def customize_threshold_value(reg_count,thresh_val):
labels2 = graph.merge_hierarchical(labels, g, thresh=thresh_val, rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
while not (reg_count <= 15):
thresh_val = 2 + thresh_val
print("Thrshold value is",thresh_val)
labels2 = graph.merge_hierarchical(labels, g, thresh=thresh_val, rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
reg_count = count_regions(labels2)
return labels2,reg_count
def get_patches(img, compactness=30, n_segments=200, rag_thresh=0.08):
"""Get list of patches from image found with SLIC."""
patches = []
img_lab = color.rgb2lab(img)
edges = filters.sobel(color.rgb2gray(img))
labels = segmentation.slic(img_lab, convert2lab=False,
compactness=compactness, n_segments=n_segments)
g = graph.rag_boundary(labels, edges)
segmented = graph.merge_hierarchical(labels, g, thresh=rag_thresh,
rag_copy=False,
in_place_merge=True,
merge_func=merge_boundary,
weight_func=weight_boundary)
idxs_sorted = np.argsort(segmented.reshape((segmented.size)))
_, idxs_start = np.unique(
segmented.reshape((segmented.size))[idxs_sorted],
return_index=True)
idxs_for_values = np.split(idxs_sorted, idxs_start[1:])
for idxs in idxs_for_values:
rows, cols = np.unravel_index(idxs, img.shape[:2])
data = img[rows, cols]
lab = np.mean(img_lab[rows, cols], axis=0)
patches.append((rows, cols, data, lab))
return patches
def get_groups(original):
"""
Finds a segmentation of image by taking an oversegmentation produced by the Priority-Flood watershed and
progressively reducing with a boundary region adjacency graph
:param original: Original RGB image to segment
:return: Segmented image. label = 0 represents an edge, label = -1 represents a pruned area
"""
original = gaussian(original, sigma=1.5, multichannel=True)
original = downscale_to(original, area_limit=2e5)
g = original[:, :, 1]
def weight_boundary(RAG, src, dst, n):
default = {'weight': 0.0, 'count': 0}
count_src = RAG[src].get(n, default)['count']
count_dst = RAG[dst].get(n, default)['count']
weight_src = RAG[src].get(n, default)['weight']
weight_dst = RAG[dst].get(n, default)['weight']
count = count_src + count_dst
return {
'count': count,
'weight': (count_src * weight_src + count_dst * weight_dst) / count
}
greyscale = rgb2gray(original)
gradient = np.hypot(sobel(greyscale, axis=0), sobel(greyscale, axis=1))
segmentation1 = watershed(gradient, markers=400, mask=greyscale > 0.3)
RAG = graph.rag_boundary(segmentation1, gradient)
segmentation2 = graph.merge_hierarchical(segmentation1,
RAG,
thresh=5e-3,
rag_copy=False,
in_place_merge=True,
merge_func=lambda *args: None,
weight_func=weight_boundary)
segmentation2[greyscale < 0.3] = -1
segmentation2 = prune(segmentation2,
abs_threshold=g.size / 1000,
default=-1)
counts, lo, hi = [], [], []
for label in set(segmentation2[segmentation2 >= 0]):
interior = distance_transform_edt(segmentation2 == label) >= 1.5
if np.sum(interior) >= 0:
counts.append(np.sum(interior))
lo.append(np.percentile(gradient[interior], q=70))
hi.append(np.percentile(gradient[interior], q=90))
edges = canny(greyscale,
low_threshold=np.average(lo, weights=counts),
high_threshold=np.average(hi, weights=counts))
edges = binary_dilation(edges, disk(2))
edges = binary_closing(edges, disk(5))
edges = remove_small_objects(edges, g.size / 1000)
edges = edges[1:-1, 1:-1]
edges = np.pad(edges, pad_width=1, mode='constant', constant_values=1)
groups = im_label(edges, background=1, connectivity=1)
groups = prune(groups, abs_threshold=g.size / 1000)
groups[greyscale <
0.15] = -2 # Ignore black areas due to mechanical vignetting
return groups
def hierarchical_merging_of_region_boundary_rags_example():
#img = data.coffee()
edges = sobel(skimage.color.rgb2gray(img))
labels = slic(img, compactness=30, n_segments=400)
g = graph.rag_boundary(labels, edges)
graph.show_rag(labels, g, img)
plt.title('Initial RAG')
labels2 = graph.merge_hierarchical(labels,
g,
thresh=0.08,
rag_copy=False,
in_place_merge=True,
merge_func=merge_boundary,
weight_func=weight_boundary)
