[Insight-users] Re: Re:filter

[Ying Zhuge]

Hi Martino,

This filter is a little bit different from that paper you mensioned.

1. This filter is designed for vector-valued image at the beginning, but it
is suitable for scalar image also.
2. This filter is for relative fuzzy connected image segmentation, please
refer to IEEE PAMI paper pp. 1485-1500, 2002.

In the code, there is a member function called compute_scale() to estimate
the scale value of each pixel, and another member function
compute_affinity() to estimate both homogeneity and object-feature based
components. The implementation of homogeneity-based affinity is different
from the definition in that paper because if we do that way, it will take a
lot of computation time for vector-valued image. w(||c-e||) is a distance
weighting function. The closer pixel e to the center c, the bigger weight
number it has. In my implementation, I use the Gaussian as the weighting
function.

Hope it helps. Please let me know if you still have questions.

--Ying


----- Original Message ----- 
From: "Martino Gabriele" <g91182 at libero.it>
To: "zhuge" <zhuge at mipg.upenn.edu>
Sent: Thursday, July 22, 2004 3:12 PM
Subject: Re:filter


> Hi Ying Zhuge,
>
> im very happy you answered my question.
> Im working with fuzzy segmentation and,
> in particular, with 2D images.I may implement
> scale-based approch and so ive read the paper
> "Scale-based Fuzzy Connected Image Segmentation"
> in which there are some alghoritms about that.
> First there is an alghoritm to estimate local scale
> for each pixel and then there is a presentation of
> homegeneity and object-feature based components
> with scale-based approch.
> Id like to know how itk implements these alghoritms,
> namely in which part of itk code i can find the object
> scale estimation of pixels and the implementation
> of the two components and what are the relations
> between itk implementation and the presentation of the paper.
> Id like to know that also because ive not understood
> very well the aim of the formulations that are in the paper,
> namely the sense of weighted sums, the sense of
>  w( || c-e || ) .
> So id like to have a general vision of that.
> But isnt this filter already implemented by itk?
> Im very sorry if my english is not good.Im a
> simple student who's dealing with these very interesting
> arguments and i think itk is the best tool we have for
> image segmentation.This software is fantastic.
>
> Thanks a lot anf good work!
>
> Gabriele
>
>
> > : Theory,Alghoritms and Validation"
>
> ---------- Initial Header -----------
>