def update(self,frame,recent_pupil_positions,events):
falloff = self.falloff.value
img = frame.img
img_shape = img.shape[:-1][::-1]#width,height
norm_gaze = [ng['norm_gaze'] for ng in recent_pupil_positions if ng['norm_gaze'] is not None]
screen_gaze = [denormalize(ng,img_shape,flip_y=True) for ng in norm_gaze]
overlay = np.ones(img.shape[:-1],dtype=img.dtype)
# draw recent gaze postions as black dots on an overlay image.
for gaze_point in screen_gaze:
try:
overlay[int(gaze_point[1]),int(gaze_point[0])] = 0
except:
pass
out = cv2.distanceTransform(overlay,cv2.cv.CV_DIST_L2, 5)
# fix for opencv binding incositency
if type(out)==tuple:
out = out[0]
overlay = 1/(out/falloff+1)
img *= cv2.cvtColor(overlay,cv2.COLOR_GRAY2RGB)
def discretPoints2ContourGrad(points, contourImg):
distField = cv2.distanceTransform(contourImg, cv2.DIST_L2, 0)
dxKernel = np.array([
[ 0, 0, 0],
[-1, 0, 1],
[ 0, 0, 0]
], dtype=np.float32) / 2.0
dyKernel = np.array([
[0, -1, 0],
[0, 0, 0],
[0, 1, 0]
], dtype=np.float32) / 2.0
#dxKernel = np.array([[]], dtype=np.float32)
dx = cv2.filter2D(distField, -1, dxKernel)
dy = cv2.filter2D(distField, -1, dyKernel)
res = np.zeros_like(points)
for i, p in enumerate(points):
res[i, 0] = takePixelInterpolated(dx, p)
res[i, 1] = takePixelInterpolated(dy, p)
print('p>', points)
print('g>', res)
return res
def segment_on_dt(img):
#http://stackoverflow.com/questions/11294859/how-to-define-the-markers-for-watershed-in-opencv
img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, img_bin = cv2.threshold(img_gray, 0, 255,cv2.THRESH_OTSU)
img_bin = cv2.morphologyEx(img_bin, cv2.MORPH_OPEN,numpy.ones((3, 3), dtype=int))
border = cv2.dilate(img_bin, None, iterations=5)
border = border - cv2.erode(border, None)
dt = cv2.distanceTransform(img_bin, 2, 3)
dt = ((dt - dt.min()) / (dt.max() - dt.min()) * 255).astype(numpy.uint8)
_, dt = cv2.threshold(dt, 180, 255, cv2.THRESH_BINARY)
lbl, ncc = label(dt)
lbl = lbl * (255/ncc)
# Completing the markers now.
lbl[border == 255] = 255
lbl = lbl.astype(numpy.int32)
cv2.watershed(img, lbl)
lbl[lbl == -1] = 0
lbl = lbl.astype(numpy.uint8)
result = 255 - lbl
result[result != 255] = 0
result = cv2.dilate(result, None)
img[result == 255] = (0, 0, 255)
return img
def distance_transform(bin_img, distance_type, mask_size):
"""Creates an image where for each object pixel, a number is assigned that corresponds to the distance to the
nearest background pixel.
Inputs:
img = Binary image data
distance_type = Type of distance. It can be CV_DIST_L1, CV_DIST_L2 , or CV_DIST_C which are 1, 2 and 3,
respectively.
mask_size = Size of the distance transform mask. It can be 3, 5, or CV_DIST_MASK_PRECISE (the latter option
is only supported by the first function). In case of the CV_DIST_L1 or CV_DIST_C distance type,
the parameter is forced to 3 because a 3 by 3 mask gives the same result as 5 by 5 or any larger
aperture.
Returns:
norm_image = grayscale distance-transformed image normalized between [0, 1]
:param bin_img: numpy.ndarray
:param distance_type: int
:param mask_size: int
:return norm_image: numpy.ndarray
"""
params.device += 1
dist = cv2.distanceTransform(src=bin_img, distanceType=distance_type, maskSize=mask_size)
norm_image = cv2.normalize(src=dist, dst=dist, alpha=0, beta=1, norm_type=cv2.NORM_MINMAX, dtype=cv2.CV_32F)
if params.debug == 'print':
print_image(norm_image, os.path.join(params.debug, str(params.device) + '_distance_transform.png'))
elif params.debug == 'plot':
plot_image(norm_image, cmap='gray')
return norm_image
def proc_border(img):
# Compute horizontal Sobel gradient
sobelx = cv2.Sobel(img, cv2.CV_64F, 1, 0)
sobelx = np.absolute(sobelx)
# Compute vertical Sobel gradient
sobely = cv2.Sobel(img, cv2.CV_64F, 0, 1)
sobely = np.absolute(sobely)
# Compute gradient magnitude image
mag = np.sqrt(sobelx ** 2 + sobely ** 2)
mag = np.uint8(mag)
del(img, sobelx, sobely)
# Threhold the gradient magnitude image
mag = cv2.threshold(mag, 2, 1, cv2.THRESH_BINARY)[1]
# Fill holes of the thresholded image
contour, hier = cv2.findContours(mag, cv2.RETR_CCOMP,
cv2.CHAIN_APPROX_SIMPLE)
for cnt in contour:
cv2.drawContours(mag, [cnt], 0, 255, -1)
# Distance transform
mag = cv2.distanceTransform(mag, cv2.cv.CV_DIST_L1,
cv2.cv.CV_DIST_MASK_PRECISE)
return mag
def swtChenAltered(img):
'takes distance transform and gives swt'
dT = cv2.distanceTransform(img,cv2.DIST_L2,5)
dT = np.uint8(np.absolute(dT))
_,dist_threshold = cv2.threshold(dT,1,255,cv2.THRESH_BINARY)
diff = img - dist_threshold
dist = dT.copy()
dist[diff==255] = 1
lookUp = {}
height,width = dist.shape
for i in range(height):
for j in range(width):
if dist.item(i,j)>0:
xlist = [j-1,j,j+1]
ylist = [i-1,i,i+1]
point = (i,j)
for y in ylist:
if (y>0 and y<height):
for x in xlist:
if (x>0 and x<width):
if dist.item(y,x)>dist.item(point):
point = (y,x)
lookUp[(i,j)] = point
swt = dist.copy()
for i in range(height):
for j in range(width):
if dist.item(i,j)>0:
point = (i,j)
while point != lookUp[point]:
point = lookUp[point]
val = swt.item(point)
swt.itemset(i,j,val)
return swt
def water(img, thresh):
kernel = np.ones((3,3),np.uint8)
opening = cv2.morphologyEx(thresh,cv2.MORPH_OPEN,kernel, iterations = 2)
# sure background area
sure_bg = cv2.dilate(opening,kernel,iterations=3)
# Finding sure foreground area
dist_transform = cv2.distanceTransform(opening,2,5)
ret, sure_fg = cv2.threshold(dist_transform,0.7*dist_transform.max(),255,0)
# Finding unknown region
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_bg,sure_fg)
# Marker labelling
ret, markers = cv2.connectedComponents(sure_fg)
# Add one to all labels so that sure background is not 0, but 1
markers += 1
# Now, mark the region of unknown with zero
markers[unknown==255] = 0
markers = cv2.watershed(img,markers)
img[markers == -1] = [255,0,0]
return sure_fg, sure_bg, markers, img
def watershed(self,img):
'''
watershedで領域分割を行う
args : ->
dst : ->
param: ->
'''
gimg = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(gimg,0,255,cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
# noise removal
kernel = np.ones((3,3),np.uint8)
opening = cv2.morphologyEx(thresh,cv2.MORPH_OPEN,kernel, iterations = 2)
# sure background area
sure_bg = cv2.dilate(opening,kernel,iterations=3)
# Finding sure foreground area
dist_transform = cv2.distanceTransform(opening,cv.CV_DIST_L2,5)
ret, sure_fg = cv2.threshold(dist_transform,0.7*dist_transform.max(),255,0)
# Finding unknown region
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_bg,sure_fg)
# Marker labelling
ret, markers = cv2.connectedComponents(sure_fg)
# Add one to all labels so that sure background is not 0, but 1
markers = markers+1
# Now, mark the region of unknown with zero
markers[unknown==255] = 0
def UpdateModelFromMask(self, mask, img, hsv):
self.avgRGB = cv.mean(img, mask)[0:3]
self.avgHSV = cv.mean(hsv, mask)[0:3]
distMap = cv.distanceTransform(1-mask, cv.cv.CV_DIST_L2, 5)[0]
self.lineProbabilityMap = (1.0/(1.0+0.1*distMap))
print self.avgRGB
print self.avgHSV
def segment(template, actual): ret3,th3 = cv2.threshold(template,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU) # noise removal kernel = np.ones((3,3),np.uint8) opening = cv2.morphologyEx(th3,cv2.MORPH_OPEN,kernel, iterations = 2) # sure background area sure_bg = cv2.dilate(opening,kernel,iterations=3) # Finding sure foreground area dist_transform = cv2.distanceTransform(opening,cv2.cv.CV_DIST_L2,5) ret, sure_fg = cv2.threshold(dist_transform,0.7*dist_transform.max(),255,0) # Finding unknown region sure_fg = np.uint8(sure_fg) unknown = cv2.subtract(sure_bg,sure_fg) # Marker labelling ret, markers = cv2.connectedComponents(sure_fg) # Add one to all labels so that sure background is not 0, but 1 markers = markers+1 # Now, mark the region of unknown with zero markers[unknown==255] = 0 markers = cv2.watershed(img,markers) img[markers == -1] = [255,0,0] return img
def segment_on_dt(a, img):
'''
find foreground
'''
fg=cv2.dilate(img,None, iterations=5)
fg=fg-cv2.erode(fg, None)
dt=cv2.distanceTransform(img, 2, 3)
dt=((dt-dt.min())/(dt.max()-dt.min())*255).astype(np.uint8)
_, dt=cv2.threshold(dt, 0, 255, cv2.THRESH_BINARY)
lbl, ncc=label(dt)
lbl=lbl*(255/ncc)
'''
Completing the markers now
'''
lbl[fg==255]=255
lbl=lbl.astype(np.int32)
cv2.watershed(a, lbl)
lbl[lbl==-1]=0
lbl=lbl.astype(np.uint8)
return 255-lbl
def check_thickness(self, stones: np.ndarray, rs=0, re=gsize, cs=0, ce=gsize, **kwargs):
""" Check that the provided stones array doesn't contain "big chunks" that wouldn't make sense in a game of Go.
Args:
stones: ndarray
The newly found stones, in other words the result to check.
rs: int - inclusive
re: int - exclusive
Row start and end indexes. Can be used to restrain check to a subregion.
cs: int - inclusive
ce: int - exclusive
Column start and end indexes. Can be used to restrain check to a subregion.
kwargs:
Allowing for keyword args enables multiple check methods to be called indifferently. See self.verify().
Return status: int
-1, 0, or 1 if the check is respectively refused, undetermined, or passed.
"""
for color in (B, W):
avatar = np.vectorize(lambda x: 1 if x is color else 0)(stones[rs:re, cs:ce].flatten())
# diagonal moves cost as much as side moves
dist = cv2.distanceTransform(avatar.reshape((re-rs, ce-cs)).astype(np.uint8), cv2.DIST_C, 3)
if 2 < np.max(dist):
# a stone surrounded by a 2-stones-thick wall of its own color is most likely not Go
return -1
return 0
def get_centerline_optimized(self, alpha=1e3, beta=1e6, gamma=0.01,
spacing=20, max_iterations=1000,
endpoints=None):
""" determines the center line of the polygon using an active contour
algorithm """
# use an active contour algorithm to find centerline
ac = ActiveContour(blur_radius=1, alpha=alpha, beta=beta,
gamma=gamma, closed_loop=False)
ac.max_iterations = max_iterations
# set the potential from the distance map
mask, offset = self.get_mask(1, ret_offset=True)
potential = cv2.distanceTransform(mask, cv2.DIST_L2, 5)
ac.set_potential(potential)
# initialize the centerline from the estimate
points = self.get_centerline_estimate(endpoints)
points = curves.make_curve_equidistant(points, spacing=spacing)
points = curves.translate_points(points, -offset[0], -offset[1])
# anchor the end points
anchor = np.zeros(len(points), np.bool)
anchor[0] = anchor[-1] = True
# find the best contour
points = ac.find_contour(points, anchor, anchor)
points = curves.make_curve_equidistant(points, spacing=spacing)
return curves.translate_points(points, *offset)
def centerline(self):
""" determine the center line of the tail """
mask, offset = self.mask
dist_map = cv2.distanceTransform(mask, cv2.DIST_L2, 5)
# setup active contour algorithm
ac = ActiveContour(blur_radius=self.centerline_blur_radius,
closed_loop=False,
alpha=0, #< line length is constraint by beta
beta=self.centerline_bending_stiffness,
gamma=self.centerline_adaptation_rate)
ac.max_iterations = self.centerline_max_iterations
ac.set_potential(dist_map)
# find centerline starting from the ventral_side
points = curves.translate_points(self.ventral_side,
-offset[0], -offset[1])
spacing = self.centerline_spacing
points = curves.make_curve_equidistant(points, spacing=spacing)
# use the active contour algorithm
points = ac.find_contour(points)
points = curves.make_curve_equidistant(points, spacing=spacing)
# translate points back into global coordinate system
points = curves.translate_points(points, offset[0], offset[1])
# orient the centerline such that it starts at the posterior end
dist1 = curves.point_distance(points[0], self.endpoints[0])
dist2 = curves.point_distance(points[-1], self.endpoints[0])
if dist1 > dist2:
points = points[::-1]
return points
def transDistancia(img):
#aplicamos un filtro gausssiano
#imgBlur = cv2.GaussianBlur(img,(5,5),0)
if len(img.shape) == 3:
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
print "convirtio la imagen a grises"
#binarizacion de otsu
ret, thresh = cv2.threshold(img,0,255,cv2.THRESH_BINARY+cv2.THRESH_OTSU)
# removiendo ruido
kernel = np.ones((3,3),np.uint8)
opening = cv2.morphologyEx(thresh,cv2.MORPH_OPEN,kernel, iterations = 2)
#area segura
sure_bg = cv2.dilate(opening,kernel,iterations=3)
#transformada de la distancia
dist_transform = cv2.distanceTransform(opening,cv2.cv.CV_DIST_L2,3)
imshow(dist_transform)
show()
img = np.uint8(dist_transform)
print "Transformada de distancia"
return img,sure_bg
def segment_on_dt(a, img, gray):
border = cv2.dilate(img, None, iterations=5)
border = border - cv2.erode(border, None)
dt = cv2.distanceTransform(img,cv2.DIST_L2,5)
plt.subplot(3,3,4)
plt.imshow(dt),plt.title('dt'),plt.xticks([]), plt.yticks([])
dt = ((dt - dt.min()) / (dt.max() - dt.min()) * 255).astype(np.uint8)
_, dt2 = cv2.threshold(dt, 0, 255, cv2.THRESH_BINARY)
dt2 = cv2.erode(dt2, None, iterations=2)
# dt2 = cv2.adaptiveThreshold(dt, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 5, 2)
# dt1 = peak_local_max(gray, indices=False, min_distance=10, labels=img, threshold_abs=5)
# dt2 = peak_local_max(dt, indices=False, min_distance=5, labels=img, threshold_abs=0)
lbl, ncc = label(dt2)
plt.subplot(3,3,5)
plt.imshow(dt2),plt.title('localMax'),plt.xticks([]), plt.yticks([])
# plt.subplot(3,3,6)
# plt.imshow(ncc),plt.title('ncc'),plt.xticks([]), plt.yticks([])
lbl = lbl * (255/ncc)
