def semantic2whole(self, mask_path=None, seg_path=None):
s_m, s_n = self.result.shape
self.result = hole_fill(
remove_small_objects(self.result.astype(np.bool), 200))
rows = np.where(self.result)[0]
columns = np.where(self.result)[1]
wsi = OpenSlide(os.path.join(self.img_WSI_dir, self.name))
big = np.zeros(
(s_m * self.stride / self.scale_ + 256 / self.scale_,
s_n * self.stride / self.scale_ + 256 / self.scale_, 3))
# get the original area
for idx, i in enumerate(rows):
print idx, '/', len(rows)
img = np.array(
wsi.read_region(
(columns[idx] * self.stride * 2**self.using_level,
i * self.stride * 2**self.using_level), self.using_level,
(self.stride, self.stride)))[..., 0:-1]
resize_img = cv2.resize(
img, (self.stride / self.scale_, self.stride / self.scale_))
big[i * self.stride / self.scale_:(i + 1) * self.stride /
self.scale_,
columns[idx] * self.stride / self.scale_:(columns[idx] + 1) *
self.stride / self.scale_, :] = resize_img
# get rid of other area
contours = measure.find_contours(self.result, 0.5)
for i in xrange(len(contours)):
for j in range(len(contours[i])):
y = int(contours[i][j, 0] * self.stride * 2**self.using_level)
x = int(contours[i][j, 1] * self.stride * 2**self.using_level)
img = io.imread(
os.path.join(seg_path,
str(x) + '_' + str(y) + '_seg.png'))
resize_img = cv2.resize(img,
(256 / self.scale_, 256 / self.scale_))
y_ = y / (self.scale_ * 2**self.using_level)
x_ = x / (self.scale_ * 2**self.using_level)
big[y_:y_ + 256 / self.scale_,
x_:x_ + 256 / self.scale_, :][resize_img == 0] = 0
io.imsave(os.path.join(mask_path, 'tif1.png'),
big.astype(np.float) / 255)
print 'semantic2whole done'
def process(mask, min_size=100):
mask = label(
hole_fill(remove_small_objects(mask.astype(np.bool), min_size)))
mask = regionprops(mask)
mask_coord = [i.centroid for i in mask]
return mask_coord
def semantic2xml(label, doc, Annotations):
stride = 36 # set stride 36
using_level = 0 # 0: max level; 1
scale_ = 4
# mrophology operation
small_like = np.zeros_like(thumbnail_mask)
small_like[np.where(thumbnail_mask == label)] = 1
shape = (3, 3)
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, shape)
image_open = cv2.morphologyEx(np.array(small_like), cv2.MORPH_OPEN, kernel)
image_close = cv2.morphologyEx(image_open, cv2.MORPH_CLOSE, kernel)
image_open = cv2.morphologyEx(image_close, cv2.MORPH_OPEN, kernel)
result = cv2.morphologyEx(image_open, cv2.MORPH_CLOSE, kernel)
result = cv2.morphologyEx(
result, cv2.MORPH_DILATE,
cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 5)))
s_m, s_n = result.shape
result = remove_small_objects(result.astype(np.bool), 200)
result = hole_fill(result)
#
rows = np.where(result)[0]
columns = np.where(result)[1]
wsi = OpenSlide(os.path.join(img_WSI_dir, 'test.svs'))
big = np.zeros((s_m * stride / scale_ + 256 / scale_,
s_n * stride / scale_ + 256 / scale_))
# get the original area
for idx, i in enumerate(rows):
print idx, '/', len(rows)
img = np.array(
wsi.read_region((columns[idx] * stride * 2**using_level,
i * stride * 2**using_level), using_level,
(stride, stride)))[..., 0:-1]
resize_img = cv2.resize(img, (stride / scale_, stride / scale_))
big[i * stride / scale_:(i + 1) * stride / scale_,
columns[idx] * stride / scale_:(columns[idx] + 1) * stride /
scale_] = np.ones(resize_img.shape[0:-1])
# get rid of other area
segpath = '/media/zzr/Data/skin_xml/semantic_result/test.svs'
contours = measure.find_contours(result, 0.5)
for i in xrange(len(contours)):
for j in range(len(contours[i])):
y = int(contours[i][j, 0] * stride * 2**using_level)
x = int(contours[i][j, 1] * stride * 2**using_level)
img = io.imread(
os.path.join(segpath,
str(x) + '_' + str(y) + '_seg.png'))
