[SciPy-user] 2D and 3D moments of an image (Hu, Zernike, ...)

Stefan van der Walt stefan@sun.ac...
Tue Jan 22 14:21:48 CST 2008

Hi Zachary

On Tue, Jan 22, 2008 at 12:52:51PM -0500, Zachary Pincus wrote:
> wasn't very much work. I would almost suggest that writing them from  
> scratch as numpy operations might be easier than trying to deal with  
> strange C-code. (And not slower -- I think that both can easily be  
> cast in terms of ufuncs and whole-matrix operations, etc.)

I'd be very interested in such a solution, and I think it may prove
somewhat challenging (calculating the grey-level co-occurrence matrix
effectively, using only numpy operations).

> I'm happy to talk at more length about the possibility of cobbling  
> together such a scikit, if anyone's interested.

I am all for the idea.  Ndimage was written before numpy was on the
scene, and now we can replace a lot of its functionality using Python
code (that would execute just as fast!).

I recall that, when the sandbox still existed, there was some
colour-space conversion code that we could incorporate.  I also have
some computer vision, linear-filtering, non-linear operator and image
restoration code, which complements the current ndimage functionality.

Looking forward to hearing more on the topic.


More information about the SciPy-user mailing list