# Completing the markers now.
lbl[border == 255] = 255
lbl = lbl.astype(np.int32)
cv2.watershed(a, lbl)
lbl[lbl == -1] = 0
lbl = lbl.astype(np.uint8)
plt.subplot(3,3,6)
plt.imshow(lbl),plt.title('lbl_out'),plt.xticks([]), plt.yticks([])
return 255 - lbl
def watershed(img, thresh):
# noise removal
kernel = np.ones((3,3), np.uint8)
opening = cv2.morphologyEx(thresh,cv2.MORPH_OPEN,kernel, iterations = 4)
# sure background area
sure_bg = cv2.dilate(opening,kernel,iterations=3)
#sure_bg = cv2.morphologyEx(sure_bg, cv2.MORPH_TOPHAT, kernel)
# Finding sure foreground area
dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 5)
ret, sure_fg = cv2.threshold(dist_transform,0.7*dist_transform.max(),255,0)
# Finding unknown region
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_bg,sure_fg)
# Marker labelling
ret, markers = cv2.connectedComponents(sure_fg)
# Add one to all labels so that sure background is not 0, but 1
markers = markers+1
# Now, mark the region of unknown with zero
markers[unknown==255] = 0
'''
imgray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
imgray = cv2.GaussianBlur(imgray, (5, 5), 0)
img = cv2.Canny(imgray,200,500)
'''
markers = cv2.watershed(img,markers)
img[markers == -1] = [255,0,0]
return sure_bg, sure_fg
def recent_events(self, events):
frame = events.get("frame")
if not frame:
return
falloff = self.falloff
img = frame.img
pts = [
denormalize(pt["norm_pos"], frame.img.shape[:-1][::-1], flip_y=True)
for pt in events.get("gaze", [])
if pt["confidence"] >= self.g_pool.min_data_confidence
]
overlay = np.ones(img.shape[:-1], dtype=img.dtype)
# draw recent gaze postions as black dots on an overlay image.
for gaze_point in pts:
try:
overlay[int(gaze_point[1]), int(gaze_point[0])] = 0
except:
pass
out = cv2.distanceTransform(overlay, cv2.DIST_L2, 5)
# fix for opencv binding inconsitency
if type(out) == tuple:
out = out[0]
overlay = 1 / (out / falloff + 1)
img[:] = np.multiply(
img, cv2.cvtColor(overlay, cv2.COLOR_GRAY2RGB), casting="unsafe"
)
def mask2dtAndEdgelsCount(mask):
maskCopy = mask.copy()
contours, hierarchy = cv2.findContours(maskCopy, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)
contoursImage = np.zeros(mask.shape, np.uint8)
cv2.drawContours(contoursImage, contours, -1, (255))
edgelsCount = len(np.nonzero(contoursImage == 255)[0])
dt = cv2.distanceTransform(~contoursImage, cv2.cv.CV_DIST_L1, cv2.cv.CV_DIST_MASK_PRECISE)
return [dt, edgelsCount]
def __segment(self, gray):
# Substract BG to current frame
mostChanges = np.abs(sum(self.W-self.BG))
# Normalize max-min values
mostChanges = (mostChanges.clip(0, max=1)*255).astype('uint8')
self.__recognition = mostChanges.copy()
# Apply a threshold
_,mostChanges = cv2.threshold(mostChanges, 150, 255, cv2.THRESH_BINARY)
mostChanges = cv2.medianBlur(mostChanges, 9)
self.__mostChanges = mostChanges.copy()
# Apply median filter and multiply with the current scene
mask = cv2.medianBlur(mostChanges, 3)
mask = np.multiply(gray, mask)
mask = (mask*255).astype('uint8')
# Apply distance filter to minimize possible collisions
mask = cv2.distanceTransform(mask, cv2.cv.CV_DIST_L1, 3)
# Normalize between 0 and 255
mask = cv2.normalize(mask, mask, 0, 255, cv2.NORM_MINMAX).astype('uint8')
# Reduce noise
mask = cv2.GaussianBlur(mask, (-3,-3), 0.5)
# Apply threshold again to mask out real low values
# (Almost black zones, thus possible unions between objects)
_,mask = cv2.threshold(mask, 10, 255, cv2.THRESH_BINARY)
self.__mask = mask.copy()
contours,_ = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
objs = []
for c in contours:
leftmost = c[:,:,0].min()
rightmost = c[:,:,0].max()
topmost = c[:,:,1].min()
bottommost = c[:,:,1].max()
area = ((rightmost - leftmost)*(bottommost-topmost))
if area < self.min_area:
continue
objs.append((leftmost, rightmost, topmost, bottommost))
found = 1
while found > 0:
found = 0
for a in objs:
for b in objs:
if a != b:
if self.__contains(a, b):
objs.remove(b)
found = 1
break
return objs
def findSeeds(img, s_kernel):
dist = cv.distanceTransform(img, cv.DIST_L2, 3) # distanceTransform(bw, dist, CV_DIST_L2, 3);
cv.normalize(dist, dist, 0, 1., cv.NORM_MINMAX);
dist = dist * 255
dist = np.uint8(dist)
# cv.imshow('seem0', dist)
kernel = cv.getStructuringElement(cv.MORPH_ELLIPSE, (s_kernel * 2, s_kernel * 2))
seed = cv.morphologyEx(dist, cv.MORPH_ERODE, kernel, iterations=4)
seed = cv.morphologyEx(seed, cv.MORPH_CLOSE, kernel, iterations=4)
return seed
def calCenterFromContour(pointset, image):
mask = ac_mask(pointset[:, :2].transpose(), image.transpose().shape).transpose()
dist = cv2.distanceTransform(mask, cv.CV_DIST_L2, 3)
value = npy.max(dist)
indx, indy = npy.where(dist == npy.max(dist))
indx = npy.cast['float32'](indx)
indy = npy.cast['float32'](indy)
center = npy.array([[npy.mean(indy), npy.mean(indx)]])
return center
def segment_on_dt(img):
dt = cv2.distanceTransform(img, 2, 3) # L2 norm, 3x3 mask
dt = ((dt - dt.min()) / (dt.max() - dt.min()) * 255).astype(numpy.uint8)
dt = cv2.threshold(dt, 100, 255, cv2.THRESH_BINARY)[1]
lbl, ncc = label(dt)
lbl[img == 0] = lbl.max() + 1
lbl = lbl.astype(numpy.int32)
cv2.watershed(cv2.cvtColor(img, cv2.COLOR_GRAY2BGR), lbl)
lbl[lbl == -1] = 0
return lbl
def update(dummy=None):
global need_update
need_update = False
thrs = cv2.getTrackbarPos('threshold', 'distrans')
mark = cv2.Canny(img, thrs, 3*thrs)
dist, labels = cv2.distanceTransform(~mark, cv.CV_DIST_L2, 5)
if voronoi:
vis = cm[np.uint8(labels)]
else:
vis = cm[np.uint8(dist*2)]
vis[mark != 0] = 255
cv2.imshow('distrans', vis)
def fg_extract(img):
# Distance transform. Output 32-bit float image.
dist_transform = cv2.distanceTransform(img, cv2.cv.CV_DIST_L1, maskSize=5)
print("max of dist_transform is " + str(dist_transform.max()))
non_zero_d = dist_transform[dist_transform != 0];
mean_dist = np.mean(non_zero_d)
print("mean of non_zero_d is " + str(mean_dist))
ostuval, fg_img = cv2.threshold(dist_transform, mean_dist, 255, cv2.THRESH_BINARY)
#struct = np.ones((5, 5), dtype=np.uint8)
#struct[1:3, 1:3] = 0
#fg_img = cv2.erode(img, struct)
return fg_img, mean_dist
def distance_transform_diameter(edge_trace, segmented):
"""
(dev comments from nefi1)
my cv2 lacks cv2.DIST_L2, it seems to have the value 2 though, so I use
that, same for MASK_PRECISE
<python3 cv2.DIST_L2 equals to 2>
"""
dt = cv2.distanceTransform(segmented, 2, 0)
edge_pixels = np.nonzero(edge_trace)
diameters = defaultdict(list)
for label, diam in zip(edge_trace[edge_pixels], 2.0 * dt[edge_pixels]):
diameters[label].append(diam)
return diameters
def detect_black_areas(filename):
m1 = cv2.imread('images/brick.jpg')
m2 = cv2.cvtColor(m1, cv2.cv.CV_BGR2GRAY)
m3 = 255 - cv2.threshold(m2, 0, 255, cv2.THRESH_OTSU)[1]
m4 = cv2.distanceTransform(m3, cv2.cv.CV_DIST_L2, 3)
m5 = cv2.normalize(m4, alpha=0., beta=1., norm_type=cv2.NORM_MINMAX)
m6 = cv2.threshold(m5, .8, 1., cv2.THRESH_BINARY)[1]
m7 = cv2.dilate(m6, np.ones((3,3), np.uint8), iterations=7)
cnt = cv2.findContours(m7.astype(np.uint8),
cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)[0]
cv2.drawContours(m1, cnt, -1, (0,0,255), 2)
cv2.imshow("image", m1)
cv2.waitKey()
def _divide(self):
cv2.distanceTransform(self.biofilm, cv.CV_DIST_L2, 5,
dst=self.distances)
rows, columns = map(list, self.dividing.nonzero())
displacements = [0 for _ in rows]
while rows:
cellIndex = np.array((rows.pop(), columns.pop()))
displacement = displacements.pop()