graph.show_rag(labels, g, img)
plt.title('RAG after hierarchical merging')
plt.figure()
out = skimage.color.label2rgb(labels2, img, kind='avg')
plt.imshow(out)
plt.title('Final segmentation')
plt.show()
def show_segmen_rag(array,
pathlibpath,
numSegments,
threshold,
cedges=177,
compactness=0.1,
sigma=5,
convert2lab=False):
print('Obtaining superstructures')
segments = slic(array,
compactness=compactness,
n_segments=numSegments,
sigma=sigma,
multichannel=False,
convert2lab=convert2lab)
print('Number of SV: ', len(np.unique(segments)))
segments += 1
edges = filters.sobel(array)
print('Obtaining RAG')
rag = graph.rag_boundary(segments, edges, connectivity=1)
segments2 = graph.merge_hierarchical(segments,
rag,
threshold=threshold,
in_place_merge=True,
rag_copy=False,
merge_func=merge_boundary,
weight_func=weight_boundary)
print('Final Number of SV: ', len(np.unique(segments2)))
graph.show_rag(segments, rag, array, edge_cmap='viridis')
save_yorno(array, segments, pathlibpath, cedges)
def merge_hierarchical_mean_color(labels,
rag,
thresh,
rag_copy=True,
in_place_merge=False):
return graph.merge_hierarchical(labels, rag, thresh, rag_copy,
in_place_merge, _pre_merge_mean_color,
_weight_mean_color)
def segmentation_reduction(img,
labels,
n_segments,
reduction=None,
kind='mix',
cs=None):
if reduction == 'selective':
# selective search
img_cvtcolor = label2rgb(labels,
img,
kind=kind,
bg_label=-1,
bg_color=(0, 0, 0))
if cs == 'lab':
img_cvtcolor = cv2.cvtColor(img_cvtcolor, cv2.COLOR_BGR2LAB)
elif cs == 'hsv':
img_cvtcolor = cv2.cvtColor(img_cvtcolor, cv2.COLOR_BGR2HSV)
merged_labels = selective_search(img_cvtcolor,
labels,
seg_num=n_segments,
sim_strategy='CTSF')
rgbmap = label2rgb(merged_labels, img, kind=kind)
elif reduction == 'cluster':
# aggregate colors in each of the labels and output
_, rbg_labels = label2rgb(labels,
img,
kind=kind,
bg_label=-1,
bg_color=(0, 0, 0),
ret_rbg_labels=True)
# apply kmeans clustering to the resulting color labels
ret, klabels, centroids = apply_kmeans(
np.array(rbg_labels, dtype=np.float32), n_segments)
reduced_colors = centroids[klabels.flatten()]
rgbmap = label2rgb(labels, img, reduced_colors=reduced_colors)
elif reduction == 'rag':
# Region Adjacency Graph (RAG)
g = rag_mean_color(img, labels)
merged_labels = merge_hierarchical(labels,
g,
thresh=35,
rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
rgbmap = label2rgb(merged_labels,
img,
kind=kind,
bg_label=-1,
bg_color=(0, 0, 0))
else:
rgbmap = img
return rgbmap
def cluster_merge(volume, labels):
g = graph.rag_mean_color(volume, labels)
labels_merged = graph.merge_hierarchical(labels,
g,
thresh=0.03,
rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=weight_mean_color)
return labels_merged
def initial_formation(self, labels, g):
threshold_value = 50
labels2 = graph.merge_hierarchical(labels,
g,
thresh=threshold_value,
rag_copy=False,
in_place_merge=True,
merge_func=self.merge_mean_color,
weight_func=self._weight_mean_color)
return labels2
def graphMergeHierarchical(self,img, labels,thresh=0.15,plot=True,save=0):
g = graph.rag_mean_color(img,labels)
new_labels = graph.merge_hierarchical(labels, g, thresh, rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
if plot:
plotLibs().dispImg(color.label2rgb(new_labels, img, kind='avg'),save=save,title='Merge Hierarchical Label rgb Image')
plotLibs().plotBoundaries(img,new_labels,save=save,title='Merge Hierarchical')
return new_labels
def segment_image(self,img,name):
labels = segmentation.slic(img, compactness=30, n_segments=200)
g = graph.rag_mean_color(img, labels)
labels2 = graph.merge_hierarchical(labels, g, thresh=80, rag_copy=False,
in_place_merge=True,
merge_func=self.merge_mean_color,
weight_func=self._weight_mean_color)
self.result = color.label2rgb(labels2, img, kind='avg')
cv2.imshow('segmentation : '+str(name), self.result)
k = cv2.waitKey(1) & 0xFF
return self.result
def partition_image(img, n_segments):
# lab_img = cv2.cvtColor(img, cv2.COLOR_BGR2LAB)
superpixels = slic(img, n_segments, sigma=sigma, multichannel=multichannel, convert2lab=convert2lab,
