How to roll a numpy 2D array to create grey level co-occurrence matrices, calculate dissimilarity

[Matteo]

I am very new to GLCM.

I have some 1D array I would like to calculate dissimilarity in the vertical direction, so I am trying to trick scikit-image by converting my 1D array into 2D.

Say I have an 16x5 2D array. I make 5 copies of it to get a 16x5 array. I want to pass rows 0-4, then rows 5-9 and so on to run greycomatrix, by just lookign at the vertical direction,  append to output, and ultimately calculate dissimilarity.

Example array:

    x =np.array([7, 2, 0, 8, 8, 3, 2, 2, 0, 5, 3, 3, 4, 8, 7, 0])
    w=5
    s = np.array([x,]*w,dtype=np.uint8).transpose()


array([
 [7, 7, 7, 7, 7],
 [2, 2, 2, 2, 2],
 [0, 0, 0, 0, 0],
 [8, 8, 8, 8, 8],
 [8, 8, 8, 8, 8],
 [3, 3, 3, 3, 3],
 [2, 2, 2, 2, 2],
 [2, 2, 2, 2, 2],
 [0, 0, 0, 0, 0],
 [5, 5, 5, 5, 5],
 [3, 3, 3, 3, 3],
 [3, 3, 3, 3, 3],
 [4, 4, 4, 4, 4],
 [8, 8, 8, 8, 8],
 [7, 7, 7, 7, 7],
 [0, 0, 0, 0, 0]], dtype=uint8)

Below I test rolling:

test = np.roll(s[:, :], -5)
test

array([[2, 2, 2, 2, 2],
       [0, 0, 0, 0, 0],
       [8, 8, 8, 8, 8],
       [8, 8, 8, 8, 8],
       [3, 3, 3, 3, 3],
       [2, 2, 2, 2, 2],
       [2, 2, 2, 2, 2],
       [0, 0, 0, 0, 0],
       [5, 5, 5, 5, 5],
       [3, 3, 3, 3, 3],
       [3, 3, 3, 3, 3],
       [4, 4, 4, 4, 4],
       [8, 8, 8, 8, 8],
       [7, 7, 7, 7, 7],
       [0, 0, 0, 0, 0],
       [7, 7, 7, 7, 7]], dtype=uint8)

Now I roll and pass top 5 rows to greycomatrix:

levels = 9# this rolls the 2D array up one row at a time, each time calculating GLCM on the top 5x5 elements
out = []
counter = 0
while counter < np.shape(s)[0]:
 s = np.roll(s[:, :], -5)
 glcm = greycomatrix(s[:5,:], [1], [np.pi/2], levels = levels, symmetric = True, normed = True)
 out.append(glcm)
 counter +=1

But unfortunately output is a 5D array:

print np.shape(out)
(16, 9, 9, 1, 1)

I tried to recast into a 4D array to pass to greycoprops in different ways, but unsuccessfully.

With: 
out1 = np.array(out).reshape((16, 9, 9, 1)) 
np.shape(out1)
(16, 9, 9, 1)

diss = greycoprops(out1, 'dissimilarity')

I get this error:

---------------------------------------------------------------------------
AssertionError Traceback (most recent call last)
<ipython-input-15-a88f05b0e3e2> in <module>()
----> 1 diss = greycoprops(out1, 'dissimilarity')
 2 diss

/Users/matteoniccoli/anaconda2/lib/python2.7/site-packages/skimage/feature/texture.pyc in greycoprops(P, prop)
 187 
 188 (num_level, num_level2, num_dist, num_angle) = P.shape
--> 189 assert num_level == num_level2
 190 assert num_dist > 0
 191 assert num_angle > 0

AssertionError:


It may be that I do not really understand the shape of the output, and the requirements of greycoprops.

print np.shape(np.array(out))
print np.shape(np.array(out[0][:][:][:][:]))
print np.shape(np.array(out[:][0][:][:][:]))
print np.shape(np.array(out[:][:][0][:][:]))
print np.shape(np.array(out[:][:][:][0][:]))
print np.shape(np.array(out[:][:][:][:][0]))


(16, 9, 9, 1, 1)
(9, 9, 1, 1)
(9, 9, 1, 1)
(9, 9, 1, 1)
(9, 9, 1, 1)
(9, 9, 1, 1)

Can anybody suggest a way to make this work?

Thanks

[Matteo]

Sorry, below I meant to write:"Say I have an 16x1 1D array. I make 5 copies of it to get a 16x5 2D array..."

[Matteo]

My apologies.

It is clear now that the issue was not in the idea for the implementation, or the code, but clearly my incomplete understanding of graycomatrix and graycoprops.

Since I am sliding along a 16x5 array, with 9 levels, I should get a 16x9x9x1x1 5D array. Now that I understand this better, it occurs to me that I just need to slide over this array also, passing each of the 16  9x9x1x1 4D partial array to greycopros, to get 16 values of dissimilarity. Does that make sense?

diss = []
for i in np.arange(np.shape(out)[0]):
    ds = greycoprops(out[i], 'dissimilarity')
    diss.append(ds)
diss  

will give:

[array([[ 3.75]]),
 array([[ 3.5]]),
 array([[ 1.5]]),
 array([[ 2.]]),
 array([[ 2.]]),
 array([[ 2.25]]),
 array([[ 2.25]]),
 array([[ 2.]]),
 array([[ 1.75]]),
 array([[ 1.5]]),
 array([[ 3.25]]),
 array([[ 4.75]]),
 array([[ 5.]]),
 array([[ 5.25]]),
 array([[ 5.5]]),
 array([[ 3.75]])]

Thank you