I have an image.
I want to obtain a 3x3 window (neighbouring pixels) for every pixel in the image.
I have this Python code:
for x in range(2,r-1,1):
for y in range(2,c-1,1):
mask5=numpy.array([cv.Get2D(copy_img,x-1,y-1),cv.Get2D(copy_img,x-1,y),cv.Get2D(copy_img,x-1,y+1),cv.Get2D(copy_img,x,y-1),cv.Get2D(copy_img,x,y),cv.Get2D(copy_img,x,y+1),cv.Get2D(copy_img,x+1,y-1),cv.Get2D(copy_img,x+1,y),cv.Get2D(copy_img,x+1,y+1)])
cent=[cv.Get2D(copy_img,x,y)]
mask5 is the 3x3 window. cent is the center pixel.
Is there a more efficient way to do this - i.e. using maps, iterators - anything but the two nested loops I've used?
This can be done faster, by reshaping and swapping axes, and then repeating over all kernel elements, like this:
im = np.arange(81).reshape(9,9)
print np.swapaxes(im.reshape(3,3,3,-1),1,2)
This gives you an array of 3*3 tiles which tessalates across the surface:
[[[[ 0 1 2] [[ 3 4 5] [[ 6 7 8]
[ 9 10 11] [12 13 14] [15 16 17]
[18 19 20]] [21 22 23]] [24 25 26]]]
[[[27 28 29] [[30 31 32] [[33 34 35]
[36 37 38] [39 40 41] [42 43 44]
[45 46 47]] [48 49 50]] [51 52 53]]]
[[[54 55 56] [[57 58 59] [[60 61 62]
[63 64 65] [66 67 68] [69 70 71]
[72 73 74]] [75 76 77]] [78 79 80]]]]
To get the overlapping tiles we need to repeat this 8 further times, but 'wrapping' the array, by using a combination of vstack
and column_stack
. Note that the right and bottom tile arrays wrap around (which may or may not be what you want, depending on how you are treating edge conditions):
im = np.vstack((im[1:],im[0]))
im = np.column_stack((im[:,1:],im[:,0]))
print np.swapaxes(im.reshape(3,3,3,-1),1,2)
#Output:
[[[[10 11 12] [[13 14 15] [[16 17 9]
[19 20 21] [22 23 24] [25 26 18]
[28 29 30]] [31 32 33]] [34 35 27]]]
[[[37 38 39] [[40 41 42] [[43 44 36]
[46 47 48] [49 50 51] [52 53 45]
[55 56 57]] [58 59 60]] [61 62 54]]]
[[[64 65 66] [[67 68 69] [[70 71 63]
[73 74 75] [76 77 78] [79 80 72]
[ 1 2 3]] [ 4 5 6]] [ 7 8 0]]]]
Doing it this way you wind up with 9 sets of arrays, so you then need to zip them back together. This, and all the reshaping generalises to this (for arrays where the dimensions are divisible by 3):
def new(im):
rows,cols = im.shape
final = np.zeros((rows, cols, 3, 3))
for x in (0,1,2):
for y in (0,1,2):
im1 = np.vstack((im[x:],im[:x]))
im1 = np.column_stack((im1[:,y:],im1[:,:y]))
final[x::3,y::3] = np.swapaxes(im1.reshape(rows/3,3,cols/3,-1),1,2)
return final
Comparing this new
function to looping through all the slices (below), using timeit
, its about 4 times faster, for a 300*300 array.
def old(im):
rows,cols = im.shape
s = []
for x in xrange(1,rows):
for y in xrange(1,cols):
s.append(im[x-1:x+2,y-1:y+2])
return s
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With