# this also takes care of the situation where updating the biofilm
# in-place breaks the distance map implmentation.
if displacement > 10: continue
foundEmpty = False
for neighbor in _randomNeighbors():
index = tuple(cellIndex + neighbor)
if not self._validIndex(index): continue
if self.biofilm[index] == EMPTY:
foundEmpty = True
self.biofilm[index] = ALIVE
if not foundEmpty:
minDistance = 999999
minNeighbor = None
for neighbor in _neighbors:
index = tuple(cellIndex + neighbor)
if not self._validIndex(index): continue
if self.distances[index] <= minDistance:
minDistance = self.distances[index]
minNeighbor = index
rows.append(minNeighbor[0])
columns.append(minNeighbor[1])
displacements.append(displacement + 1)
def frames(self):
for frame in self.inputf.frames():
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
# invert so that background=black, object=white
thresh = (255-thresh)
cv2.imwrite("is00_thresh.jpg", thresh)
# noise removal
kernel = np.ones((3, 3), np.uint8)
opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=2)
cv2.imwrite("is01_opening.jpg", opening)
# now white covers foreground but has foreground false positives
# may be sufficient if we don't care about separating touching objects;
# also I think there will be cases like the spider where the threshould level
# will lose the joints of its legs
# black cover background with background false positives
sure_bg = cv2.dilate(opening, kernel, iterations=3)
cv2.imwrite("is02a_sure_bg.jpg", sure_bg)
# Finding sure foreground area
dist_transform = cv2.distanceTransform(opening,cv2.DIST_L2,5)
ret, sure_fg = cv2.threshold(dist_transform,self.dt_param*dist_transform.max(),255,0)
cv2.imwrite("is02b_sure_fg.jpg", sure_fg)
# unkown region, i.e. part of image we are unsure about
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_bg,sure_fg)
cv2.imwrite("is03_unknown.jpg", unknown)
# Marker labelling
ret, markers = cv2.connectedComponents(sure_fg)
# Add one to all labels so that sure background is not 0, but 1
markers = markers+1
cv2.imwrite("is04_markers.jpg", markers)
# Now, mark the region of unknown with zero
markers[unknown==255] = 0
cv2.imwrite("is05_markers_u0.jpg", markers)
markers = cv2.watershed(frame, markers)
frame[markers == -1] = [255,0,0]
cv2.imwrite("is07_final.jpg", frame)
break
yield None
def get_mask_patch(img_orig, mask):
dist = cv2.distanceTransform(mask, cv.CV_DIST_L1, 3)
max_val = np.amax(dist)
x_mask, y_mask = mask.shape
x_orig, y_orig = img_orig.shape
x_coeff = np.divide(x_orig, x_mask)
y_coeff = np.divide(y_orig, y_mask)
# force the patch to be max 50 x 50
# (in the downsampled image)
if max_val >= 100:
delta = 25
else:
delta = max_val / 4
x, y = np.unravel_index(np.argmax(dist), mask.shape)
return img_orig[(x-delta) * x_coeff:(x+delta) * x_coeff,
(y-delta) * y_coeff:(y+delta) * y_coeff]
def detect(a):
# convert data into image format
img = cv2.imdecode(a, cv2.IMREAD_COLOR) # like cv2.imread()
# function to snip the image of fly glue trap
#img = scanProcess(img)
img = cropped(img)
# virtually save the image of new cut-out image
# convert the image format into streaming data
ret, imgCrop = cv2.imencode('.jpg', img) # like cv2.imwrite()
# cnvert into image format for later use
imgCrop = cv2.imdecode(imgCrop, cv2.IMREAD_COLOR)
#perform grayscale conversion and thresholding
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(gray, 0, 255,
cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
thresh = gridline(gray, thresh)
#Setting up size of objects(number of pixel) to segment
ret, markers, stats, centroids = cv2.connectedComponentsWithStats(thresh)
size = stats[1:, -1]
ret = ret - 1
#declaration for answer image
imgB = np.zeros((markers.shape))
#to keep only the component in the image that is above min size
for i in range(0, ret):
if (size[i] > 350):
imgB[markers == i + 1] = 255
#convert image data type to 8 bit unsigned interger
imgB = np.uint8(imgB)
img = cover350(imgB, img)
# noise removal
kernel = np.ones((2, 3), np.uint8)
kernel2 = np.ones((1, 2), np.uint8)
#opening - erosion then dilate
opening = cv2.morphologyEx(imgB, cv2.MORPH_OPEN, kernel, iterations=2)
#closing - dilate then erosion
closing = cv2.morphologyEx(opening, cv2.MORPH_CLOSE, kernel2, iterations=1)
#sure background area
sureBack = cv2.dilate(imgB, kernel, iterations=1)
#finding sure foreground area
dist_transform = cv2.distanceTransform(closing, cv2.DIST_L2, 3)
# Threshold
ret, sureFore = cv2.threshold(dist_transform, 0.1 * dist_transform.max(),
255, 0)
#finding unknown region
sureFore = np.uint8(sureFore)
unknown = cv2.subtract(sureBack, sureFore)
#marker labelling
ret, markers, stats, centroids = cv2.connectedComponentsWithStats(sureFore)
#add one to all labels so that sure background is not 0, but 1
markers = markers + 1
#mark the region of unknown with zero
markers[unknown == 255] = 0
# apply watershed
markers = cv2.watershed(img, markers)
img[markers == -1] = [0, 0, 255] #label in red
img[markers > 1] = [255, 0, 0] #label in blue
# to count object detection with area greateer than 350
am, pm = ((np.unique(markers, return_counts=True)))
countB = 0
for i in range(len(am)):
if am[i] > 1 and pm[i] > 350:
countB += 1
# virtually save the result image
ret, imgMark = cv2.imencode('.jpg', img)
imgDiff = cv2.imdecode(imgMark, cv2.IMREAD_COLOR)
# to calculate difference of initial and after detection image
# use to find coverage percentage
difference = cv2.absdiff(imgDiff, imgCrop)
result = difference.astype(
np.uint8) #if difference is all zeros it will return False
percentage = (np.count_nonzero(result) * 100 / result.size)
if (percentage < 66):
percentage = percentage * 1.5
coverage = ("Coverage of the flies is {0:.2f}%".format(percentage))
result = ("[INFO] {} unique segments found".format(countB))
cv2.waitKey(0)
cv2.destroyAllWindows()
return imgMark, result, coverage
def utils_preprocess_img(img,
resize=256,
denoise=False,
remove_color=False,
morphology=False,
segmentation=False,
plot=False,
figsize=(20, 13)):
## original
img_processed = img
lst_imgs = [img_processed]
lst_titles = ["original: " + str(img_processed.shape)]
## scale
#img_processed = img_processed/255
## resize
if resize is not False:
img_processed = cv2.resize(img_processed, (resize, resize),
interpolation=cv2.INTER_LINEAR)
lst_imgs.append(img_processed)
lst_titles.append("resized: " + str(img_processed.shape))
## denoise (blur)
if denoise is True:
img_processed = cv2.GaussianBlur(img_processed, (5, 5), 0)
lst_imgs.append(img_processed)
lst_titles.append("blurred: " + str(img_processed.shape))
## remove color
if remove_color is True:
img_processed = cv2.cvtColor(img_processed, cv2.COLOR_RGB2GRAY)
lst_imgs.append(img_processed)
lst_titles.append("removed color: " + str(img_processed.shape))
## morphology
if morphology is True:
if len(img_processed.shape) > 2:
ret, mask = cv2.threshold(img_processed, 255 / 2, 255,
cv2.THRESH_BINARY)
else:
mask = cv2.adaptiveThreshold(img_processed, 255,
cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 11, 2)
img_processed = cv2.morphologyEx(mask,
cv2.MORPH_OPEN,
np.ones((3, 3), np.uint8),
iterations=2)
lst_imgs.append(img_processed)
lst_titles.append("morphology: " + str(img_processed.shape))
## segmentation (after morphology)
if segmentation is True:
background = cv2.dilate(img_processed,
np.ones((3, 3), np.uint8),
iterations=3)
if len(img_processed.shape) > 2:
print("--- need to remove color to segment ---")
else:
ret, foreground = cv2.threshold(
cv2.distanceTransform(img_processed, cv2.DIST_L2, 5), 0.7 *
cv2.distanceTransform(img_processed, cv2.DIST_L2, 5).max(),
255, 0)
foreground = np.uint8(foreground)
ret, markers = cv2.connectedComponents(foreground)
markers = markers + 1
unknown = cv2.subtract(background, foreground)
markers[unknown == 255] = 0
img_processed = cv2.watershed(
cv2.resize(img,
img_processed.shape,
interpolation=cv2.INTER_LINEAR), markers)
lst_imgs.append(img_processed)
lst_titles.append("segmented: " + str(img_processed.shape))
if (segmentation is True) and (morphology is False):
print("--- need to do morphology to segment ---")
## plot
if plot is True:
plot_imgs(lst_imgs, lst_titles, figsize)
return img_processed
ycoord=[]
radius=[]
silver=[]
for cnt in contours:
cntarea=cv2.contourArea(cnt)
if cntarea<=700:
continue
#see if its a circle
#find inscribing circle{
mask=np.zeros((480,640),np.uint8)
cv2.drawContours(mask, [cnt],-1, (255), -1)
#cv2.drawContours(frame, [cnt],-1, (0,0,255), 3)
#_,mask=cv2.threshold(frame,127,255,cv2.THRESH_BINARY)
dist_transform=cv2.distanceTransform(mask,cv2.DIST_L2,5)
argmax=dist_transform.argmax()
#print(argmax)
x=argmax%640
y=int(argmax/640)
r=dist_transform[y][x]
#maxDT=np.unravel_index(argmax, dist_transform.shape)
'''
circlearea=r*r*3.14159
if((cntarea-circlearea)>5000):
continue
'''
#}
_,R=cv2.minEnclosingCircle(cnt)
if R-r>60:
def main():
"""
"""
if len(sys.argv) > 1:
image = cv2.imread(sys.argv[1], 0)
else:
print "load default image"
image = cv2.imread('NucleiDAPIconfocal.png', 0)
#Array de imagenes
imagenes = [
#title, image, method
['original', image, 'gray'],
]
#imagen a escala de grises
#si aparece algun mensaje de error en pantalla es por esto
try:
print "conversion a gris"
image_gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
except:
#el ejemplo ya estaba en escala de grises :S
#no conversion
image_gray = copy.copy(image)
imagenes.append(['image_gray', image_gray, 'gray'])
#usar el difuminado gausiano no fue muy practico
gnucleo = (3, 3)
gblur_image = cv2.GaussianBlur(image, gnucleo, 0)
imagenes.append(['gblur_image', gblur_image, 'gray'])
ret, thresh = cv2.threshold(gblur_image, 80, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
#ret, thresh = cv2.threshold(image_gray, 80, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
imagenes.append(['thresh', thresh, 'gray'])
kernel = numpy.ones((2,2), numpy.uint8) #nucleo
cl_image = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel, iterations=7)
imagenes.append(['cl_image', cl_image, 'gray'])
## quitar puntos internos
#erosionar imagen
fg_image = cv2.erode(cl_image, None, iterations=2)
imagenes.append(['fg_image', fg_image, 'gray'])
fg_image8 = numpy.uint8(fg_image)
#dilatacion
bgt_image = cv2.dilate(cl_image, None, iterations=5)
ret, bg_image = cv2.threshold(bgt_image, 1, 255, cv2.THRESH_BINARY)
imagenes.append(['bgt_image', bgt_image, 'gray'])
imagenes.append(['bg_image', bg_image, 'gray'])
dist_transform = cv2.distanceTransform(cl_image, cv2.DIST_L2, 5)
imagenes.append(['dist_transform', dist_transform, 'gray'])
ret, sure_fg = cv2.threshold(dist_transform, 0.65*dist_transform.max(), 255, 0)
imagenes.append(['sure_fg', sure_fg, 'gray'])
sure_fg8 = numpy.uint8(sure_fg)
ot, contornos, hier = cv2.findContours(sure_fg8, mode=cv2.RETR_CCOMP, method=cv2.CHAIN_APPROX_SIMPLE)
imagenes.append(['sure_fg8', sure_fg8, 'gray'])
dist_transform_b = cv2.distanceTransform(bg_image, cv2.DIST_L2, 5)
imagenes.append(['dist_transform_b', dist_transform_b, 'gray'])
ret, sure_fg_b = cv2.threshold(dist_transform_b, 0.65*dist_transform.max(), 255, 0)
imagenes.append(['sure_fg_b', sure_fg_b, 'gray'])
sure_fg_b8 = numpy.uint8(sure_fg_b)
ot, contornos_b, hier = cv2.findContours(sure_fg_b8, mode=cv2.RETR_CCOMP, method=cv2.CHAIN_APPROX_SIMPLE)
imagenes.append(['sure_fg_b8', sure_fg_b8, 'gray'])
#comparo cual fue la imager que mejores resultados proporciono
ca = len(contornos or [])
cb = len(contornos_b or [])
if ca >= cb:
print "Cantidad de Objectos: {}".format(ca)
else:
print "Cantidad de Objectos: {}".format(cb)
contornos = contornos_b
circ_img = copy.copy(image)
num_img = copy.copy(image)
i = 1
for cnt in contornos:
(x, y), rad = cv2.minEnclosingCircle(cnt)
center = (int(x), int(y))
center2 = (int(x)-10, int(y)-10)
rad = int(rad)
if rad < 3: #algunos contornos detectados son muy pequenios
rad += 6
cv2.circle(circ_img, center, rad, (0,0,0), -1)
cv2.putText(num_img, str(i), center2, cv2.FONT_HERSHEY_SIMPLEX, 0.8, (0,0,0), 2, cv2.LINE_AA)
i += 1
imagenes.append(['num_img', num_img, 'gray'])
imagenes.append(['circ_img', circ_img, 'gray'])
save_images(imagenes)
#muestra todas las imagenes
plot(imagenes)
pyplot.show()
def run_svm_OTSU(modelFile, test_dir):
global total_Discocyte
global total_Echinocyte
global total_Stomatocyte
global total_Others
global prediction
# load model
pick = open(modelFile, 'rb') # change model depending on SVC parameter
model = pickle.load(pick)
pick.close()
# image pipeline
# img1 = glob.glob("test_image/*.jpg")
# for images in img1:
# # load image
# bld_img = cv2.imread(images)
# bld_img = cv2.resize(bld_img, dsize)
# scan_image(bld_img, winW, winH, model)
# load single image
blood_img = cv2.imread(test_dir)
# sharpen
# kernel = np.array([[1, 1, 1], [1, -8, 1], [1, 1, 1]], dtype = np.float32)
# laplacian = cv2.filter2D(blood_img, cv2.CV_32F, kernel)
# sharp = np.float32(blood_img)
# blood_img = sharp - laplacian
# blood_img = np.clip(blood_img, 0, 255)
# blood_img = blood_img.astype('uint8')
# laplacian = np.clip(laplacian, 0, 255)
# laplacian = np.uint8(laplacian)
#cv2.imshow("sharpened", blood_img)
cv2.waitKey(0)
# nagcreate lang ako ng copy sa line na to
#blood_img2 = blood_img
# load image as grayscale
blood_img_gray = cv2.cvtColor(blood_img, cv2.COLOR_BGR2GRAY)
# perform OTSU's binarization method
ret, blood_img_otsu = cv2.threshold(blood_img_gray, 0, 255, cv2.THRESH_OTSU)
cv2.imshow("otsu", blood_img_otsu)
# inverse otsu (black bg, white foreground)
invblood_otsu = cv2.bitwise_not(blood_img_otsu)
# cv2.imshow("inverse otsu", invblood_otsu)
# noise removal
kernel1 = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (7,7)) #7,7
opening = cv2.morphologyEx(invblood_otsu, cv2.MORPH_OPEN, kernel1, iterations=1)
cv2.imshow("erode-dilate", opening)
cv2.waitKey(0)
# sure background area
sure_bg = cv2.dilate(opening, kernel1, iterations=10)
cv2.imshow("bg", sure_bg)
cv2.waitKey(0)
# Finding sure foreground area
dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 3)
cv2.normalize(dist_transform, dist_transform, 0, 1.0, cv2.NORM_MINMAX)
cv2.imshow("normalized", dist_transform)
cv2.waitKey(0)
ret, sure_fg = cv2.threshold(dist_transform, 0.157 * dist_transform.max(), 255, 0) #0.157
cv2.imshow("fg", sure_fg)
# sure_fg = cv2.dilate(sure_fg, kernel1)
# cv2.imshow("peaks", sure_fg)
# Finding unknown region
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_bg, sure_fg)
cv2.waitKey(0)
# Marker labelling
ret, markers = cv2.connectedComponents(sure_fg)
# Add one to all labels so that sure background is not 0, but 1
markers = markers + 1
# Now, mark the region of unknown with zero
markers[unknown == 255] = 0
markers = cv2.watershed(blood_img, markers)
blood_img[markers == -1] = [0, 255, 255]
ws = color.label2rgb(markers, bg_label=0)
cv2.imshow("overlay on original image", blood_img)
cv2.imshow("watershed", ws)
cv2.waitKey(0)
#
# # # inverse otsu (black bg, white foreground)
# # invblood_otsu = cv2.bitwise_not(blood_img_otsu)
# #
# # cv2.imshow("inverse otsu", invblood_otsu)
#
# sure_fg_8u = sure_fg.astype('uint8')
#
# contours, hierarchy = cv2.findContours(sure_fg_8u, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
#
# markers = np.zeros(sure_fg.shape, dtype=np.int32)
#
# for i in range(len(contours)):
# cv2.drawContours((markers, contours, i, (i+1), -1))
#
# cv2.circle(markers, (5,5), 3, (255,255,255), -1)
# cv2.imshow('Markers', markers * 10000)
#
# cv2.waitKey(0)
contours, hierarchy = cv2.findContours(opening, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
for cnt in contours:
cv2.drawContours(opening, [cnt], 0, 255, -1)
cv2.imshow("filled", opening)
img = opening
# # fill holes found on Inverse Otsu
# contours, hierarchy = cv2.findContours(invblood_otsu, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# for cnt in contours:
# cv2.drawContours(invblood_otsu, [cnt], 0, 255, -1)
#
# cv2.imshow("filled", invblood_otsu)
#
# # Clean cell edges
# kernel = np.ones((2, 2), np.uint8)
# img = cv2.morphologyEx(invblood_otsu, cv2.MORPH_OPEN, kernel, iterations=3)
# #cv2.imshow('Morphological', img)
# BlobDetector Parameters
params = cv2.SimpleBlobDetector_Params()
params.minThreshold = 0
params.maxThreshold = 256
params.filterByArea = True
params.maxArea = 13000
params.minArea = 1000 #2000
params.filterByColor = True
params.blobColor = 255
params.filterByCircularity = False
params.maxCircularity = 1.00
params.minCircularity = 0.75
params.filterByConvexity = True
params.maxConvexity = 1.00
params.minConvexity = 0.00
params.filterByInertia = True
params.maxInertiaRatio = 1.00
params.minInertiaRatio = 0.00
# BlobDetector
detector = cv2.SimpleBlobDetector_create(params)
keypoints = detector.detect(img)
draw = cv2.drawKeypoints(img, keypoints, np.array([]), (0, 0, 255), cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
# for keypoints in keypoints:
# print(keypoints.size)
cv2.imshow("Keypoints", draw)
cv2.waitKey(0)
#General Test Param
# # Position offset variables
# ul = 50
# jl = 100
#
# #patch size
# PTCH_SZ = 55 #52
#Stomatest_Param
# Position offset variables
ul = 35
jl = 70
# patch size
PTCH_SZ = 30 # 52
# #Echinotest Param
# # Position offset variables
# ul = 40
# jl = 70
#
# # patch size
# PTCH_SZ = 40 # 52
# # find center of each blob
# contours, hierarchy = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# # contours, hierarchy = cv2.findContours(blood_img_otsu, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
#
# cv2.imshow("pre-moment", img)
#
# for c in contours:
# M = cv2.moments(c)
#
# if M["m00"] != 0:
# cX = int(M["m10"] / M["m00"])
# cY = int(M["m01"] / M["m00"])
# else:
# cX, cY = 0, 0
#
# cv2.circle(blood_img, (cX, cY), 5, (255, 0, 0), -1)
# cv2.imshow("Centroids", blood_img)
#cv2.putText(blood_img, "centroid", (cX - 25, cY - 25), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 2)
for kp in keypoints:
cX = int(kp.pt[0])
cY = int(kp.pt[1])
cv2.circle(blood_img, (cX, cY), 3, (255, 0, 0), -1)
# cv2.waitKey(0)
# Position variables
# initial location variables (x1.y1)
# final location variables (x2,y2)
# variables used to create ROI on the parts of the image
x1 = cX - ul
y1 = cY + ul
x2 = x1 + jl
y2 = y1 - jl
# x1 = cX - w
# y1 = cY + h
#
# x2 = x1 + w
# y2 = y1 - h
# create rectangle on each identified blob forming around the center of the blob
cv2.rectangle(blood_img, (x1, y1), (x2, y2), (0, 255, 200), 1)
cv2.imshow("Moments", blood_img)
cv2.waitKey(0)