slic_zero=slic_zero, enforce_connectivity=False)
ent_rag = graph.rag_mean_color(img, superpixels)
regions = graph.merge_hierarchical(superpixels, ent_rag, thresh=hier_thresh, rag_copy=rag_copy,
in_place_merge=in_place_merge,
merge_func=merge_mean_color,
weight_func=weight_mean_color)
# out = color.label2rgb(regions, img, kind='avg')
# out = segmentation.mark_boundaries(out, labels2, (0, 0, 0))
return regions
def _apply(self, imgmsg, maskmsg):
bridge = cv_bridge.CvBridge()
img = bridge.imgmsg_to_cv2(imgmsg)
if img.ndim == 2:
img = gray2rgb(img)
mask = bridge.imgmsg_to_cv2(maskmsg, desired_encoding='mono8')
mask = mask.reshape(mask.shape[:2])
mask = gray2rgb(mask)
# compute label
roi = closed_mask_roi(mask)
roi_labels = masked_slic(img=img[roi],
mask=mask[roi],
n_segments=20,
compactness=30)
if roi_labels is None:
return
labels = np.zeros(mask.shape, dtype=np.int32)
# labels.fill(-1) # set bg_label
labels[roi] = roi_labels
if self.is_debugging:
# publish debug slic label
slic_labelmsg = bridge.cv2_to_imgmsg(labels)
slic_labelmsg.header = imgmsg.header
self.pub_slic.publish(slic_labelmsg)
# compute rag
g = rag_solidity(labels, connectivity=2)
if self.is_debugging:
# publish debug rag drawn image
rag_img = draw_rag(labels, g, img)
rag_img = img_as_uint(rag_img)
rag_imgmsg = bridge.cv2_to_imgmsg(rag_img.astype(np.uint8),
encoding='rgb8')
rag_imgmsg.header = imgmsg.header
self.pub_rag.publish(rag_imgmsg)
# merge rag with solidity
merged_labels = merge_hierarchical(labels,
g,
thresh=1,
rag_copy=False,
in_place_merge=True,
merge_func=_solidity_merge_func,
weight_func=_solidity_weight_func)
merged_labels += 1
merged_labels[mask == 0] = 0
merged_labelmsg = bridge.cv2_to_imgmsg(merged_labels.astype(np.int32))
merged_labelmsg.header = imgmsg.header
self.pub.publish(merged_labelmsg)
if self.is_debugging:
out = label2rgb(merged_labels, img)
out = (out * 255).astype(np.uint8)
out_msg = bridge.cv2_to_imgmsg(out, encoding='rgb8')
out_msg.header = imgmsg.header
self.pub_label.publish(out_msg)
def get_segments(self, frame):
rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
labels = slic(rgb, self.n_segments)
edges = filters.sobel(color.rgb2gray(rgb))
g = graph.rag_boundary(labels, edges)
labels2 = graph.merge_hierarchical(labels,
g,
thresh=0.02,
rag_copy=False,
in_place_merge=True,
merge_func=merge_boundary,
weight_func=weight_boundary)
return labels2.astype(np.uint8)
def process(self, data):
with Timer('superpixelnode_2'):
im_l, im_r, rgb, labels = data
edges = filters.sobel(color.rgb2gray(rgb))
g = graph.rag_boundary(labels, edges)
labels2 = graph.merge_hierarchical(labels,
g,
thresh=0.02,
rag_copy=False,
in_place_merge=True,
merge_func=merge_boundary,
weight_func=weight_boundary)
return im_l, im_r, labels2.astype(np.uint8)
def preview(self, ips, para):
lab = self.app.get_img(para['lab']).img
connect = ['4-connected', '8-connected'].index(para['connect']) + 1
g = graph.rag_mean_color(ips.snap, lab, connect, para['mode'],
para['sigma'])
lab = graph.merge_hierarchical(lab,
g,
thresh=para['thresh'],
rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
ips.img[:] = color.label2rgb(lab, ips.snap, kind='avg')
def _merge_superpixels(
self,
input_image: np.ndarray,
initial_superpixels: np.ndarray,
tissue_mask: np.ndarray = None,
) -> np.ndarray:
"""Merge the initial superpixels to return merged superpixels
Args:
image (np.array): Input image
initial_superpixels (np.array): Initial superpixels
tissue_mask (None, np.array): Tissue mask
Returns:
np.array: Output merged superpixel tensor
"""
if tissue_mask is not None:
# Remove superpixels belonging to background or having < 10% tissue
# content
ids_initial = np.unique(initial_superpixels, return_counts=True)
ids_masked = np.unique(tissue_mask * initial_superpixels,
return_counts=True)
ctr = 1
superpixels = np.zeros_like(initial_superpixels)
for i in range(len(ids_initial[0])):
id = ids_initial[0][i]
if id in ids_masked[0]:
idx = np.where(id == ids_masked[0])[0]
