| one more question on object classification _or_ local/adaptive thresholding [message #30309] |
Wed, 17 April 2002 09:42 |
dmartin
Messages: 5
Registered: April 2002
|
Junior Member |
|
|
I am trying to find objects in a 2-D picture. By find I preferably
mean create a binary image of the objects with which I can use
label_region(). Individually, they have approximately uniform
intensity on a brighter background. However, the illumination varies
widely across the image. Some boring details: the images are 640x480,
and the objects are 20-100 pixels in diameter, and are not necessarily
round, or even close, and are separated by 5-100 pixels.
I have the feeling that some sort of adaptive thresholding is what I
need, but I am not sure of an efficient way of implementing it (I'm
working with tens of thousands of these images).
Right now, I am working with smoothing over the images, and
subtracting this effective background - but this leads to holes and
bright spots (see a few posts ago), which I cannot remove efficiently
(erosion/dilation often joins the objects together).
Would anyone be willing to share a solution to either the overall
problem or the local thresholding problem?
Thanks a million,
Doug Martin
University of Texas at Austin
PS: I am trying to get this to work in both high and low contrast
situations. I haven't had much luck with simplistic edge finding
(Sobel), but perhaps there is an intelligent form that would work
better?
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