# cv2.imshow("Image", blood)
# pwedeng icomment muna itong since ung currently trained dataset, nakabase sa grayscale lang.
# If patch of image to be extracted exceeds the image size (640x640),
# ignore patch, then move on to next patch
if x1 < 0 or x2 > 640 or y1 > 640 or y2 < 0:
continue
try:
clone = blood_img_otsu.copy()
patch1 = clone[cY-PTCH_SZ:cY+PTCH_SZ, cX-PTCH_SZ:cX+PTCH_SZ]
# patch1 = clone[y1:y1 + jl, x1:x1 + jl]
# patch1 = clone[x1:x2, y1:y2]
# patch1 = clone[y1:y2, x1,x2]
# patch1 = clone[x1:x1 + jl, y1:y1 + jl]
# patch2 = cv2.cvtColor(patch1, cv2.COLOR_BGR2GRAY)
patch3 = cv2.resize(patch1, (48, 48))
cv2.namedWindow("cropped", cv2.WINDOW_NORMAL)
cv2.resizeWindow("cropped", 48 * 6, 48 * 6)
cv2.imshow("cropped", patch3)
# cv2.waitKey(0)
patch4 = np.array(patch3).flatten()
# patch_final = patch4[0]
patch_final = patch4.reshape(-1, 2304)
# prediction = None
prediction = model.predict(patch_final)
# prediction[0] = 1
if prediction[0] == 0:
total_Others = total_Others + 1
cv2.rectangle(blood_img, (x1, y1), (x2, y2), (255, 200, 100), 2)
cv2.putText(blood_img, "others", (cX - 25, cY - 25), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 200, 100), 2)
elif prediction[0] == 1:
total_Discocyte = total_Discocyte + 1
cv2.rectangle(blood_img, (x1, y1), (x2, y2), (0, 0, 255), 2)
cv2.putText(blood_img, "discocyte", (cX - 25, cY - 25), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 2)
elif prediction[0] == 2:
total_Echinocyte = total_Echinocyte + 1
cv2.rectangle(blood_img, (x1, y1), (x2, y2), (0, 255, 0), 2)
cv2.putText(blood_img, "echinocyte", (cX - 25, cY - 25), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
elif prediction[0] == 3:
total_Stomatocyte = total_Stomatocyte + 1
cv2.rectangle(blood_img, (x1, y1), (x2, y2), (255, 80, 0), 2)
cv2.putText(blood_img, "stomatocyte", (cX - 25, cY - 25), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 80, 0),2)
cv2.imshow("Image", blood_img)
except BaseException as ex:
#print(ex.message, ex.args)
continue
img = cv2.imread('../img/coins_connected.jpg')
rows, cols = img.shape[:2]
cv2.imshow('original', img)
mean = cv2.pyrMeanShiftFiltering(img, 20, 50)
cv2.imshow('mean', mean)
gray = cv2.cvtColor(mean, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray, (3,3), 0)
_, thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)
cv2.imshow('thresh', thresh)
dst = cv2.distanceTransform(thresh, cv2.DIST_L2, 3)
dst = ( dst / (dst.max() - dst.min()) * 255 ).astype(np.uint8)
cv2.imshow('dst', dst)
localMx = cv2.dilate(dst, np.ones((50,50), np.uint8))
lm = np.zeros((rows, cols), np.uint8)
lm[(localMx==dst) & (dst != 0)] = 255
cv2.imshow('localMx', lm)
seeds = np.where(lm ==255)
seed = np.stack( (seeds[1], seeds[0]), axis=-1)
fill_mask = np.zeros((rows+2, cols+2), np.uint8)
for x,y in seed:
ret = cv2.floodFill(mean, fill_mask, (x,y), (255,255,255), (10,10,10), (10,10,10))
C_flag[min_prv_mask_idy, min_prv_mask_idx - 2] = 1
C_flag[min_prv_mask_idy + 2, min_prv_mask_idx] = 1
C_flag[min_prv_mask_idy - 2, min_prv_mask_idx] = 1
print("Previous mask min at value " + str(min_prv_mask_val) + " K")
C_Core = C_flag[:]
#%% Start spreading
#% First round, sepearate the connect blobs relating to the cyclone
C_binary = np.where(C_BTemp > 280, 0, C_BTemp)
C_binary = np.where(C_binary > 1, 1, C_binary)
C_binary8 = C_binary.astype(np.uint8)
I_idx = get_coord_to_idx(I_lat[C_i], I_lon[C_i])
kernel = np.ones((3, 3), np.uint8)
opening = cv2.morphologyEx(C_binary8, cv2.MORPH_OPEN, kernel, iterations=2)
dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 0)
# dist_matrix = dist_transform[I_idx[0]-100:I_idx[0]+100,I_idx[1]-100:I_idx[1]+100]
# dist_max = np.unravel_index(np.argmax(dist_matrix, axis=None), dist_matrix.shape)
# dist_max_big = [I_idx[0]-100 + dist_max[0],I_idx[1] - 100 + dist_max[1]]
# markers_one = np.zeros([DIM_LAT,DIM_LON])
# markers_one[dist_max_big[0],dist_max_big[1]] = 1
# dist_transform_mod = np.where(dist_transform<20,0,dist_transform)
labels_ws = watershed(-dist_transform, C_flag, mask=C_binary8)
C_binary8_second = np.where(labels_ws > 0, C_binary8, 0)
#%% Second round - divide the previous mask into blobs
seed = np.copy(C_binary8_second)
seed[1:-1, 1:-1] = C_binary8_second.max()
async def setup_learner():
await download_file(export_file_url, path / export_file_name)
try:
img_new = cv2.imread(str(path), cv2.IMREAD_COLOR)
img_new = cv2.fastNlMeansDenoisingColored(img_new, None, 10, 10, 7, 21)
img_sv = img_new
img_test = img_sv
img = img_new
r = img.copy()
r[:, :, 0] = 0
r[:, :, 1] = 0
img = r
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
test_r = img
im_gray = cv2.cvtColor(r, cv2.COLOR_BGR2GRAY)
im_blur = cv2.GaussianBlur(im_gray, (5, 5), 0)
kernel_sharpening = np.array([[-1, -1, -1], [-1, 9, -1], [-1, -1, -1]])
im_sharp = cv2.filter2D(im_blur, -1, kernel_sharpening)
img = im_sharp
th3 = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 11, 2)
img = im_sharp
ret, thresh3 = cv2.threshold(img, 127, 255, cv2.THRESH_TRUNC)
kernel = np.ones((3, 3), np.uint8)
opening = cv2.morphologyEx(thresh3,
cv2.MORPH_OPEN,
kernel,
iterations=2)
ret, th = cv2.threshold(thresh3, 0, 255,
cv2.THRESH_BINARY + cv2.THRESH_OTSU)
kernel = np.ones((3, 3), np.uint8)
opening = cv2.morphologyEx(th, cv2.MORPH_OPEN, kernel, iterations=2)
sure_bg = cv2.dilate(th, kernel, iterations=3)
dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 5)
ret, sure_fg = cv2.threshold(dist_transform,
0.005 * dist_transform.max(), 255, 0)
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_bg, sure_fg)
ret, markers = cv2.connectedComponents(sure_fg)
markers = markers + 1
markers[unknown == 255] = 0
markers = cv2.watershed(img_sv, markers)
img_sv[markers == -1] = [0, 255, 0]
img_sv[markers == 1] = [0, 0, 0]
from skimage.filters import threshold_otsu
thresh_val = threshold_otsu(opening)
mask = np.where(opening > thresh_val, 1, 0)
if np.sum(mask == 0) < np.sum(mask == 1):
mask = np.where(mask, 0, 1)
from scipy import ndimage
labels, nlabels = ndimage.label(mask)
all_cords = {}
for label_ind, label_coords in enumerate(ndimage.find_objects(labels)):
cell = im_gray[label_coords]
all_cords[label_ind] = label_coords
if np.product(cell.shape) < 10:
mask = np.where(labels == label_ind + 1, 0, mask)
labels, nlabels = ndimage.label(mask)
i = 0
res = {}
for ii, obj_indices in enumerate(
ndimage.find_objects(labels)[0:nlabels]):
cell = img_sv[obj_indices]
res[i] = cell
i = i + 1
i = 0
features = {}
for i in res:
gcolor = res[i]
h, w, _ = gcolor.shape
ggray = np.array([[gcolor[i, j, 2] for j in range(w)]
for i in range(h)],
dtype=np.uint8)
area = np.sum(
np.sum([[1.0 for j in range(w) if ggray[i, j]]
for i in range(h)]))
mean_area = area / (h * w)
r, b, g = np.sum(
[gcolor[i, j] for j in range(w)
for i in range(h)], axis=0) / (area * 256)
_, _, eigen_value = pca(ggray)
eccentricity = eigen_value[0] / eigen_value[1]
l = [
mean_area, r, b, g, eigen_value[0], eigen_value[1],
eccentricity
]
features[i] = np.array(l)
i = i + 1
out = {}
#learn = load_learner(path, export_file_name)
#Change Working directory of pkl file
model = keras.models.load_model('../input/weight/weights.pkl')
out = {}
for i in features:
out[i] = model.predict(np.array([features[i]]))
good = not_good = 0
for i in out:
s = res[i]
if np.argmax(out[i][0]) == 0:
good += 1
x1 = all_cords[i][0].start
y1 = all_cords[i][1].start
x2 = all_cords[i][1].stop
y2 = all_cords[i][0].stop
cv2.rectangle(img_test, (x2, x1), (y1, y2), (255, 0, 0), 8)
else:
x1 = all_cords[i][0].start
y1 = all_cords[i][1].start
x2 = all_cords[i][1].stop
y2 = all_cords[i][0].stop
not_good += 1
cv2.rectangle(img_test, (x2, x1), (y1, y2), (0, 0, 255), 3)
p = (good / (good + not_good) * 100)
print("Number of good grain :", good)
print("Number of impure grains or impurity:", not_good)
print("Percentage Purity is:", p)
return p
except RuntimeError as e:
if len(e.args) > 0 and 'CPU-only machine' in e.args[0]:
print(e)
message = "\n\nThis model was trained with an old version of fastai and will not work in a CPU environment.\n\nPlease update the fastai library in your training environment and export your model again.\n\nSee instructions for 'Returning to work' at https://course.fast.ai."
