def calc_patch_mask(self, src, dst, coord=None, overlap=None):
"""
Calculates the blend mask between two patches according to the min cut
in their top and left overlapping areas.
It computes the min-cuts of the difference between the two patches both
in their top and left overlapping areas. Then it uses them to build the
blend mask of the patch.
___________
|/\_/\_/\_/_| min-cut in the top overlap
min-cut |\| |
in the |/| PATCH |
left |\| |
overlap |_|_________|
Args:
src: "background" patch
dst: "foreground" patch
coord: coordinate in tile-size
overlap: depth of the overlap regions
Returns:
blend mask: 0 where background, 1 where foreground
"""
[i, j] = coord
overlap = overlap or self.overlap
# Initialize the mask to all ones
mask = np.ones((src.shape[0], src.shape[1]))
# if it is the first one, or if src==dst, or is dst is empty
if (i == 0 and j == 0) or np.all(src == dst) or not np.any(dst):
return mask
# We have a left overlap
if j > 0:
# difference
diff = power(rgb2gray(src[:, :overlap] - dst[:, :overlap]), 2)
# min-cut
cut = mincut(diff, 0)
# find the mask
mask[:, :overlap] = cut
# We have a top overlap
if i > 0:
# difference
diff = power(rgb2gray(src[:overlap, :] - dst[:overlap, :]), 2)
# min-cut
cut = mincut(diff, 1)
# find the mask
mask[:overlap, :] = multiply(mask[:overlap, :], cut)
# Write to the destination using the mask
return mask
def test_size(self):
"""
Test the returned matrix has the same shape of the input one
"""
a = np.array([[30, 39, 48, 1, 10, 19, 28], [38, 47, 7, 9, 18, 27, 29],
[46, 6, 8, 17, 26, 35, 37], [5, 14, 16, 25, 34, 36, 45],
[13, 15, 24, 33, 42, 44, 4]])
result = mincut(a)
np.testing.assert_array_equal(a.shape, result.shape)
def test_values(self):
"""
Test the returned matrix contains only 0 and 1 values.
"""
a = np.array([[30, 39, 48, 1, 10, 19, 28], [38, 47, 7, 9, 18, 27, 29],
[46, 6, 8, 17, 26, 35, 37], [5, 14, 16, 25, 34, 36, 45],
[13, 15, 24, 33, 42, 44, 4]])
result = mincut(a)
np.testing.assert_array_equal([0, 1], np.unique(result))
def test_small(self):
"""
Test expected output with a small matrix.
"""
a = np.array([[7, 8, 9, 0, 0, 6], [5, 9, 0, 4, 3, 3],
[8, 1, 3, 0, 6, 6]])
result = mincut(a)
expected = np.asarray([[0, 0, 0, 0, 0, 1], [0, 0, 0, 1, 1, 1],
[0, 0, 0, 0, 1, 1]])
self.assertEqual(expected.shape, result.shape)
np.testing.assert_array_equal(expected, result)
def test_transpose(self):
"""
Test expected output with inverted matrix direction.
"""
a = np.array([[30, 39, 48, 1, 10, 19, 28], [38, 47, 7, 9, 18, 27, 29],
[46, 6, 8, 17, 26, 35, 37], [5, 14, 16, 25, 34, 36, 45],
[13, 15, 24, 33, 42, 44, 4]])
result = mincut(a, direction=1)
expected = np.asarray([[0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 1, 1, 1, 1],
[0, 0, 1, 1, 1, 1, 1], [0, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1]])
self.assertEqual(expected.shape, result.shape)
np.testing.assert_array_equal(expected, result)