ratio = ids_masked[1][idx] / ids_initial[1][i]
if ratio >= 0.1:
superpixels[initial_superpixels == id] = ctr
ctr += 1
initial_superpixels = superpixels
# Merge superpixels within tissue region
g = self._generate_graph(input_image, initial_superpixels)
merged_superpixels = graph.merge_hierarchical(
initial_superpixels,
g,
thresh=self.threshold,
rag_copy=False,
in_place_merge=True,
merge_func=self._merging_function,
weight_func=self._weighting_function,
)
merged_superpixels += 1 # Handle regionprops that ignores all values of 0
mask = np.zeros_like(initial_superpixels)
mask[initial_superpixels != 0] = 1
merged_superpixels = merged_superpixels * mask
return merged_superpixels
def _apply(self, imgmsg, maskmsg):
bridge = cv_bridge.CvBridge()
img = bridge.imgmsg_to_cv2(imgmsg)
if img.ndim == 2:
img = gray2rgb(img)
mask = bridge.imgmsg_to_cv2(maskmsg, desired_encoding='mono8')
mask = mask.reshape(mask.shape[:2])
mask = gray2rgb(mask)
# compute label
roi = closed_mask_roi(mask)
roi_labels = masked_slic(img=img[roi], mask=mask[roi],
n_segments=20, compactness=30)
if roi_labels is None:
return
labels = np.zeros(mask.shape, dtype=np.int32)
# labels.fill(-1) # set bg_label
labels[roi] = roi_labels
if self.is_debugging:
# publish debug slic label
slic_labelmsg = bridge.cv2_to_imgmsg(labels)
slic_labelmsg.header = imgmsg.header
self.pub_slic.publish(slic_labelmsg)
# compute rag
g = rag_solidity(labels, connectivity=2)
if self.is_debugging:
# publish debug rag drawn image
rag_img = draw_rag(labels, g, img)
rag_img = img_as_uint(rag_img)
rag_imgmsg = bridge.cv2_to_imgmsg(
rag_img.astype(np.uint8), encoding='rgb8')
rag_imgmsg.header = imgmsg.header
self.pub_rag.publish(rag_imgmsg)
# merge rag with solidity
merged_labels = merge_hierarchical(
labels, g, thresh=1, rag_copy=False,
in_place_merge=True,
merge_func=_solidity_merge_func,
weight_func=_solidity_weight_func)
merged_labels += 1
merged_labels[mask == 0] = 0
merged_labelmsg = bridge.cv2_to_imgmsg(merged_labels.astype(np.int32))
merged_labelmsg.header = imgmsg.header
self.pub.publish(merged_labelmsg)
if self.is_debugging:
out = label2rgb(merged_labels, img)
out = (out * 255).astype(np.uint8)
out_msg = bridge.cv2_to_imgmsg(out, encoding='rgb8')
out_msg.header = imgmsg.header
self.pub_label.publish(out_msg)
def rag_merging_example():
img = data.coffee()
labels = slic(img, compactness=30, n_segments=400)
g = graph.rag_mean_color(img, labels)
labels2 = graph.merge_hierarchical(labels,
g,
thresh=35,
rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
out = skimage.color.label2rgb(labels2, img, kind='avg')
out = mark_boundaries(out, labels2, (0, 0, 0))
skimage.io.imshow(out)
skimage.io.show()
def get_segments(self, frame):
if not self.initialized:
self.rgb = np.empty_like(frame)
self.lab = np.empty_like(frame)
self.initialized = True
#with Timer('super-labels1'):
cv2.cvtColor(frame, cv2.COLOR_BGR2RGB, dst=self.rgb)
labels = slic(self.rgb, self.n_segments)
#with Timer('super-labels2'):
edges = filters.sobel(color.rgb2gray(self.rgb))
g = graph.rag_boundary(labels, edges)
labels2 = graph.merge_hierarchical(labels,
g,
thresh=0.02,
rag_copy=False,
in_place_merge=True,
merge_func=merge_boundary,
weight_func=weight_boundary)
return labels2.astype(np.uint8)
def run(self, ips, imgs, para=None):
if not para['stack']:
imgs, labs = [ips.img], [self.app.get_img(para['lab']).img]
else:
labs = self.app.get_img(para['lab']).imgs
if len(imgs) != len(labs):
labs = [self.app.get_img(para['lab']).img] * len(imgs)
for i in range(len(imgs)):
img, lab = imgs[i], labs[i]
connect = ['4-connected', '8-connected'].index(para['connect']) + 1
g = graph.rag_mean_color(img, lab, connect, para['mode'],
para['sigma'])
lab = graph.merge_hierarchical(lab,
g,
thresh=para['thresh'],
rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
img[:] = color.label2rgb(lab, img, kind='avg')
self.progress(i, len(imgs))
def rag_merge_threshold(edges, labels, threshold, size_pen):
'''
Merge adjacent segments using region adjacency graph based on strength of
edge between them.