raise RuntimeError(message)
else:
raise
def build_dataset(base="COMP9517 20T2 Group Project Image Sequences/DIC-C2DH-HeLa",
kernel_sz=5,
scale=1):
# Initialize the kernel used for morphological operations
ker = cv.getStructuringElement(cv.MORPH_CROSS,(kernel_sz, kernel_sz))
# ker = np.ones((kernel_sz, kernel_sz), np.uint8)
samples = list()
# Build a list of samples
for d in os.listdir(base):
if "Mask" in d:
d_n = base + '/' + d
for f in os.listdir(d_n):
f_n = d_n + '/' + f
samples.append((f_n, f_n.replace(" Masks", "").replace("mask", "")))
# Iterate through each sample and its mask
for i, (mask_path, img_path) in enumerate(samples):
# Load the image and its mask
img = cv.imread(img_path, cv.IMREAD_GRAYSCALE)
mask = cv.imread(mask_path, cv.IMREAD_ANYDEPTH)
seg = np.zeros(mask.shape, np.uint8)
bor = np.zeros(mask.shape, np.uint8)
dst_exp = np.zeros(mask.shape, np.uint8)
# Iterate through unique values in the mask image
for j in np.unique(mask):
if j == 0:
continue
# Create an image with just that object, erode it to emphasize the
# border and add it to the current segmentation mask
tmp = (mask == j).astype(np.uint8)
dst = cv.distanceTransform(tmp, cv.DIST_L2, 3)
dst = ((np.exp((np.log(256) / dst.max())) ** dst) - 1).astype(np.uint8)
dst_exp += dst
tmp = cv.erode(tmp, ker)
seg += tmp
# Dilate the same object mask, and add it to the border image
tmp = cv.dilate((mask == j).astype(np.uint8), ker, iterations=3)
bor += tmp
# If any pixel has a value greater than 1, then at least 2 dilated
# objects occupy that space. Filter to get the border, then subtract the
# object segmentation mask to get only the borders
bor = (bor > 1).astype(np.uint8)
bor *= (1 - seg)
img = cv.resize(img, None, fx=scale, fy=scale)
# img = img - img.min()
# img = img * (255 // img.max())
seg = cv.resize(seg, None, fx=scale, fy=scale,
interpolation=cv.INTER_NEAREST)
bor = cv.resize(bor, None, fx=scale, fy=scale,
interpolation=cv.INTER_NEAREST)
dst_exp = cv.resize(dst_exp, None, fx=scale, fy=scale)
# Write images to their respective folders
cv.imwrite(f"data/img/{i}.png", img)
cv.imwrite(f"data/seg/{i}.png", seg)
cv.imwrite(f"data/bnd/{i}.png", bor)
cv.imwrite(f"data/dst/{i}.png", dst_exp)
def main(img_path):
src = cv.imread(img_path)
# Show source image
cv.imshow('Source Image', src)
## [load_image]
gray = cv.cvtColor(src, code=cv.COLOR_RGB2GRAY)
thr_b = getTresholdByGrad(gray)
# thr_b = 40
print('\n\n\n thr_b: {}'.format(thr_b))
## [black_bg]
# Change the background from white to black, since that will help later to extract
# better results during the use of Distance Transform
# src[np.all(src == 255, axis=2)] = 0
src[np.all(src < thr_b, axis=2)] = 0 # 背景变黑
# src[np.all(src == 0, axis=2)] = 255
# Show output image
# cv.imshow('Black Background Image', src)
## [black_bg]
## [sharp]
# Create a kernel that we will use to sharpen our image an approximation of second derivative, a quite strong kernel
kernel = np.array([[1, 1, 1], [1, -8, 1], [1, 1, 1]], dtype=np.float32)
# do the laplacian filtering as it is
# well, we need to convert everything in something more deeper then CV_8U
# because the kernel has some negative values,
# and we can expect in general to have a Laplacian image with negative values
# BUT a 8bits unsigned int (the one we are working with) can contain values from 0 to 255
# so the possible negative number will be truncated
imgLaplacian = cv.filter2D(src, cv.CV_32F, kernel)
sharp = np.float32(src)
imgResult = sharp - imgLaplacian
# convert back to 8bits gray scale
imgResult = np.clip(imgResult, 0, 255)
imgResult = imgResult.astype('uint8')
imgLaplacian = np.clip(imgLaplacian, 0, 255)
imgLaplacian = np.uint8(imgLaplacian)
cv.imshow('Laplace Filtered Image', imgLaplacian)
cv.imshow('New Sharped Image', imgResult)
## [sharp]
## [bin]
# Create binary image from source image
bw = cv.cvtColor(imgResult, cv.COLOR_BGR2GRAY)
_, bw = cv.threshold(bw, thr_b, 255, cv.THRESH_BINARY)
cv.imshow('Binary Image', bw)
## [bin]
## [dist]
# Perform the distance transform algorithm
dist = cv.distanceTransform(bw, cv.DIST_L2, 3)
# Normalize the distance image for range = {0.0, 1.0}
# so we can visualize and threshold it
cv.normalize(dist, dist, 0, 1.0, cv.NORM_MINMAX)
cv.imshow('Distance Transform Image', dist)
## [dist]
## [peaks]
# Threshold to obtain the peaks
# This will be the markers for the foreground objects
_, dist = cv.threshold(dist, 0.4, 1.0, cv.THRESH_BINARY)
# Dilate a bit the dist image
kernel1 = np.ones((3, 3), dtype=np.uint8)
dist = cv.dilate(dist, kernel1)
cv.imshow('Peaks', dist)
## [peaks]
## [seeds]
# Create the CV_8U version of the distance image
# It is needed for findContours()
dist_8u = dist.astype('uint8')
# Find total markers
_, contours, _ = cv.findContours(dist_8u, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_SIMPLE)
# Create the marker image for the watershed algorithm
markers = np.zeros(dist.shape, dtype=np.int32)
# Draw the foreground markers
for i in range(len(contours)):
# drawContours(image, contours, contourIdx, color, thickness=None,
# lineType=None, hierarchy=None, maxLevel=None, offset=None)
# fills the area bounded by the contours if thickness<0
print('len(contours): {}'.format(len(contours)))
cv.drawContours(markers, contours, i, (2, ), thickness=-1)
# Draw the background marker
# cv.circle(markers, (5,5), 122, (255,255,255), -1)
cv.imshow('Markers', markers*10000)
## [seeds]
## [watershed]
# Perform the watershed algorithm
cv.watershed(imgResult, markers)
#mark = np.zeros(markers.shape, dtype=np.uint8)
mark = markers.astype('uint8')
mark = cv.bitwise_not(mark)
# uncomment this if you want to see how the mark
# image looks like at that point
# cv.imshow('Markers_v2', imgResult)
# Generate random colors
colors = []
for contour in contours:
colors.append((rng.randint(0, 256), rng.randint(0, 256), rng.randint(0, 256)))
# Create the result image
dst = np.zeros((markers.shape[0], markers.shape[1], 3), dtype=np.uint8)
# Fill labeled objects with random colors
for i in range(markers.shape[0]):
for j in range(markers.shape[1]):
index = markers[i,j]
if index > 0 and index <= len(contours):
dst[i,j,:] = colors[index-1]
# Visualize the final image
cv.imshow('Final Result', dst)
## [watershed]
cv.waitKey()
def main():
img = cv2.imread(filename)
imgray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
## Binary (OTSU's method)
ret, thresh = cv2.threshold(imgray, 0, 255,
cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
#ret, thresh = cv2.threshold(imgray, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
print 'Threshold (pixel) : ', ret
## Remove noise
kernel = np.ones((3, 3), np.uint8)
opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=2)
## Distance transformation
dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 5)
# dist_transform = cv2.distanceTransform(opening,cv2.DIST_L1,3)
# dist_transform = cv2.distanceTransform(opening,cv2.DIST_C,3)
dist_transform_8u = np.uint8(dist_transform)
## Find particles
particles = find_peak(dist_transform_8u, k_size)
print particles
print 'Number of particles : ', len(particles)
## Shape discription (Snake algorithm)
init_list = []
init_list2 = []
snake_list = []
particle_list = []
k = 0
for p in particles:
k += 1
print k
init = set_init(p[0], p[1], p[2], p[3])
init_list.append(init), init_list2.extend(init)
contour = (active_contour(gaussian(thresh, 3),
init,
alpha=0.015,
beta=10,
gamma=0.001))
snake_list.append(contour)
particle_list.extend(contour)
print np.array(snake_list)
print cv2.contourArea(np.array(snake_list))
with open("snake_list_%s.csv" % filename.split('.')[0], "wb") as f:
writer = csv.writer(f)
writer.writerows(snake_list)
f.close()
## Plot figures
plt.figure(), plt.imshow(img), plt.title('Original'), plt.xticks(
[]), plt.yticks([])
plt.savefig("figures_%s_%s.png" % (filename.split('.')[0], 'Original'))
plt.figure(), plt.imshow(imgray), plt.title('Gray'), plt.xticks(
[]), plt.yticks([])
plt.savefig("figures_%s_%s.png" % (filename.split('.')[0], 'Gray'))
plt.figure(), plt.imshow(thresh), plt.title('Binary'), plt.xticks(
[]), plt.yticks([])
plt.savefig("figures_%s_%s.png" % (filename.split('.')[0], 'Binary'))
plt.figure(), plt.imshow(
dist_transform_8u,
cmap=plt.cm.gray), plt.title('Distance transform'), plt.xticks(
[]), plt.yticks([])
plt.savefig("figures_%s_%s.png" % (filename.split('.')[0], 'Distance'))
plt.figure(), plt.hist(imgray.ravel(), 256,
color='black'), plt.title('Histogram'), plt.xticks(
[]), plt.yticks([])
plt.savefig("figures_%s_%s.png" % (filename.split('.')[0], 'Histogram'))
plt.close()
plt.figure(), plt.imshow(thresh), plt.title('Binary+Snake'), plt.xticks(
[]), plt.yticks([])
for p in range(len(particles)):
#plt.plot(particles[p][0], particles[p][2], 'og')#, lw=3)
plt.plot(init_list[p][:, 0], init_list[p][:, 1], '--r') #, lw=3)
plt.plot(snake_list[p][:, 0], snake_list[p][:, 1], '-b') #, lw=3)
plt.savefig("figures_%s_Binary_Snake.png" % filename.split('.')[0])
plt.close()
plt.figure(), plt.imshow(
dist_transform_8u,
cmap=plt.cm.gray), plt.title('Distance+Snake'), plt.xticks(
[]), plt.yticks([])
for p in range(len(particles)):
#plt.plot(particles[p][0], particles[p][2], 'og')#, lw=3)
plt.plot(init_list[p][:, 0], init_list[p][:, 1], '--r') #, lw=3)
plt.plot(snake_list[p][:, 0], snake_list[p][:, 1], '-b') #, lw=3)
plt.savefig("figures_%s_Distance_Snake.png" % filename.split('.')[0])
plt.close()
plt.figure(), plt.imshow(img), plt.title('Original+Snake'), plt.xticks(
[]), plt.yticks([])
for p in range(len(particles)):
#plt.plot(particles[p][0], particles[p][2], 'og')#, lw=3)
#plt.plot(init_list[p][:, 0], init_list[p][:, 1], '--r')#, lw=3)
plt.plot(snake_list[p][:, 0], snake_list[p][:, 1], '-b') #, lw=3)
plt.savefig("figures_%s_Original_Snake.png" % filename.split('.')[0])
plt.close()
def estimate_navigation_channels(floor_mask,
idealized_height_image,
m_per_pix,
display_on=False,
verbose=False):
# min_robot_width_m : The best case minimum width of the robot in meters when moving forward and backward.
# threshold based on minimum radius of the robot
min_robot_radius_m = 0.4
# subtract pixels to provide an optimistic estimate and account for possible quantization
min_robot_radius_pix = (min_robot_radius_m / m_per_pix) - 2.0
# create optimistic traversable region mask
traversable_image = np.zeros_like(floor_mask, dtype=np.uint8)
# if the region is floor or unobserved, consider it to be traversable
traversable_selector = (floor_mask > 0) | (idealized_height_image == 0)
traversable_image[traversable_selector] = 255
if display_on:
cv2.imshow('traversable image', traversable_image)
# compute distance map: distance from untraversable regions
original_dist_transform = cv2.distanceTransform(traversable_image,
cv2.DIST_L2, 5)
dist_transform = original_dist_transform.copy()
norm_dist_transform = cv2.normalize(dist_transform, None, 0, 255,
cv2.NORM_MINMAX, cv2.CV_8U)
if display_on:
cv2.imshow('distance transform of the traversable image',
norm_dist_transform)
########
# remove unobserved and untraversable pixels from the distance map
# set parts of the distance transform that are off of the observed
# floor to zero
floor_selector = (floor_mask < 1)
dist_transform[floor_selector] = 0
# set parts of the distance transform that represent free space
# less than the robot would require to zero
robot_radius_selector = (dist_transform < min_robot_radius_pix)
dist_transform[robot_radius_selector] = 0
if display_on:
norm_dist_transform = cv2.normalize(dist_transform, None, 0, 255,
cv2.NORM_MINMAX, cv2.CV_8U)
cv2.imshow('cropped distance transform', norm_dist_transform)
########
# create kernel for morphological operations
kernel_width_pix = 11
kernel_radius_pix = (kernel_width_pix - 1) / 2
kernel = np.zeros((kernel_width_pix, kernel_width_pix), np.uint8)
cv2.circle(kernel, (kernel_radius_pix, kernel_radius_pix),
kernel_radius_pix, 255, -1)
# fill in floor mask holes
closed_floor_mask = cv2.morphologyEx(floor_mask, cv2.MORPH_CLOSE, kernel)
# find the boundary of the floor
dilated_floor = cv2.dilate(closed_floor_mask, kernel, iterations=1)
floor_boundary = dilated_floor - closed_floor_mask
# Estimate the locations of exits. Creates a binary image with exits marked with 255.
min_connectivity_dist_m = min_robot_radius_m
min_connectivity_dist_pix = min_robot_radius_pix
connectivity_image = original_dist_transform.copy()
connectivity_image[connectivity_image < min_connectivity_dist_pix] = 0
connectivity_image[floor_boundary == 0] = 0
map_exits = np.zeros_like(connectivity_image, dtype=np.uint8)
map_exits[connectivity_image > 0] = 255
exit_dilation = True
if exit_dilation:
# attempt to increase the chance of a vertex being labeled as an exit
kernel = np.ones((21, 21), np.uint8)
map_exits = cv2.dilate(map_exits, kernel, iterations=1)
# find exit regions
number_of_exits, exit_label_image = simple_connected_components(map_exits)
print 'number_of_exits =', number_of_exits
distance_map = original_dist_transform.copy()
distance_map[closed_floor_mask < 1] = 0
exit_locations = []
for i in range(number_of_exits)[1:]:
distance = distance_map.copy()
not_component_selector = (exit_label_image != i)
distance[not_component_selector] = 0
y, x = np.unravel_index(np.argmax(distance), distance.shape)
exit_locations.append([x, y])
if display_on:
norm_connectivity_image = cv2.normalize(connectivity_image, None, 0,
255, cv2.NORM_MINMAX,
cv2.CV_8U)
cv2.imshow('connectivity image', norm_connectivity_image)
cv2.imshow('estimated map exits', map_exits)
cv2.imshow('floor boundary', floor_boundary)
norm_distance_map = cv2.normalize(distance_map, None, 0, 255,
cv2.NORM_MINMAX, cv2.CV_8U)
cv2.imshow('raw distance map', norm_distance_map)
return dist_transform, map_exits, distance_map, exit_locations
import cv2 # Import relevant libraries
import numpy as np
img = cv2.imread('cameraman.png', 0) # Read in image
height = img.shape[0] # Get the dimensions
width = img.shape[1]
# Define mask
mask = 255 * np.ones(img.shape, dtype='uint8')
# Draw circle at x = 100, y = 70 of radius 25 and fill this in with 0
cv2.circle(mask, (100, 70), 25, 0, -1)
# Apply distance transform to mask
out = cv2.distanceTransform(mask, cv2.DIST_L2, 3)
# Define scale factor
scale_factor = 10
# Create output image that is the same as the original
filtered = img.copy()
# Create floating point copy for precision
img_float = img.copy().astype('float')
# Number of channels
if len(img_float.shape) == 3:
num_chan = img_float.shape[2]
else:
# If there is a single channel, make the images 3D with a singleton
opening = cv2.morphologyEx(gray_mask, cv2.MORPH_CLOSE, kernel)
erosion1 = cv2.erode(opening, kernel, iterations=1)
opening2 = cv2.morphologyEx(erosion1,
cv2.MORPH_CLOSE,
kernel,
iterations=2)
#####################################################################
sure_img = cv2.dilate(opening2, kernel, iterations=3)
######################################################################
# contours, hierarchy = cv2.findContours(sure_img, cv2.RETR_TREE+cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
#######################################################################
dist_transform = cv2.distanceTransform(opening2, cv2.DIST_L2, 5)
ret, sure_fg = cv2.threshold(dist_transform, 0.5 * dist_transform.max(),
255, 0)
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_img, sure_fg)
ret, markers = cv2.connectedComponents(sure_fg)
# Add one to all labels so that sure background is not 0, but 1
markers = markers + 1
# Now, mark the region of unknown with zero
markers[unknown == 250] = 0
markers = cv2.watershed(bilateralFilter, markers)
# print(markers.shape, markers.dtype)
def textDetectWatershed(thresh):
""" Text detection using watershed algorithm - NOT IN USE """
# According to: http://docs.opencv.org/trunk/d3/db4/tutorial_py_watershed.html
img = cv2.cvtColor(cv2.imread("data/textdet/%s.jpg" % IMG),
cv2.COLOR_BGR2RGB)
img = resize(img, 3000)
thresh = resize(thresh, 3000)
# noise removal
kernel = np.ones((3, 3), np.uint8)
opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=3)
# sure background area
sure_bg = cv2.dilate(opening, kernel, iterations=3)
# Finding sure foreground area
dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 5)
ret, sure_fg = cv2.threshold(dist_transform, 0.01 * dist_transform.max(),
255, 0)
# Finding unknown region
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_bg, sure_fg)
# Marker labelling
ret, markers = cv2.connectedComponents(sure_fg)
# Add one to all labels so that sure background is not 0, but 1
markers += 1
# Now, mark the region of unknown with zero
markers[unknown == 255] = 0
markers = cv2.watershed(img, markers)
implt(markers, t='Markers')
image = img.copy()
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
for mark in np.unique(markers):
# mark == 0 --> background
if mark == 0:
continue
# Draw it on mask and detect biggest contour
mask = np.zeros(gray.shape, dtype="uint8")
mask[markers == mark] = 255
cnts = cv2.findContours(mask.copy(), cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)[-2]
c = max(cnts, key=cv2.contourArea)
# Draw a bounding rectangle if it contains text
x, y, w, h = cv2.boundingRect(c)
cv2.drawContours(mask, c, 0, (255, 255, 255), cv2.FILLED)
maskROI = mask[y:y + h, x:x + w]
# Ratio of white pixels to area of bounding rectangle
r = cv2.countNonZero(maskROI) / (w * h)
# Limits for text
if r > 0.2 and 2000 > w > 15 and 1500 > h > 15:
cv2.rectangle(image, (x, y), (x + w, y + h), (0, 255, 0), 2)
implt(image)
def segment_grabcut(image, window=None, seeds=[]):
"""Segments an image using grabcut technique. Initialised with edges.
