add my version
(TFPBase2Interpolation) with a property Interpolation where another
interpolation can be set and make Filter and MaxSupport of
TFPBaseInterpolation public. 


Mattias