def generate_skeleton_img(skeleton, i_keep, shape, dtype=np.uint8):
max_value = np.iinfo(dtype).max
skeleton_img = np.zeros(shape, dtype=dtype)
for i in i_keep:
px_coords = skeleton.path_coordinates(i).astype(int)
for px_coord in px_coords:
skeleton_img[px_coord[0], px_coord[1]] = max_value
## closing
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
img = skeleton_img
img = cv2.dilate(skeleton_img, kernel, iterations=1)
img = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)
return bin2img(skeletonize(img2bin(img)))
def update_skeleton(skeleton_img, skeleton, i_remove):
dtype = skeleton_img.dtype
max_value = np.iinfo(dtype).max
skeleton_prune_img = skeleton_img.copy()
for i in i_remove:
px_coords = skeleton.path_coordinates(i).astype(int)
rows = px_coords[:, 0]
cols = px_coords[:, 1]
skeleton_prune_img[rows, cols] = 0
## closing
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
img = cv2.dilate(skeleton_prune_img, kernel, iterations=1)
img = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)
img = remove_blobs(img)
return bin2img(skeletonize(img / max_value))
def skeletonize_img(img):
return bin2img(skeletonize(img2bin(img)))
def label2img(labels, label, dim):
data = labels.ravel() == label
img = data.reshape(dim)
return bin2img(img)
def segment_truss(img_hue,
img_a=None,
save="False",
name="",
pwd="",
my_settings=None):
if my_settings is None:
my_settings = settings.segment_image()
n = 3
centers_prior_hue = {
'tomato': np.deg2rad(0), # [rad]
'peduncle': np.deg2rad(90), # [rad]
'background': np.deg2rad(240)
} # [rad]
centers_prior_a = {'tomato': 1.0, 'peduncle': 0.5, 'background': 0.0}
centers_prior = {'hue': centers_prior_hue, 'a': centers_prior_a}
centers = {
'hue': centers_prior['hue'].values(),
'a': centers_prior['a'].values()
}
if img_a is None:
centers, labels = k_means_hue(img_hue, n, centers=centers) # centers
else:
a_min = np.min(img_a)
a_max = np.max(img_a)
img_a_norm = normalize_image(img_a)
centers, _ = k_means_hue_a(img_hue,
img_a_norm,
n,
my_settings,
centers=centers) # centers
labels = assign_labels(img_hue,
centers,
hue_radius=my_settings['hue_radius'],
img_a=img_a)
# determine which center corresponds to which segment
lbl = {}
lbl["tomato"] = np.argmin(
angular_difference(centers['hue'], centers_prior['hue']['tomato']))
lbl["background"] = np.argmin(
angular_difference(centers['hue'], centers_prior['hue']['background']))
lbl["peduncle"] = list(set(range(0, n)) - set(lbl.values()))[0]
# compute masks
dim = img_hue.shape
tomato = label2img(labels, lbl["tomato"], dim)
if np.abs(centers['hue'][lbl["background"]] -
centers['hue'][lbl["peduncle"]]) < np.deg2rad(10):
print "did not detect a peduncle"
peduncle = bin2img(np.zeros(dim))
background = cv2.bitwise_not(tomato)
else:
peduncle = label2img(labels, lbl["peduncle"], dim)
background = label2img(labels, lbl["background"], dim)
if save:
both_hist(img_hue,
img_a_norm,
centers,
lbl,
a_bins=a_max - a_min + 1,
pwd=pwd,
name=name,
hue_radius=my_settings['hue_radius'])
hue_hist(img_hue, np.rad2deg(centers['hue']), lbl, name, pwd)
if img_a is not None:
a_hist(img_a_norm,
centers['a'],
lbl,
bins=a_max - a_min + 1,
name=name,
pwd=pwd)
return background, tomato, peduncle