Parameters
----------
image : (M, N, 3) array
Image to process.
window : tuple, (x, y, w, h)
Optional subwindow in image.
Returns
-------
(rects, display) : list, (M, N, 3) array
Region results and visualization image.
"""
if window:
subimage = np.array(image)
x, y, w, h = window
image = subimage[y:y + h, x:x + w]
rects, display = segment_edges(image,
variance_threshold=100,
line_filter=0,
size_filter=0)
gray = cv2.cvtColor(image, cv2.cv.CV_BGR2GRAY)
(k, mag0) = cv2.threshold(gray, 128, 255,
cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)
h, w = image.shape[:2]
initial = np.zeros((h, w), np.uint8)
initial[:] = cv2.GC_BGD
initial[mag0 == 255] = cv2.GC_FGD
for i, rect in enumerate(rects):
cv2.drawContours(initial, [rect[4]], -1, int(cv2.GC_FGD), -1)
initial[display[:, :, 0] > 0] = cv2.GC_PR_FGD
bgmodel = np.zeros((1, 65), np.float64)
fgmodel = np.zeros((1, 65), np.float64)
mask = initial
rect = None
cv2.grabCut(image, mask, rect, bgmodel, fgmodel, 1, cv2.GC_INIT_WITH_MASK)
mask2 = np.where((mask == 2) | (mask == 0), 0, 1).astype('uint8')
contours, hierarchy = cv2.findContours(mask2.copy(), cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
rects = [cv2.boundingRect(c) for c in contours]
display = np.dstack(3 * [255 * mask2.astype(np.uint8)])
if seeds:
distance = cv2.distanceTransform(mask2, cv2.cv.CV_DIST_L2, 3)
markers = np.zeros(mask2.shape, dtype=np.int32)
markers[mask2 == 0] = 255
for i, seed in enumerate(seeds):
sx, sy = seed
markers[sy, sx] = i + 1
if USE_OPENCV_WATERSHED:
cv2.watershed(display, markers)
else:
markers = watershed(mask2, markers, mask=mask2)
new_rects = []
for i, seed in enumerate(seeds):
mask = np.array(markers == i + 1, dtype=np.uint8)
contours, _ = cv2.findContours(mask, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
if contours:
contours.sort(
lambda x, y: cmp(cv2.contourArea(y), cv2.contourArea(x)))
new_rects.append(cv2.boundingRect(contours[0]))
colour = [randint(100, 255), randint(100, 255), 0]
cv2.drawContours(display, [contours[0]], -1, colour, -1)
rects = new_rects
if window:
new_rects = []
gray = cv2.cvtColor(image, cv2.cv.CV_BGR2GRAY)
for rect in rects:
dx, dy = 0, 0
new_rect = (rect[0] + x - dx, rect[1] + y - dy, rect[2] + 2 * dx,
rect[3] + 2 * dy)
im = gray[rect[1]:rect[1] + rect[3], rect[0]:rect[0] + rect[2]]
if np.var(im) > 200 and rect[2] * rect[3] > w * h / 4E3:
new_rects.append(new_rect)
rects = new_rects
return rects, display
def get_distances(map):
"""
Computes distance
transform.
"""
return cv.distanceTransform(map, cv.DIST_L2, 5)
def main():
## load datqa
#img = cv.imread("water_coins.jpg")
img, singleJJ = my_data(show=False, return_stamp=True)
# let's look at a small part
'''
## try 3D plot
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
xs = np.linspace(0,1,200)
ys = np.linspace(0,1,80)
xy = np.array( np.meshgrid(xs,ys)[0] )
ax.plot_surface(xy,xy,singleJJ)
plt.title("Cut")
plt.show()
return 0;
#'''
## convert to openCV's favorite version
gray = cv.cvtColor(img, cv.COLOR_BGR2GRAY)
plt.figure()
plt.title("Gray")
plt.imshow(gray, cmap='gray')
## threshold the devices to find only exposed region:
## short story: Gaussian blur then filter gives clean-looking segmentation
## the true magic, is the OTSU method
blur = cv.GaussianBlur(gray, (5, 5), 0)
ret, thresh = cv.threshold(blur,
0,
255,
type=cv.THRESH_BINARY_INV + cv.THRESH_OTSU)
plt.figure()
plt.title("thresh")
plt.imshow(thresh, cmap='gray')
return 0
'''
#ret, thresh = cv.threshold(gray,0,255,cv.THRESH_BINARY_INV+cv.THRESH_OTSU)
ret, thresh = cv.threshold(gray,0,255,type=cv.THRESH_BINARY_INV+cv.THRESH_OTSU)
plt.figure()
plt.title("thresh")
plt.imshow(thresh, cmap='gray')
## try a Gaussian blur first
blur = cv.GaussianBlur(gray,(5,5),0)
ret, thresh_blur = cv.threshold(blur,0,255,type=cv.THRESH_BINARY_INV+cv.THRESH_OTSU)
plt.figure()
plt.title("blur --> thresh")
plt.imshow(thresh_blur, cmap='gray')
## or perhaps an adaptive thresholding
th3 = cv.adaptiveThreshold(gray,255,cv.ADAPTIVE_THRESH_GAUSSIAN_C,\
cv.THRESH_BINARY,11,2)
plt.figure()
plt.title("Adaptive")
plt.imshow(th3, cmap='gray')
'''
## first estimation of noise
'''
# noise removal
kernel = np.ones((3,3),np.uint8)
opening = cv.morphologyEx(thresh,cv.MORPH_OPEN,kernel, iterations = 2)
plt.figure()
plt.imshow(opening, cmap='gray')
plt.title("Opening")
# sure background area
sure_bg = cv.dilate(opening,kernel,iterations=3)
plt.figure()
plt.imshow(sure_bg, cmap='gray')
plt.title("sure_bg")
'''
# Finding sure foreground area
dist_transform = cv.distanceTransform(opening, cv.DIST_L2, 5)
ret, sure_fg = cv.threshold(dist_transform, 0.7 * dist_transform.max(),
255, 0)
plt.figure()
plt.imshow(sure_fg, cmap='gray')
plt.title("sure_fg")
# Finding unknown region
sure_fg = np.uint8(sure_fg)
unknown = cv.subtract(sure_bg, sure_fg)
# Marker labelling
ret, markers = cv.connectedComponents(sure_fg)
# Add one to all labels so that sure background is not 0, but 1
markers = markers + 1
# Now, mark the region of unknown with zero
markers[unknown == 255] = 0
markers = cv.watershed(img, markers)
img[markers == -1] = [255, 0, 0]
plt.figure()
plt.imshow(markers, cmap='gray')
plt.title("Markers")
def spatter(x, severity=1):
c = [(0.65, 0.3, 4, 0.69, 0.6, 0), (0.65, 0.3, 3, 0.68, 0.6, 0),
(0.65, 0.3, 2, 0.68, 0.5, 0), (0.65, 0.3, 1, 0.65, 1.5, 1),
(0.67, 0.4, 1, 0.65, 1.5, 1)][severity - 1]
x_PIL = x
x = np.array(x, dtype=np.float32) / 255.
liquid_layer = np.random.normal(size=x.shape[:2], loc=c[0], scale=c[1])
liquid_layer = gaussian(liquid_layer, sigma=c[2])
liquid_layer[liquid_layer < c[3]] = 0
if c[5] == 0:
liquid_layer = (liquid_layer * 255).astype(np.uint8)
dist = 255 - cv2.Canny(liquid_layer, 50, 150)
dist = cv2.distanceTransform(dist, cv2.DIST_L2, 5)
_, dist = cv2.threshold(dist, 20, 20, cv2.THRESH_TRUNC)
dist = cv2.blur(dist, (3, 3)).astype(np.uint8)
dist = cv2.equalizeHist(dist)
ker = np.array([[-2, -1, 0], [-1, 1, 1], [0, 1, 2]])
dist = cv2.filter2D(dist, cv2.CV_8U, ker)
dist = cv2.blur(dist, (3, 3)).astype(np.float32)
m = cv2.cvtColor(liquid_layer * dist, cv2.COLOR_GRAY2BGRA)
m /= np.max(m, axis=(0, 1))
m *= c[4]
# water is pale turqouise
color = np.concatenate(
(175 / 255. * np.ones_like(m[..., :1]), 238 / 255. *
np.ones_like(m[..., :1]), 238 / 255. * np.ones_like(m[..., :1])),
axis=2)
color = cv2.cvtColor(color, cv2.COLOR_BGR2BGRA)
if len(x.shape) < 3 or x.shape[2] < 3:
add_spatter_color = cv2.cvtColor(np.clip(m * color, 0, 1),
cv2.COLOR_BGRA2BGR)
add_spatter_gray = rgb2gray(add_spatter_color)
return np.clip(x + add_spatter_gray, 0, 1) * 255
else:
x = cv2.cvtColor(x, cv2.COLOR_BGR2BGRA)
return cv2.cvtColor(np.clip(x + m * color, 0, 1),
cv2.COLOR_BGRA2BGR) * 255
else:
m = np.where(liquid_layer > c[3], 1, 0)
m = gaussian(m.astype(np.float32), sigma=c[4])
m[m < 0.8] = 0
x_rgb = np.array(x_PIL.convert('RGB'))
# mud brown
color = np.concatenate((63 / 255. * np.ones_like(x_rgb[..., :1]),
42 / 255. * np.ones_like(x_rgb[..., :1]),
20 / 255. * np.ones_like(x_rgb[..., :1])),
axis=2)
color *= m[..., np.newaxis]
if len(x.shape) < 3 or x.shape[2] < 3:
x *= (1 - m)
return np.clip(x + rgb2gray(color), 0, 1) * 255
else:
x *= (1 - m[..., np.newaxis])
return np.clip(x + color, 0, 1) * 255
def find_exits(floor_mask,
max_height_image,
m_per_pix,
min_robot_width_m,
robot_x_pix,
robot_y_pix,
display_on=False):
# A mask with optimistic traversable pixels consisting of floor
# and unobserved pixels. This should result in traversable paths
# that move from observed floor to unobserved floor, which can be
# used to find exits.
traversable_selector = (floor_mask > 0) | (max_height_image == 0)
traversable_mask = 255 * np.uint8(traversable_selector)
# Fill in small non-floor regions (likely noise)
# TODO: improve this. For example, fill in isolated pixels that
# are too small to trust as obstacles. Options include a hit or
# miss filter, matched filter, or speckle filter.
fill_in = True
if fill_in:
kernel = np.ones((3, 3), np.uint8)
traversable_mask = cv2.morphologyEx(traversable_mask, cv2.MORPH_CLOSE,
kernel)
# Optimistic estimate of robot width. Use ceil to account for
# possible quantization. Consider adding a pixel, also.
min_robot_radius_pix = np.ceil((min_robot_width_m / 2.0) / m_per_pix)
# compute distance map: distance from untraversable regions
#
# cv2.DIST_L2 : Euclidean distance
#
# 5 is the mask size : "finds the shortest path to the nearest
# zero pixel consisting of basic shifts: horizontal, vertical,
# diagonal, or knight's move (the latest is available for a 5x5
# mask)" - OpenCV documentation
distance_map = cv2.distanceTransform(traversable_mask, cv2.DIST_L2, 5)
# fill in floor mask holes
#kernel = np.ones((11,11), np.uint8)
kernel_width_pix = 11
kernel_radius_pix = (kernel_width_pix - 1) / 2
kernel = np.zeros((kernel_width_pix, kernel_width_pix), np.uint8)
cv2.circle(kernel, (kernel_radius_pix, kernel_radius_pix),
kernel_radius_pix, 255, -1)
closed_floor_mask = cv2.morphologyEx(floor_mask, cv2.MORPH_CLOSE, kernel)
# find the boundary of the floor
dilated_floor = cv2.dilate(closed_floor_mask, kernel, iterations=1)