'''
# create region adjacency graph using the edge info to produce
# edge weights between the nodes.
click.echo('creating Region Adjacency Graph')
rag = graph.rag_boundary(labels, edges)
# calculate pixel counts for each node
lab, counts = np.unique(labels.ravel(), return_counts=True)
click.echo('starting with {} segments'.format(lab.max()))
counts = (counts / counts.max()) * size_pen
for i, n in enumerate(lab):
rag.node[n].update({'pixels': counts[i]})
# update initial edge weights to weight by mean size of nodes
for n1, n2 in rag.edges_iter():
total_pix = rag.node[n1]['pixels'] + rag.node[n2]['pixels']
rag.edge[n1][n2]['weight'] += total_pix
rag.edge[n2][n1]['weight'] += total_pix
# calculate a value from the threshold percentile of edge weights.
edge_weights = [x[2]['weight'] for x in rag.edges(data=True)]
t = np.percentile(edge_weights, threshold)
# merge adjacent labels iteratively if their edge weight is below the
# required threshold.
click.echo('merging segments with edge weights below {} percentile'.format(
threshold))
refined_labels = graph.merge_hierarchical(
labels, rag, t, rag_copy=True, in_place_merge=True,
merge_func=merge_nodes, weight_func=update_edge_weights)
refined_labels, *_ = segmentation.relabel_sequential(refined_labels + 1,
offset=1)
click.echo('merged into {} segments'.format(refined_labels.max()))
return refined_labels
def merge_hier_boundary(labels, image, thresh=0.03, show_rag=False):
"""
Merges the given labels using a RAG based on boundaries.
Parameters
----------
labels: ndarray
image: ndarray
thresh: float
show_rag: bool
Returns
-------
rag: RAG
labels: ndarray
Merged labels.
"""
edges = filters.sobel(color.rgb2gray(image))
rag = graph.rag_boundary(labels, edges)
rag_copy = False
if show_rag:
rag_copy = True
fig, ax = plt.subplots(1, 2, figsize=(10, 10))
labels = graph.merge_hierarchical(labels,
rag,
thresh=thresh,
rag_copy=rag_copy,
in_place_merge=True,
merge_func=merge_boundary,
weight_func=weight_boundary)
if show_rag:
graph.show_rag(labels, rag, image, ax=ax[0])
ax[0].title('Initial RAG')
graph.show_rag(labels, graph.rag_boundary(labels, edges), ax=ax[1])
ax[1].title('Final RAG')
return rag, labels
def merge_hier_color(labels, image, thresh=0.08, show_rag=False):
"""
Merges the given labels using a RAG based on the mean color.
Parameters
----------
labels: ndarray
image: ndarray
thresh: float
show_rag: bool
Returns
-------
rag: RAG
labels: ndarray
Merged labels.