resize_img = cv2.resize(img, (256 / scale_, 256 / scale_))
y_ = y / (scale_ * 2**using_level)
x_ = x / (scale_ * 2**using_level)
big[y_:y_ + 256 / scale_,
x_:x_ + 256 / scale_][resize_img == 0] = 0
# get xml
big = remove_small_objects(big.astype(np.bool), 1000)
big = hole_fill(big)
contours = measure.find_contours(big, 0.5)
for i in xrange(len(contours)):
Annotation = doc.createElement("Annotation")
Annotation.setAttribute("Name", 'Semantic_' + str(i))
Annotation.setAttribute("Type", "Polygon")
Annotation.setAttribute("PartOfGroup", "None")
Annotation.setAttribute("Color", "#aa5500")
Annotations.appendChild(Annotation)
Coordinates = doc.createElement("Coordinates")
Annotation.appendChild(Coordinates)
for j in range(len(contours[i])):
y = int(contours[i][j, 0] * scale_) * 2**using_level
x = int(contours[i][j, 1] * scale_) * 2**using_level
Coordinate1 = doc.createElement("Coordinate")
Coordinate1.setAttribute("Order", str(j))
Coordinate1.setAttribute("Y", str(y))
Coordinate1.setAttribute("X", str(x))
Coordinates.appendChild(Coordinate1)
return doc
def semantic2xml(self, resultFile=None, seg_path=None):
"""
:param resultFile:
:param seg_path: the result of segnet
:return:
"""
s_m, s_n = self.result.shape
self.result = hole_fill(
remove_small_objects(self.result.astype(np.bool), 200))
rows = np.where(self.result)[0]
columns = np.where(self.result)[1]
big = np.zeros((s_m * self.stride / self.scale_ + 256 / self.scale_,
s_n * self.stride / self.scale_ + 256 / self.scale_))
# get the original area
for idx, i in enumerate(rows):
# print idx, '/', len(rows)
big[i * self.stride / self.scale_:(i + 1) * self.stride /
self.scale_,
columns[idx] * self.stride / self.scale_:(columns[idx] + 1) *
self.stride / self.scale_] = np.ones(
[self.stride / self.scale_, self.stride / self.scale_])
# get rid of other area
contours = measure.find_contours(self.result, 0.5)
for i in xrange(len(contours)):
for j in range(len(contours[i])):
y = int(contours[i][j, 0] * self.stride * 2**self.using_level)
x = int(contours[i][j, 1] * self.stride * 2**self.using_level)
img = io.imread(
os.path.join(seg_path,
str(x) + '_' + str(y) + '_seg.png'))
resize_img = cv2.resize(img,
(256 / self.scale_, 256 / self.scale_))
y_ = y / (self.scale_ * 2**self.using_level)
x_ = x / (self.scale_ * 2**self.using_level)
big[y_:y_ + 256 / self.scale_,
x_:x_ + 256 / self.scale_][resize_img == 0] = 0
# get xml
big = hole_fill(remove_small_objects(big.astype(np.bool), 1000))
# io.imsave('/media/zzr/Data/skin_xml/mask_result/tif/tif_3.png', big.astype(np.float)) # binary map
contours = measure.find_contours(big, 0.5)
# np.save('contours.npy', contours)
if resultFile:
doc = minidom.Document()
ASAP_Annotation = doc.createElement("ASAP_Annotations")
doc.appendChild(ASAP_Annotation)
Annotations = doc.createElement("Annotations")
ASAP_Annotation.appendChild(Annotations)
for i in xrange(len(contours)):
Annotation = doc.createElement("Annotation")
Annotation.setAttribute("Name", "_" + str(i))
Annotation.setAttribute("Type", "Polygon")
Annotation.setAttribute("PartOfGroup", "None")
Annotation.setAttribute("Color", "#F4FA58")
Annotations.appendChild(Annotation)
Coordinates = doc.createElement("Coordinates")
Annotation.appendChild(Coordinates)
for j in range(len(contours[i])):
y = int(
contours[i][j, 0] * self.scale_) * 2**self.using_level
x = int(
contours[i][j, 1] * self.scale_) * 2**self.using_level
Coordinate1 = doc.createElement("Coordinate")
Coordinate1.setAttribute("Order", str(j))
Coordinate1.setAttribute("Y", str(y))
Coordinate1.setAttribute("X", str(x))
Coordinates.appendChild(Coordinate1)
f = file(resultFile, "w")
doc.writexml(f)
f.close()
print 'semantic2xml done'
return big