[Insight-developers] RecursiveGaussianImageFilter

[Gunnar Farneback]

Jim wrote:
> You might want to take a look at
> 
>  * T. Lindeberg. "On Scale Selection for Differential Operators.", Proc.
>  * 8th Scandinavian Conference on Image Analysis. May 1993. pp. 857-866.
> 
> I think you can grab if off of Lindeberg's website.

Yes.

> I'd have to look at the paper again but I think you end up
> multiplying by sigma to normalize across scales. This paper is not
> about the recursive gaussian but derivative gaussian calculations
> across scales. But it may give some insight.

If I understand the paper correctly the idea is to multiply by
sigma^N, where N is the order of the derivative, so that at a given
scale the maximum amplification over the pure sinusoids (the optimal
frequency is given by sqrt(N)/sigma) is constant. But that is not at
all what the current implementation does; it's multiplying by sigma,
regardless of N.

/Gunnar