"""
rag = graph.rag_mean_color(image, labels)
rag_copy = False
if show_rag:
rag_copy = True
fig, ax = plt.subplots(1, 2, figsize=(10, 10))
labels = graph.merge_hierarchical(labels,
rag,
thresh=thresh,
rag_copy=rag_copy,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
# labels2 = graph.cut_normalized(img_slic, rag, thresh=30)
# labels2 = graph.cut_threshold(img_slic, rag, 0.2)
if show_rag:
graph.show_rag(labels, rag, image, ax=ax[0])
ax[0].title('Initial RAG')
graph.show_rag(labels, graph.rag_mean_color(image, labels), ax=ax[1])
ax[1].title('Final RAG')
return rag, labels
def rag_merging_segmentation(image):
# input - PIL Image
img = np.array(image)
labels = segmentation.slic(img, compactness=30, n_segments=350)
g = graph.rag_mean_color(img, labels)
labels2 = graph.merge_hierarchical(labels,
g,
thresh=45,
rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
# for x in np.unique(labels2):
# vol = sum(sum(labels2 == x))
# if vol > 1200 or vol < 100:
# labels2[labels2 == x] = 0
return np.array(drop_invalid_objects(labels2,
lower_threshold=100,
upper_threshold=1200),
dtype='int16')
def merge_RAG_hierarchical(input_img, input_labels, threshold):
# create Region Adjacency Graph with rag mean color function
g = graph.rag_mean_color(input_img, input_labels)
labels2 = graph.merge_hierarchical(input_labels,
g,
thresh=threshold,
rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
print("labels2 complete")
# This print statement will print out the array of merged segments
# print(labels2[0])
# out = color.label2rgb(labels2, input_img, kind='avg')
# out = segmentation.mark_boundaries(out, labels2, (0, 0, 0))
# fig = plt.figure(figsize=(12,12))
# plt.imshow(out)
# plt.show()
return labels2
def graph_seg(img, max_dist=200, thresh=80, sigma=255.0):
"""
segment an image using quickshift and merge_hierarchical
from scikit-image. In principle, any segmentation method
can be used, this is just one example.
"""
img = img.astype("float")
seg = segmentation.quickshift(
img,
max_dist=max_dist,
)
g = graph.rag_mean_color(
img,
seg,
sigma=sigma,
)
seg = graph.merge_hierarchical(seg,
g,
thresh=thresh,
rag_copy=False,
in_place_merge=True,
merge_func=_merge_mean_color,
weight_func=_weight_mean_color)
return seg
def _apply(self, imgmsg, maskmsg):
bridge = cv_bridge.CvBridge()
img = bridge.imgmsg_to_cv2(imgmsg)
if img.ndim == 2:
img = gray2rgb(img)
mask = bridge.imgmsg_to_cv2(maskmsg, desired_encoding='mono8')
mask = mask.reshape(mask.shape[:2])
# compute label
roi = closed_mask_roi(mask)
roi_labels = masked_slic(img=img[roi],
mask=mask[roi],
n_segments=20,
compactness=30)
if roi_labels is None:
return
labels = np.zeros(mask.shape, dtype=np.int32)
# labels.fill(-1) # set bg_label
labels[roi] = roi_labels
if self.is_debugging:
# publish debug slic label
slic_labelmsg = bridge.cv2_to_imgmsg(labels)
slic_labelmsg.header = imgmsg.header
self.pub_slic.publish(slic_labelmsg)
# compute rag
g = rag_solidity(labels, connectivity=2)
if self.is_debugging:
# publish debug rag drawn image
if LooseVersion(skimage.__version__) >= '0.13.0':
fig, ax = plt.subplots(figsize=(img.shape[1] * 0.01,
img.shape[0] * 0.01))
show_rag(labels, g, img, ax=ax)
ax.axis('off')
plt.subplots_adjust(0, 0, 1, 1)
fig.canvas.draw()
w, h = fig.canvas.get_width_height()
rag_img = np.fromstring(fig.canvas.tostring_rgb(),
dtype=np.uint8)
rag_img.shape = (h, w, 3)
plt.close()
else:
rag_img = draw_rag(labels, g, img)
rag_img = img_as_uint(rag_img)
rag_imgmsg = bridge.cv2_to_imgmsg(rag_img.astype(np.uint8),
encoding='rgb8')
rag_imgmsg.header = imgmsg.header
self.pub_rag.publish(rag_imgmsg)
# merge rag with solidity
merged_labels = merge_hierarchical(labels,
g,
thresh=1,
rag_copy=False,
in_place_merge=True,
merge_func=_solidity_merge_func,
weight_func=_solidity_weight_func)
merged_labels += 1
merged_labels[mask == 0] = 0
merged_labelmsg = bridge.cv2_to_imgmsg(merged_labels.astype(np.int32))
merged_labelmsg.header = imgmsg.header
self.pub.publish(merged_labelmsg)
if self.is_debugging:
out = label2rgb(merged_labels, img)
out = (out * 255).astype(np.uint8)
out_msg = bridge.cv2_to_imgmsg(out, encoding='rgb8')
out_msg.header = imgmsg.header
self.pub_label.publish(out_msg)
def merge_hierarchical_mean_color(labels, rag, thresh, rag_copy=True,
in_place_merge=False):
return graph.merge_hierarchical(labels, rag, thresh, rag_copy,
in_place_merge, _pre_merge_mean_color,
_weight_mean_color)
This method computes the mean color of `dst`.