floor_boundary = dilated_floor - closed_floor_mask
# Estimate the locations of exits. Creates a binary image with exits marked with 255.
min_connectivity_dist_pix = min_robot_radius_pix #(min_connectivity_dist_mm / mm_per_pix) - 2.0
connectivity_image = distance_map.copy()
# remove regions that are too narrow for the robot to drive through
connectivity_image[connectivity_image < min_connectivity_dist_pix] = 0
# Only keep pixels that are at the boundary of the observed
# floor. This should cut between the observed and unobserved
# traversible regions.
connectivity_image[floor_boundary == 0] = 0
# convert float image to uint8 image
map_exits = np.zeros_like(connectivity_image, dtype=np.uint8)
# exit pixels have a value of 255
map_exits[connectivity_image > 0] = 255
# enlarge the map exit pixels so that they will intersect with the floor
exit_dilation = True
if exit_dilation:
# attempt to increase the chance of a vertex being labeled as an exit
# create kernel for morphological operations
kernel_width_pix = 11
kernel_radius_pix = (kernel_width_pix - 1) / 2
kernel = np.zeros((kernel_width_pix, kernel_width_pix), np.uint8)
cv2.circle(kernel, (kernel_radius_pix, kernel_radius_pix),
kernel_radius_pix, 255, -1)
map_exits = cv2.dilate(map_exits, kernel, iterations=1)
# Select the connected component of the floor on which the robot
# is located.
h, w = closed_floor_mask.shape
accessible_floor_mask = np.zeros((h + 2, w + 2), np.uint8)
#possibly add to floodFill in the future: flags = cv2.FLOODFILL_FIXED_RANGE
cv2.floodFill(floor_mask, accessible_floor_mask,
(robot_x_pix, robot_y_pix), 255)
accessible_floor_mask = 255 * accessible_floor_mask[1:-1, 1:-1]
# only consider map exit candidates that are connected to parts of
# the floor that can be reached
map_exits[accessible_floor_mask == 0] = 0
# Ignore exits that are very close to the robot. These can be due
# to the robot's own body occluding the floor. Otherwise, they can
# represent real exits that the robot is already next to and does
# not need to navigate to.
ignore_radius_pix = int(4.0 * min_robot_radius_pix)
cv2.circle(map_exits, (robot_x_pix, robot_y_pix), ignore_radius_pix, 0, -1)
# Dilate exits in order to merge exits that are very close to one
# another.
kernel_width_pix = 21 #11 #15
kernel_radius_pix = (kernel_width_pix - 1) / 2
kernel = np.zeros((kernel_width_pix, kernel_width_pix), np.uint8)
cv2.circle(kernel, (kernel_radius_pix, kernel_radius_pix),
kernel_radius_pix, 255, -1)
map_exits = cv2.dilate(map_exits, kernel, iterations=1)
# only consider map exit candidates that are connected to parts of
# the floor that can be reached
map_exits[accessible_floor_mask == 0] = 0
# find exit regions
number_of_exits, exit_label_image = simple_connected_components(
map_exits, connectivity=8)
# find geometric centers of the exits
label_indices = range(number_of_exits)[1:]
label_image = exit_label_image
centers_of_mass = nd.measurements.center_of_mass(label_image, label_image,
label_indices)
ones = np.ones_like(label_image)
sums = nd.measurements.sum(ones, label_image, label_indices)
print('centers_of_mass =', centers_of_mass)
if display_on:
print('find_exits: number_of_exits =', number_of_exits)
h, w = max_height_image.shape
color_im = np.zeros((h, w, 3), np.uint8)
color_im[:, :, 0] = max_height_image
color_im[:, :, 1] = max_height_image
color_im[:, :, 2] = map_exits
for s, (c_y, c_x) in zip(sums, centers_of_mass):
if s > 50:
c_x = int(round(c_x))
c_y = int(round(c_y))
radius = 5 #3
cv2.circle(color_im, (c_x, c_y), radius, [255, 255, 255], -1)
scale_divisor = 2
nh = h / scale_divisor
nw = w / scale_divisor
color_im = cv2.resize(color_im, (nw, nh))
cv2.imshow('find_exits: exits on the map', color_im)
map_exits_mask = map_exits
min_area = 50
exit_points = [[int(round(c_x)), int(round(c_y))]
for s, (c_y, c_x) in zip(sums, centers_of_mass)
if s > min_area]
return exit_points, map_exits_mask, number_of_exits, exit_label_image
# BUT a 8bits unsigned int (the one we are working with) can contain values from 0 to 255
# so the possible negative number will be truncated
imgLaplacian = cv.filter2D(src, cv.CV_32F, kernel)
sharp = np.float32(src)
imgResult = sharp - imgLaplacian
# convert back to 8bits gray scale
imgResult = np.clip(imgResult, 0, 255)
imgResult = imgResult.astype('uint8')
imgLaplacian = np.clip(imgLaplacian, 0, 255)
imgLaplacian = np.uint8(imgLaplacian)
#cv.imshow('Laplace Filtered Image', imgLaplacian)
cv.imshow('New Sharped Image', imgResult)
bw = cv.cvtColor(imgResult, cv.COLOR_BGR2GRAY)
_, bw = cv.threshold(bw, 40, 255, cv.THRESH_BINARY | cv.THRESH_OTSU)
cv.imshow('Binary Image', bw)
dist = cv.distanceTransform(bw, cv.DIST_L2, 3)
# Normalize the distance image for range = {0.0, 1.0}
# so we can visualize and threshold it
cv.normalize(dist, dist, 0, 1.0, cv.NORM_MINMAX)
cv.imshow('Distance Transform Image', dist)
_, dist = cv.threshold(dist, 0.4, 1.0, cv.THRESH_BINARY)
# Dilate a bit the dist image
kernel1 = np.ones((3, 3), dtype=np.uint8)
dist = cv.dilate(dist, kernel1)
cv.imshow('Peaks', dist)
dist_8u = dist.astype('uint8')
# Find total markers
_, contours = cv.findContours(dist_8u, cv.RETR_EXTERNAL,
cv.CHAIN_APPROX_SIMPLE)
# Create the marker image for the watershed algorithm
markers = np.zeros(dist.shape, dtype=np.int32)
def incomingDepthData(self, data):
self.started()
t0 = rospy.Time.now()
img = self.ros2cv(data, dt="passthrough")
img = np.asarray(img, dtype=np.float32)
img /= self.maxDepth * 1000 # convert to 0-1 units, 1=maxDepth
img[np.bitwise_not(np.isfinite(img))] = 0
img[img < 0.01] = 1
img = np.clip(img, 0.0, 1.0)
img = img.reshape((480, 640))
t1 = rospy.Time.now()
# Only draw images if we actually display them
showing = self.pub_depth.get_num_connections(
) > 0 or self.pub_camera.get_num_connections() > 0
# Dont do anything until we first create a background
if self.counterSteps > 0:
# gather background
if self.counter < self.counterSteps:
self.sig_backgroundStep.emit(self.counter, self.counterSteps)
img[img < 0.001] = 1
self.acc[:, :, self.counter] = img
if showing:
show = np.asarray(cv2.cvtColor(img * 255,
cv2.COLOR_GRAY2BGR),
dtype=np.uint8)
cv2.putText(show, str(self.counterSteps - self.counter),
(5, 30), cv2.FONT_HERSHEY_PLAIN, 2,
(255, 0, 0))
self.counter += 1
# create background
if self.counter == self.counterSteps:
if self.method == 0: # Median
med = np.median(self.acc, 2)
max = np.max(self.acc, 2)
std = np.std(self.acc, 2)
self.depth = max
for i in range(self.counterSteps):
local = self.acc[:, :, i]
mask = cv2.absdiff(med, local) < std
new_min = cv2.min(self.depth, local)
self.depth[mask] = new_min[mask]
else:
min = np.min(self.acc, 2)
self.depth = min
rospy.loginfo("Built background")
self.sig_backgroundStep.emit(self.counter, self.counterSteps)
self.sig_background.emit(True)
self.counter += 1
# extract foreground
if self.counter > self.counterSteps:
t2 = rospy.Time.now()
img = np.clip(img, 0.0, 1.0)
# Our BW image
img = cv2.min(self.depth, img)
# Binary image of close objects
d = np.subtract(self.depth, img)
d = cv2.morphologyEx(d,
cv2.MORPH_OPEN,
self.kernel,
iterations=self.morphIterations)
bm = cv2.threshold(d, self.bg_thresh / self.maxDepth, 255,
cv2.THRESH_BINARY)[1]
bm = np.asarray(bm, dtype=np.uint8)
dsts = cv2.distanceTransform(bm, cv2.cv.CV_DIST_L2, 3)
t3 = rospy.Time.now()
if showing:
pass
show = np.asarray(cv2.cvtColor(img * 255,
cv2.COLOR_GRAY2BGR),
dtype=np.uint8)
show[:, :, 2] = cv2.max(bm, show[:, :, 2])
t4 = rospy.Time.now()
# Detect possible flies
flies = []
while (True):
# Look for biggest blob
mn, dist, mnl, mxl = cv2.minMaxLoc(dsts, mask=bm)
if dist < 3:
break
# Extract blob info
h, w = img.shape[:2]
mask = np.zeros((h + 2, w + 2), np.uint8)
retval, rect = cv2.floodFill(bm,
mask,
mxl,
0,
flags=8
| cv2.FLOODFILL_FIXED_RANGE
| cv2.FLOODFILL_MASK_ONLY)
# Get pose %TODO should compute average over whole blob
if self.depthEstMode == 0:
# Center Pixel Depth
x, y, z = self.projectRay(
mxl[0], mxl[1],
self.maxDepth * img[mxl[1], mxl[0]])
else:
x, y, z = self.projectRay(
mxl[0], mxl[1],
self.maxDepth * img[mxl[1], mxl[0]])
rospy.logerr(
"Not implemented depth estimation mode [%d], defaulting to center depth",
self.depthEstMode)
#elif self.depthEstMode == 1:
# # Median depth of blob
#elif self.depthEstMode == 2:
# # Closest point of blob
# Estimate width
xD, yD, zD = self.projectRay(
mxl[0] + dist, mxl[1] + dist,
self.maxDepth * img[mxl[1], mxl[0]])
w = math.sqrt((x - xD)**2 + (y - yD)**2) * 2
# Estimatea distance from background
b_diff = cv2.mean(d, mask=mask[1:-1,
1:-1])[0] * self.maxDepth
if abs((w - self.estSize
)) > self.estSizeTol: #if w<0.02 or w>0.10:
if showing:
cv2.circle(show, mxl, 1, (0, 200, 0))
cv2.circle(show, mxl, int(dist), (120, 0, 0), 1)
#cv2.putText(show, str(round(w*100, 1))+"cm", mxl, cv2.FONT_HERSHEY_PLAIN, 1.2, (255,0,0))
else:
flies.append([dist, mxl, x, y, z, w, b_diff])
bm -= mask[1:-1, 1:-1] * 255
t5 = rospy.Time.now()
if len(flies) > 0:
# Sort by distance from background#
#TODO
if self.priority == 0:
# Distance from background
flies = sorted(flies, key=itemgetter(6), reverse=True)
elif self.priority == 1:
# Closest to goal
self.getCameraGoal()
flies = sorted([
f + [(f[2] - self.goal[0])**2 +
(f[3] - self.goal[1])**2 +
(f[4] - self.goal[2])**2] for f in flies
],
key=itemgetter(7),
reverse=False)
elif self.priority == 2:
# Depth
flies = sorted(flies, key=itemgetter(4), reverse=False)
# Closest to GT
#gt = self.getGTDist()
#flies = sorted([f+[(f[2]-gt[0])**2+(f[3]-gt[1])**2+(f[4]-gt[2])**2] for f in flies], key=itemgetter(7),reverse=False)
elif self.priority == 3:
# Estimated object size
flies = sorted(
[f + [abs(self.estSize - f[5])] for f in flies],
key=itemgetter(7),
reverse=False)
elif self.priority == 4:
# Blob Size
flies = sorted(flies, key=itemgetter(0), reverse=True)
else:
rospy.logerr("Unknown priority type: %d",
self.priority)
if len(flies) > 0:
dist, mxl, x, y, z, w, b_diff = flies[0][0:7]
# publish with level rotation
t, r, = self.getKinectPoint()
#self.pub_tf.sendTransform([z,-x-0.02,-y], (0,0,0,1), rospy.Time.now(), "cf_xyz", "camera_depth_frame")#TODO maybe we need to rotate 90" so x is aligned with optical axis
#self.pub_tf.sendTransform([x,y,z], (-r[0],-r[1],-r[2],r[3]), rospy.Time.now(), "cf_xyz", "camera_depth_optical_frame")#TODO maybe we need to rotate 90" so x is aligned with optical axis
# Point in world frame with R from crayzflie
self.pub_tf.sendTransform(
[x, y, z], quaternion_inverse(r), rospy.Time.now(),
"cf_xyz", "camera_depth_optical_frame"
) #TODO maybe we need to rotate 90" so x is aligned with optical axis
t6 = rospy.Time.now()
if showing:
pass
for i, flie in enumerate(flies):
dist, mxl, x, y, z, w, b_diff = flies[i][0:7]
if i == 0:
cv2.circle(show, mxl, int(dist), (0, 255, 0),
4)
else:
cv2.circle(show, mxl, int(dist), (12, 50, 12),
1)
#cv2.circle(show, mxl, 2, (0, 255, 0))
#cv2.circle(show, mxl, int(b_diff*10), (255, 0, 0))
#cv2.putText(show, str(i), (mxl[0]-30, mxl[1]), cv2.FONT_HERSHEY_PLAIN, 1.4, (0, 255, 255))
#cv2.putText(show, str(round(z, 2))+"m", (mxl[0]+14, mxl[1]+00), cv2.FONT_HERSHEY_PLAIN, 1.4, (255, 255, 0))
#cv2.putText(show, str(round(w*100, 1))+"cm", (mxl[0]+14, mxl[1]+20), cv2.FONT_HERSHEY_PLAIN, 1.4, (255,0,255))
#cv2.putText(show, str(round(b_diff,2))+"m", (mxl[0]+14, mxl[1]+40), cv2.FONT_HERSHEY_PLAIN, 1.4, (255, 0, 0))
t7 = rospy.Time.now()
rospy.loginfo(
"%7.1f %7.1f %7.1f %7.1f %7.1f %7.1f = %10.1f ",
1000. * (t1 - t0).to_sec(), 1000. * (t2 - t1).to_sec(),
1000. * (t3 - t2).to_sec(), 1000. * (t4 - t3).to_sec(),
1000. * (t5 - t4).to_sec(), 1000. * (t7 - t6).to_sec(),
1000 * (t7 - t0).to_sec())
