[Insight-developers] Flood iterator ++ method

[Damion Shelton]

Jim:

Could you elaborate a little on your suggestion? The current implementation,
in a nutshell, is:

-----------------
Create tempImage as an n-dimensional image of unsigned chars the same size
as source image.
Pixel values in tempImage are:

0=unchecked
1=checked, not in function
2=checked, in function, neighbor check incomplete
3=checked, in function, neighbor check complete

Push the seed index onto the stack.

Do{
Examine the pixel at the index on top of the stack
If the pixel has not been checked (i.e. tempImage[index] = 0)
    If this pixel is inside the function
         tempImage[index] = 2
    Else (it's not inside the function)
        tempImage[index] = 1
        Pop top index off the stack
        Continue (next iteration of do loop)
    Now check all neighbors
        If there's a neighbor that's hasn't been checked
(i.e.tempImage[neighborindex] = 0)
            Push neighbor's index onto stack
        Else
            tempImage[index] = 3
            Pop top index off stack (we've checked all the neighbors)
}While there are no indices on the stack
-----------------

The stack keeps track of where I've been, and the temp image store info
about the state of all of the pixels in the original image
(in/out/checked/unchecked). The stack seems neccessary since it allows me to
reverse my forward progress and ensure that all neighbors are eventually
checked. But, if you have a way around this that would be great.

I'm also not sure how RLE would help - recomputing a RLE compressed temp
image at every iterator step would be an expensive operation, but I'm
probably misunderstanding what you mean.

-Damion-

----- Original Message -----
From: "Miller, James V (CRD)" <millerjv@crd.ge.com>
To: "'Damion Shelton'" <dmsst59@pitt.edu>; "'insight-Developers'"
<insight-developers@public.kitware.com>
Sent: Wednesday, June 06, 2001 7:29 AM
Subject: RE: [Insight-developers] Flood iterator ++ method


> Damion,
>
> Instead of a stack of indices indicating where you have been, you could
keep a mask (image). This
> could be an image of a low bit count or a run length encoded image. You
really only need one bit per
> pixel to indicate that you have visited that pixel.
>
> While this is analogous to your stack storage, it would require less
memory.
>
> Jim