Parameters
----------
graph : RAG
The graph under consideration.
src, dst : int
The vertices in `graph` to be merged.
"""
graph.node[dst]['total color'] += graph.node[src]['total color']
graph.node[dst]['pixel count'] += graph.node[src]['pixel count']
graph.node[dst]['mean color'] = (graph.node[dst]['total color'] /
graph.node[dst]['pixel count'])
img = data.coffee()
labels = segmentation.slic(img, compactness=30, n_segments=400)
g = graph.rag_mean_color(img, labels)
labels2 = graph.merge_hierarchical(labels, g, thresh=35, rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
g2 = graph.rag_mean_color(img, labels2)
out = color.label2rgb(labels2, img, kind='avg')
out = segmentation.mark_boundaries(out, labels2, (0, 0, 0))
io.imshow(out)
io.show()
This method computes the mean color of `dst`.
Parameters
----------
graph : RAG
The graph under consideration.
src, dst : int
The vertices in `graph` to be merged.
"""
graph.nodes[dst]['total color'] += graph.nodes[src]['total color']
graph.nodes[dst]['pixel count'] += graph.nodes[src]['pixel count']
graph.nodes[dst]['mean color'] = (graph.nodes[dst]['total color'] /
graph.nodes[dst]['pixel count'])
img = io.imread("images/sample.jpg")
labels = segmentation.slic(img, compactness=30, n_segments=400, start_label=1)
g = graph.rag_mean_color(img, labels)
labels2 = graph.merge_hierarchical(labels,
g,
thresh=35,
rag_copy=False,
in_place_merge=True,
merge_func=merge_mean_color,
weight_func=_weight_mean_color)
out = color.label2rgb(labels2, img, kind='avg', bg_label=0)
out = segmentation.mark_boundaries(out, labels2, (0, 0, 0))
io.imshow(out)
io.show()
def merge_boundary(graph, src, dst):
"""Call back called before merging 2 nodes.
In this case we don't need to do any computation here.
"""
pass
img = data.coffee()
edges = filters.sobel(color.rgb2gray(img))
labels = segmentation.slic(img, compactness=30, n_segments=400)
g = graph.rag_boundary(labels, edges)
graph.show_rag(labels, g, img)
plt.title('Initial RAG')
labels2 = graph.merge_hierarchical(labels, g, thresh=0.08, rag_copy=False,
in_place_merge=True,
merge_func=merge_boundary,
weight_func=weight_boundary)
graph.show_rag(labels, g, img)
plt.title('RAG after hierarchical merging')
plt.figure()
out = color.label2rgb(labels2, img, kind='avg')
plt.imshow(out)
plt.title('Final segmentation')
plt.show()
def main():
import argparse
import matplotlib
import matplotlib.pyplot as plt
import os
import time
from skimage.io import imread
from util import closed_mask_roi
from util import masked_slic
parser = argparse.ArgumentParser()
parser.add_argument('img_file')
parser.add_argument('depth_file')
args = parser.parse_args()
img_file = args.img_file
depth_file = args.depth_file
img = imread(img_file)
depth = imread(depth_file)
depth_rgb = gray2rgb(depth)
plt.figure(figsize=(24, 10))
plt.subplot(231)
plt.axis('off')
plt.imshow(depth_rgb)
t_start = time.time()
roi = closed_mask_roi(depth)
roi_labels = masked_slic(img=depth_rgb[roi], mask=depth[roi],
n_segments=20, compactness=30)
print('slic: takes {} [secs]'.format(time.time() - t_start))
labels = np.zeros_like(depth)
labels[roi] = roi_labels
print('n_labels: {}'.format(len(np.unique(labels))))
out = label2rgb(labels, img, bg_label=0)
plt.subplot(232)
plt.axis('off')
plt.imshow(out)
t_start = time.time()
g = rag_solidity(labels, connectivity=2)
print('rag_solidity: takes {} [secs]'.format(time.time() - t_start))
# draw rag: all
out = graph.draw_rag(labels, g, img)
plt.subplot(233)
plt.axis('off')
plt.imshow(out)
# draw rag: good edges
cmap = matplotlib.colors.ListedColormap(['#0000FF', '#FF0000'])
out = graph.draw_rag(labels, g, img, node_color='#00FF00', colormap=cmap,
thresh=1, desaturate=True)
plt.subplot(234)
plt.axis('off')
plt.imshow(out)
t_start = time.time()
merged_labels = graph.merge_hierarchical(
labels, g, thresh=1, rag_copy=False,
in_place_merge=True,
merge_func=_solidity_merge_func, weight_func=_solidity_weight_func)
print('graph.merge_hierarchical: takes {} [secs]'.format(time.time() - t_start))
out = label2rgb(merged_labels, img, bg_label=0)
plt.subplot(235)
plt.axis('off')
plt.imshow(out)
merged_g = rag_solidity(merged_labels)
# draw rag: all
out = graph.draw_rag(merged_labels, merged_g, img)
plt.subplot(236)
plt.axis('off')
plt.imshow(out)
# plt.show()
plt.savefig('{}_solidity_rag_merge.png'.format(time.time()), dpi=100)
def _merge_labels(self, source_image, labels, **kwargs):
def weight_boundary(graph, src, dst, n):
"""
Handle merging of nodes of a region boundary region adjacency graph.
