Re: Selective Gaussian -> Early test app posted


"Mattias Andersson" <mattias@xxxxxxxxxxxxx> wrote:
>Lord Crc wrote:
>>
>> I belive the Selective Gaussian is not separable, because you do not
>> use all pixels when computing it (hence the name of the algorithm :).
>> Since you've lost this, you might aswell allow arbitrary kernels.

[snip]

>Now, consider the following two expressions:
>
> Sum[ y = y1..y2 ] Sum[ x = x1..x2 ] ( p[x, y] a[x, y] exp(-x^2/c)
>exp(-y^2/c) )
>
> <=>
>
> Sum[ y = y1..y2 ] ( exp(-y^2/c) Sum[ x = x1..x2 ] ( p[x, y] a[x, y]
>exp(-x^2/c) ) )

Since a[] changes for each reference pixel, I can't see how you CANNOT perform SOME kind of operation at each pixel within the radius of the reference pixel. The least you have to do is to actually create a[] (ie do the test). If you then sum each horizontal row in the kernel and then sum vertically, you're actually performing more work than in the original.

So no, I still don't see how this helps at all :)

Then again, it's 4:40am :)

- Asbjørn
.

Relevant Pages

  • Re: Selective Gaussian -> Early test app posted
    ... >> I belive the Selective Gaussian is not separable, ... >> use all pixels when computing it. ... And how exactly do you handle the weight scaling? ... I can't see a way to separate it while ...
    (borland.public.delphi.language.basm)
  • Re: Motion estimation vs. background subtraction
    ... Computing the mean pixel intesity is okay, ... create a new image set as the background. ... of pixels containing of all pixels from ...
    (comp.soft-sys.matlab)