Erode/DIlate invent pixel values

[kannitverstan]

Hi,

I greatly appreciate your software! It is very well thought out and extremely useful.

Besides of that, I'm pretty sure I have found a bug in erode/dilate. I have attached the TEST.PNG plus the python test file below.

As I understand erosion/dilation, it is the min or max of the local neighbourhood. So it should not produce new pixel values which are not contained in the original image, right?

See attached python file with image. I also cross checked with the scpiy ndimage version, which does not produce the problem.

Regards,
Clemens Steckner

import mahotas
im= mahotas.imread('test.png')

find list of all values of pixels gt 0

def pixelvalues(imi):
im=imi.copy()
lst=[]
mn=im.min()
while True:
val= im.max()
assert (im==val).any()
if val == mn:
break
lst.append(val)
im[im==val] = mn
assert (im==mn).all()
lst.append(mn)
return [i for i in lst if i > 0]

show pixelvalues which have been 'invented' by mahotas

print set(pixelvalues(mahotas.erode(im))) - set(pixelvalues(im))
print set(pixelvalues(mahotas.dilate(im))) - set(pixelvalues(im))

erode invents pixel value 254

print (mahotas.erode(im) == 254).any(), (im==254).any()

dilate invents pixel value 103

print (mahotas.dilate(im) == 103).any(), (im==103).any()

[luispedro]

I don't like this interface 100%, but I adopted it from pymorph (scipy has the same, though).

from scipy import ndimage
cross = np.zeros((3,3), np.uint8)
cross[1] = 1
cross.T[1] = 1
set(pixelvalues(ndimage.grey_dilation(im, structure=cross))) - set(pixelvalues(im))

prints {103}

 set(pixelvalues(ndimage.grey_erosion(im, structure=cross))) - set(pixelvalues(im))

prints {0, 254}.

I think this is a duplicate of #29, so I'm closing here, but I still agree that the issue is confusing.

[luispedro]

If you do want local maxima/minima, there is a function mahotas.morph.locmax which does take a structuring element defining a region. Perhaps check that out.