def _find_regions(self, node):
if node.width == 1:
return 0., 0.
# resigned to split into 2
if node.width == 2:
is_homogeneous = np.allclose(node.pixels[0], node.pixels[1], rtol=5e-2, atol=0)
if is_homogeneous:
idx = int(node.seq, 2)
base = 2 ** len(node.seq)
start = idx * node.width // base
end = start + node.width
else:
stat, isSimilar = normal_test(node.pixels)
if isSimilar:
start =
end = start + node.width
return stat
stat = 0.
node.split()
stat += self._find_regions(node.left_child)
stat += self._find_regions(node.right_child)
return stat
def homogeneity(self, node):
# constant intensity
if (node.pixels == node.pixels[:, 0]).all():
return 0., True
if node.width == 2:
sse = math.sqrt((node.pixels[:, 0] - node.pixels[:, 1]) ** 2)
is_homogeneous = np.allclose(left, right, rtol=5e-2, atol=0)
return sse.sum(), is_homogeneous
# fit a low-order polynomial
x = [i for i in range(node.width)]
degree = 1
p, residuals, rank, _, _ = np.polyfit(x, node.pixels, degree, full=True)
return residuals.sum(), normal_test(residuals)
def homogenity_test(root):
# constant intensity
if (root.pixels == root.pixels[0]).all():
return True
if root.width == 2:
return closeness_test(root.pixels[0], root.pixels[1])
# fit a low-order polynomial
x = [i for i in range(root.width)]
degree = 1
p, residuals, rank, _, _ = np.polyfit(x, root.pixels, degree, full=True)
# the residuals are only some noise
stat, hypothesis = normal_test(residuals)
return hypothesis
def _grow_column_tree(self, node):
if node.width == 1:
return 0.
if node.width == 2:
is_homogeneous = np.allclose(node.pixels[0], node.pixels[1], rtol=5e-2, atol=0)
if is_homogeneous:
return
elif normal_test(root):
return
root.split()
self._grow_column_tree(self, root.left_child)
self._grow_column_tree(self, root.right_child)
return
def similarity_test(self, left_tree, right_tree): diff = left_tree.pixels - right_tree.pixels quality = np.linalg.norm(left_tree.pixels - right_tree.pixels) stat, similar = normal_test(left_tree.pixels - right_tree.pixels) return quality, stat, similar