[SciPy-User] possible bug in scipy.ndimage.interpolation.shift

[Gary Ruben]

Thanks for the explanation Zach. I had suspected it might be spline 
related. Also I think you're right that I am provoking ringing at the 
edges in my case but I think I can improve that in my case.

Gary R.

Zachary Pincus wrote:
>> However, perhaps another optional argument, e.g.,  
>> "floor={True,False}" say, (and code to implement, of course) could  
>> be added to the function?
> 
> Eh, ndimage does the right thing when using integral-type outputs,  
> even unsigned ones. And if you're using floating-point images, zero  
> isn't necessarily a meaningful "floor". (Most of my floating-point  
> images are centered around zero, for example.)
> 
> It's probably OK the way it is, just with the caveats that usually  
> apply to fancy spline interpolation, which is that there will be  
> ringing at sharp edges that may go outside of the range of the  
> original values (whatever that range is).
> 
> Zach
> 
> 
> In : ndimage.shift([0,8,0,5,0,3,0,10], -.75, output=float)
> array([ 6.65579912,  1.6582038 ,  3.94576069,  1.04312844,  2.60047557,
>         -0.3512807 ,  8.69527224,  0.        ])
> 
> In : ndimage.shift([0,8,0,5,0,3,0,10], -.75, output=int)
> array([7, 2, 4, 1, 3, 0, 9, 0])