| Re: Active Contours and Snakes! Oh, my! [message #37169] |
Thu, 04 December 2003 07:03  |
David Fanning
Messages: 11724
Registered: August 2001
|
Senior Member |
|
|
Haje Korth writes:
> I tested your program and found the subject interesting, too. Presently I
> have no use for the algorithm, but it is always good to know what is out
> there already. Just out of curiosity, which part is crippled. I had some
> trouble snaking the contours directly along the edges. Is that because the
> code is crippled, or did I not play enough with the settings?
Nothing in the snaking algorithm is crippled. I just don't
allow you to pass images to the object directly (you have
to read images from a file) and I don't allow you to
specify the scale factors that would allow you to
report contour perimeter and area in physical values.
These are reported strictly in pixel values in the
demo. Pretty lame crippling (if you enjoy puns).
I'm not sure what you mean by "directly along the
edges". In the application I was working with
originally (the motivation for building an active
contour capability), we first "clipped" the image
with a rectangular mask. This resulted in some
extremely straight edges, which I thought would
be perfect for the algorithm. Not so. The snake
almost always undulates in these areas instead
of converging on a straight line. These kinds
of undulations can be dampened, to some extent,
by increasing either the viscosity (the difficulty
the snake has moving through the medium), or the
elasticity (making it hard to "stretch" the snake).
There is no question you have to play with the
parameters. In fact, that is why the darn application
got so elaborate. You need an infrastructure that allows
you to interact with the algorithm or you are never
going to get it right. I really didn't understand what
I was doing (the bulk of those 100+ hours) until
I had a GUI to play with it.
With some of the snaking algorithms I played with, if
you didn't get the snake parameters just right, the snake
would zip itself up and disappear! Disconcerting after
all those hours making it a pet. :-)
The GVF algorithm was the best of the ones I used
in producing reliable results...well... let me
say *understandable* results most of the time.
Cheers,
David
--
David W. Fanning, Ph.D.
Fanning Software Consulting, Inc.
Phone: 970-221-0438, E-mail: david@dfanning.com
Coyote's Guide to IDL Programming: http://www.dfanning.com/
Toll-Free IDL Book Orders: 1-888-461-0155
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
