You need to use the nearest op in this case. The other alternative is the sample op, but it interpolates, which messes up an erode type operation. nearest takes a coord from [0, 1] to determine which pixel is grabs. To get the coordinate of a pixel offset from the current pixel being processed, you can use the value 1/dim as the delta between pixels:

Thanks for the response and the patch.

I seem to get only a few pixels returned from the output of the patch though.

This is your patch v cv.jit.erode

The description from the help patch about what the operator does isn't the entire story. It's actually more general. Here's a modified patch that does what the object is doing:

Here's the actual OpenCV function it's using: http://opencv.willowgarage.com/documentation/cpp/image_filtering.html#cv-erode

Ah this is perfect. Thanks for the open CV source also. I need to practice visualising equations in max. I've always had trouble with that.