This function computes the `"weight"` and the count `"count"`
attributes of the edge between `n` and the node formed after
merging `src` and `dst`.
Parameters
----------
graph : RAG
The graph under consideration.
src, dst : int
The vertices in `graph` to be merged.
n : int
A neighbor of `src` or `dst` or both.
Returns
-------
data : dict
A dictionary with the "weight" and "count" attributes to be
assigned for the merged node.
"""
default = {"weight": 0.0, "count": 0}
count_src = graph[src].get(n, default).get("count", 0)
count_dst = graph[dst].get(n, default).get("count", 0)
weight_src = graph[src].get(n, default).get("weight", 0)
weight_dst = graph[dst].get(n, default).get("weight", 0)
count = count_src + count_dst
return {
"count":
count,
"weight":
(count_src * weight_src + count_dst * weight_dst) / count,
}
def merge_boundary(graph, src, dst):
"""Call back called before merging 2 nodes.
In this case we don't need to do any computation here.
"""
pass
threshold = self.get_value_of("hierarchy_threshold", 35)
res = None
try:
g = graph.rag_mean_color(source_image, labels)
merged_labels = graph.merge_hierarchical(
labels,
g,
thresh=threshold,
rag_copy=False,
in_place_merge=True,
merge_func=merge_boundary,
weight_func=weight_boundary,
)
merged_labels[merged_labels == -1] = 0
labels2 = ((merged_labels - merged_labels.min()) /
(merged_labels.max() - merged_labels.min()) *
255).astype(np.uint8)
res = cv2.applyColorMap(255 - labels2, DEFAULT_COLOR_MAP)
self._wrapper.store_image(res, f"rag_vis", text_overlay=True)
self.print_segmentation_labels(res, labels2, dbg_suffix="rag")
except Exception as e:
logger.exception(f'FAIL label merging, exception: "{repr(e)}"')
finally:
return res
mag = ndi.generic_gradient_magnitude(prueba1, ndi.sobel,
float) # Float for rag_boundary
#mag *= 255.0 / np.max(mag) # normalize (Q&D)
ragb = graph.rag_boundary(superp, mag, connectivity=1)
# CLUST "CORTAR"
umbralc = 1000
superp_co = graph.cut_threshold(superp, ragb, umbralc)
print('Intermediate Number of SV: ', len(np.unique(superp_co)))
superp_co += 1
# CLUST "UNIR"
umbralb = 3500
ragb2 = graph.rag_boundary(superp_co, mag, connectivity=1)
superp_un = graph.merge_hierarchical(superp_co,
ragb2,
umbralb,
in_place_merge=True,
rag_copy=False,
merge_func=ji.merge_boundary,
weight_func=ji.weight_boundary)
superp_un += 1
directorio = Path(
"C:/Users/Juan Ignacio/Documents/Movistar Cloud/TFM/Prueba_Feima_multiframe/"
)
print('Final Number of SV: ', len(np.unique(superp_un)))
print('Parametros:', '\n', ' Num. Segmentos:', numSegments, '\n',
' umbral corte: ', umbralc, '\n', ' umbral union: ', umbralb)
#nom1 = "ragb_co.tif"
ruta_co = directorio
ji.save_imagen(prueba1, superp_un, ruta_co)
