[Insight-users] discrete gaussian smoothing (documentation)‏

[miki maiorca]

 Hi all,

I'm looking for a more detailed documentation of the DiscreteGaussianImageFilter. Unfortunately in the itk software guide there is only a little documentation: it states "gaussian kernels in ITK are constructed according to the theory of Tony Lindeberg[49]". That's it (almost).
In fact, the theory of Tony Lindeberg is huge, I can understand that in few lines it is impossible to syntetize a whole book ([49]). But probably it is too short. i.e. it is not clear to me the criteria behind the MaximumKernelWidth: if I run the code for some 3d images it looks okay, for other (similar, but different size) 3d images there is a warning message(with the same kernel size-variance). 
Also I have no clue about the correlation between the kernel size and the spatial spacing of the image: the figure "Figure 6.21: Discretized Gaussian" of the software guide to me is not clear. In particular it is not clear the meaning of X coords: the X axis represents the number of pixels? If so, kernel size of "6" should be approximately as large as 1 pixel?!? 
Do you know where there is an exhaustive documentation of DiscreteGaussianImageFilter? Or do you have an answer to my questions? ;-) 

thanks,
Miki



_________________________________________________________________
Tutta la posta in un solo posto… scarica Windows Live Mail!!!
http://get.live.com/wlmail/overview