[FFmpeg-devel] Question: indeo5 scalability DWT

[Roberto Ragusa]

Maxim wrote:
> Hello guys,
> 
> 
> {h0} = {-1, 2, 6, 2, -1}1/8
> 
> But the high-pass filter coefficients don't match:
> 
> {h1} = {1, -2, 1}1/4
> 
> My question: could anybody skilled in wavelets (Michael?) clarify me
> this situation? Is that a modification of the LeGall wavelet or a
> completely different 5/3 one?

What did you expect for h1? Maybe

{h1}={-1 2 -1}1/2

It is completely equivalent to the one you wrote.
There is just a sign inversion in the high pass filter
(which is basically arbitrary) and a scale factor
of 2 (which again is arbitrary and can be compensated
on the other side of the analysis/synthesis).

Consider this:
I can give you two numbers a and b.
Or I can give you another representation: a+b and a-b.
You will be able to solve for a and b.
What if I give you instead a+b and 2*(b-a)?
Almost nothing has changed. You just solve little differently.

Of course if the numbers were a+b and 2*(a+b)
you would be in real trouble. :-)

Best regards.
-- 
   Roberto Ragusa    mail at robertoragusa.it