The application is CT imaging with tagged materials. If there is a thin layer of tissue (0 HU) with brightly tagged materials (>400 HU) one side and very dark air (< -600 HU) on the other, the gradient magnitude at the tissue becomes very high, essentially marking the tissue as an edge. This leads to significant fold erosion in my application.<br>
<br><div class="gmail_quote">On Sun, Jun 19, 2011 at 1:30 PM, David Doria <span dir="ltr"><<a href="mailto:daviddoria@gmail.com">daviddoria@gmail.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class="im">On Sun, Jun 19, 2011 at 1:19 PM, Neil Panjwani <<a href="mailto:paniwani@gmail.com">paniwani@gmail.com</a>> wrote:<br>
> I know the the tophat filters find local minima and maxima, but I'm looking<br>
> for an operation that finds a region that has dark on one side and light on<br>
> the other. Is there anything like that?<br>
<br>
</div>Unless "side" is defined very strongly, like "the region of the image<br>
to the left of a vertical line", I think you'll need to use a<br>
collection of kernels rather than just a single one. The values of<br>
these kernels would be the same as a normal edge detection kernel<br>
(Sobel, etc), but then you would rotate the kernel every few degrees,<br>
and then take the maximal responses over the collection of<br>
convolutions with all of the kernels.<br>
<br>
What is the application? Will a standard edge detection not work?<br>
<font color="#888888"><br>
David<br>
</font></blockquote></div><br>