else:
show = np.asarray(cv2.cvtColor(img * 255, cv2.COLOR_GRAY2BGR),
dtype=np.uint8)
if showing:
pass
try:
img = self.bridge.cv2_to_imgmsg(show, "bgr8")
self.cameraInfo.header.stamp = rospy.Time.now()
img.header = self.cameraInfo.header
self.pub_camera.publish(self.cameraInfo)
self.pub_depth.publish(img)
except CvBridgeError, e:
rospy.logerr("Image sending problem: %s", e)
k = 1
if k % 2 == 0:
k += 1
b1 = b1 / 100
okern = cv2.getStructuringElement(cv2.MORPH_RECT, (k, k))
ekern = cv2.getStructuringElement(cv2.MORPH_RECT, (ek, ek))
dkern = cv2.getStructuringElement(cv2.MORPH_RECT, (dk, dk))
blurimg = cv2.GaussianBlur(img, (blur, blur), 0)
noise = cv2.morphologyEx(blurimg, cv2.MORPH_OPEN, okern, iterations=oi)
imggray = cv2.cvtColor(noise, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(imggray, 0, 255,
cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
back = cv2.dilate(thresh, dkern, iterations=di)
#fore=cv2.erode(thresh,ekern,iterations=ei)
dist_transform = cv2.distanceTransform(thresh, cv2.DIST_L2, a1)
ret, fore = cv2.threshold(dist_transform, b1 * dist_transform.max(), 255,
0)
dist_transform = cv2.normalize(dist_transform, dist_transform, 0, 255,
cv2.NORM_MINMAX)
dist_transform = np.uint8(dist_transform)
cv2.imshow('dist_transform', dist_transform)
fore = np.uint8(fore)
unknown = cv2.subtract(back, fore)
ret, markers = cv2.connectedComponents(fore)
markers += 1
markers[unknown == 255] = 0
markers = cv2.watershed(img, markers)
img[markers == -1] = [255, 0, 0]
cv2.imshow('fore', fore)
cv2.imshow('image', img)
def threshold_callback(params):
global img_gray, thresh, gamma, img_gamma
thresh = params
#blur grey 3*3
img_gray = cv2.medianBlur(img_gray, 5)
# img_gray = cv2.GaussianBlur(img_gray,(5,5), 0)
# ret, thresh_out = cv2.threshold(img_gray, thresh, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
# img_gray = adjust_gamma(img_gray, gamma)
# cv2.imshow('gamma', img_gray)
thresh_out = cv2.adaptiveThreshold(img_gray, 255,
cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY, 11, 2)
# thresh_out = cv2.adaptiveThreshold(img_gray, 255, \
# cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY,11,2)
# cv2.imshow('thresh out', thresh_out)
# noise removal
kernel = np.ones((3, 3), np.uint8)
opening = cv2.morphologyEx(thresh_out,
cv2.MORPH_GRADIENT,
kernel,
iterations=5)
#opening = cv2.morphologyEx(thresh_out, cv2.MORPH_OPEN, kernel, iterations = 2)
#opening = cv2.morphologyEx(thresh_out, cv2.MORPH_CLOSE, kernel, iterations = 2)
#opening = cv2.morphologyEx(thresh_out, cv2.MORPH_CLOSE + cv2.MORPH_OPEN, kernel, iterations = 2)
# cv2.imshow('opening', opening)
# sure background area
sure_bg = cv2.dilate(opening, kernel, iterations=5)
# cv2.imshow('sure_bg', sure_bg)
dist = cv2.distanceTransform(opening, cv2.DIST_L2, 3)
cv2.normalize(dist, dist, 0, 255, cv2.NORM_MINMAX)
dist = np.uint8(dist)
# cv2.imshow('dist', dist)
# ret, sure_fg = cv2.threshold(dist, dist.max()*0.7, 255, cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
sure_fg = cv2.adaptiveThreshold(dist, 255, \
cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
# cv2.imshow('sure_fg', sure_fg)
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_bg, sure_fg)
# cv2.imshow('unknown', unknown)
#unknown = cv2.subtract(unknown, thresh_out)
#cv2.imshow('unknown', unknown)
# laplacian = cv2.Laplacian(unknown,cv2.CV_64F)
# cv2.imshow('laplacian', laplacian)
# circles = cv2.HoughCircles(unknown ,cv2.HOUGH_GRADIENT,1,20,
# param1=50,param2=30,minRadius=0,maxRadius=0)
# im2, contours, hierarchy = cv2.findContours(dist, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
# im2, contours, hierarchy = cv2.findContours(unknown, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
im2, contours, hierarchy = cv2.findContours(unknown, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_SIMPLE)
# hull = []
minRect = []
# minEllipse = []
drawing = np.zeros(thresh_out.shape, np.uint8)
# drawing = np.zeros(dist.size, np.uint8)
# for i in range(len(contours)):
# hull.append(cv2.convexHull(contours[i]))
hsv_drawing = hsv.copy()
ret = []
for i in range(len(contours)):
minRect.append(cv2.minAreaRect(contours[i]))
# if len(contours[i]) < 5:
# minRect.append(cv2.minAreaRect(contours[i]))
# minEllipse.append(cv2.fitEllipse(contours[i]))
# cv2.ellipse(drawing, minEllipse[i], (255,0,0), 2)
# cv2.drawContours(drawing, contours, int(i), (0,255,0), 0)
# cv2.drawContours(drawing, contours, 0, (0,0,255), 1)
for i in range(len(contours)):
# if contours[i].size > 100:
# cv2.drawContours(drawing, contours, i, (0,255,0), 1)
# cv2.drawContours(drawing, hull, i, (0,255,0), 0)
box = cv2.boxPoints(minRect[i])
box = np.int0(box)
if box[1][1] < box[3][1] and box[0][0] < box[2][0]:
img_roi = hsv[int(box[1][1]):int(box[3][1]),
int(box[0][0]):int(box[2][0])]
img_rgb_roi = cv2.cvtColor(img_roi, cv2.COLOR_HSV2BGR)
img_roi = cv2.cvtColor(img_roi, cv2.COLOR_BGR2GRAY)
#cv2.imshow('ROI', img_roi)
# create a CLAHE object (Arguments are optional).
clahe = cv2.createCLAHE(clipLimit=1.0, tileGridSize=(3, 3))
cl1 = clahe.apply(img_roi)
# res = np.hstack((img_roi, cl1))
# equ = cv2.equalizeHist(img_roi)
# res = np.hstack((img_roi, img_roi))
# cv2.imshow('equ', res)
# lower_blue = np.array([110,50,50])
# upper_blue = np.array([130,255,255])
# mask = cv2.inRange(img_roi, lower_blue, upper_blue)
# res = cv2.bitwise_and(img_roi,img_roi, mask= mask)
circles = cv2.HoughCircles(cl1,
cv2.HOUGH_GRADIENT,
1,
1,
param1=80,
param2=20,
minRadius=0,
maxRadius=0)
# circles = cv2.HoughCircles(img_roi, cv2.HOUGH_GRADIENT, 1, 1, param1=80, param2=20, minRadius=0, maxRadius=0)
if circles != None:
for i in circles[0, :]:
# draw the outer circle
# cv2.imshow('ROI', img_roi)
cv2.circle(hsv_drawing,
(int(box[1][0] + i[0]), int(box[1][1] + i[1])),
i[2], (0, 255, 0), 2)
# draw the center of the circle
# cv2.circle(hsv,(i[0],i[1]),2,(0,0,255),3)
cv2.circle(hsv_drawing,
(int(box[1][0] + i[0]), int(box[1][1] + i[1])),
2, (0, 0, 255), 3)
# cv2.imshow('circle', hsv_drawing)
# cv2.imshow('img_rgb_roi', img_rgb_roi)
# create a water index pixel mask
w = img_rgb_roi[0, 0]
# print w
b, g, r = cv2.split(img_rgb_roi)
mask = np.zeros(img_rgb_roi.shape[:2], np.uint8)
mask[:, :] = w[0]
#mask_inv = cv2.bitwise_not(mask)
# cv2.imshow('b_mask',mask)
b_masked_img = cv2.subtract(b, mask)
# cv2.imshow('b_masked_img', b_masked_img)
b_histr, bins = np.histogram(b_masked_img.ravel(), 256,
[0, 256])
b_cdf = b_histr.cumsum()
# b_cdf_normalized = b_cdf * b_histr.max()/ b_cdf.max()
mask[:, :] = w[1]
#mask_inv = cv2.bitwise_not(mask)
#cv2.imshow('g_mask',mask)
g_masked_img = cv2.subtract(g, mask)
# cv2.imshow('g_masked_img', g_masked_img)
g_histr, bins = np.histogram(g_masked_img.ravel(), 256,
[0, 256])
g_cdf = g_histr.cumsum()
# g_cdf_normalized = g_cdf * g_histr.max()/ g_cdf.max()
mask[:, :] = w[2]
#mask_inv = cv2.bitwise_not(mask)
#cv2.imshow('r_mask',mask)
r_masked_img = cv2.subtract(r, mask)
# cv2.imshow('r_masked_img', r_masked_img)
r_histr, bins = np.histogram(r_masked_img.ravel(), 256,
[0, 256])
r_cdf = r_histr.cumsum()
# r_cdf_normalized = r_cdf * r_histr.max()/ r_cdf.max()
# m_img = cv2.merge((b_masked_img,g_masked_img,r_masked_img))
# cv2.imshow('m_img', m_img)
# b_histr = cv2.calcHist([img_rgb_roi],[0],None,[256],[0,256])
# g_histr = cv2.calcHist([img_rgb_roi],[1],None,[256],[0,256])
# r_histr = cv2.calcHist([img_rgb_roi],[2],None,[256],[0,256])
total = (b_cdf.max() - b_histr[0]) + (
g_cdf.max() - g_histr[0]) + (r_cdf.max() - r_histr[0])
Pb = (float(b_cdf.max() - b_histr[0]) / float(total))
Pg = (float(g_cdf.max() - g_histr[0]) / float(total))
Pr = (float(r_cdf.max() - r_histr[0]) / float(total))
if total == 0:
Pb = 1.0
Pg = 1.0
Pr = 1.0
else:
Pb = (float(b_cdf.max() - b_histr[0]) / float(total))
Pg = (float(g_cdf.max() - g_histr[0]) / float(total))
Pr = (float(r_cdf.max() - r_histr[0]) / float(total))
if abs(Pg - Pr) < 0.2:
color = 'y'
elif Pg > Pb and Pg > Pr:
color = 'g'
elif Pr > Pb and Pr > Pg:
color = 'r'
#print "Pb{}Pg{}Pr{}".format(Pb, Pg, Pr)
# hist,bins = np.histogram(img_roi.flatten(), 256,[0,256])
# cdf = hist.cumsum()
# cdf_normalized = cdf * hist.max()/ cdf.max()
# plt.plot(cdf_normalized, color = 'b')
# plt.hist(img.flatqten(),256,[0,256], color = 'r')
# plt.xlim([0,256])
# plt.legend(('cdf','histogram'), loc = 'upper left')
# plt.show()
# return (int(box[1][0]+i[0]), int(box[1][1]+i[1]), i[2], b_histr.max(), #g_histr.max(), r_histr.max())
#return (int(box[1][0]+i[0]), int(box[1][1]+i[1]), int(i[2]), Pb, Pg, Pr, color)
ret.append({"width":img_gray.shape[1], "height":img_gray.shape[0], "origin_x":int(box[1][0]+i[0]), \
"origin_y":int(box[1][1]+i[1]), "radius":int(i[2]), "prob_blue":Pb, "prob_green":Pg, "prob_red":Pr, "color":color})
#buoy = "window_x={7},window_y={8},origin_x={0},origin_y={1},radius={2},prob_blue={3},prob_green={4},prob_red={5},color={6}" \
# .format(ret[i][0], ret[i][1], ret[i][2], ret[i][3], ret[i][4], ret[i][5], ret[i][6], img_gray.shape[1], img_gray.shape[0])
print("Count all: {}".format(len(ret)))
rospy.loginfo(ret)
return ret
total = len(img_files)
for count, i in enumerate(img_files):
image_name = i.split("/")[-1]
print("Progress : ", count, "/", total)
img = cv2.imread(i)
# Split the 3 channels into Blue,Green and Red
b, g, r = cv2.split(img)
# Apply Basic Transformation
b = basicTransform(b)
r = basicTransform(r)
g = basicTransform(g)
# Perform the distance transform algorithm
b = cv2.distanceTransform(b, cv2.DIST_L2, 5) # ELCUDIAN
g = cv2.distanceTransform(g, cv2.DIST_L1, 5) # LINEAR
r = cv2.distanceTransform(r, cv2.DIST_C, 5) # MAX
# Normalize
r = cv2.normalize(r, r, 0, 1.0, cv2.NORM_MINMAX)
g = cv2.normalize(g, g, 0, 1.0, cv2.NORM_MINMAX)
b = cv2.normalize(b, b, 0, 1.0, cv2.NORM_MINMAX)
# Merge the channels
dist = cv2.merge((b, g, r))
dist = cv2.normalize(dist, dist, 0, 4.0, cv2.NORM_MINMAX)
dist = cv2.cvtColor(dist, cv2.COLOR_BGR2GRAY)
# In order to save as jpg, or png, we need to handle the Data
# format of image
def image_detect(filename):
from PIL import Image
import numpy as np
import cv2
from matplotlib import pyplot as plt
from PIL import Image
from skimage.morphology import extrema
from skimage.morphology import watershed as skwater
print(filename)
#canny starts
def auto_canny(image, sigma=0.33):
# compute the median of the single channel pixel intensities
v = np.median(image)
# apply automatic Canny edge detection using the computed median
lower = int(max(0, (1.0 - sigma) * v))
upper = int(min(255, (1.0 + sigma) * v))
edged = cv2.Canny(image, lower, upper)
# return the edged image
return edged
def ShowImage(title, img, ctype):
plt.figure(figsize=(10, 10))
if ctype == 'bgr':
b, g, r = cv2.split(img) # get b,g,r
rgb_img = cv2.merge([r, g, b]) # switch it to rgb
plt.imshow(rgb_img)
elif ctype == 'hsv':
rgb = cv2.cvtColor(img, cv2.COLOR_HSV2RGB)
plt.imshow(rgb)
elif ctype == 'gray':
plt.imshow(img, cmap='gray')
elif ctype == 'rgb':
plt.imshow(img)
else:
raise Exception("Unknown colour type")
plt.axis('off')
plt.title(title)
plt.show()
img = cv2.imread(filename)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
#ShowImage('Brain MRI',gray,'gray')
ret, thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_OTSU)
#ShowImage('Thresholding image',thresh,'gray')
ret, markers = cv2.connectedComponents(thresh)
#Get the area taken by each component. Ignore label 0 since this is the background.
marker_area = [
np.sum(markers == m) for m in range(np.max(markers)) if m != 0
]
#Get label of largest component by area
largest_component = np.argmax(
marker_area) + 1 #Add 1 since we dropped zero above
#Get pixels which correspond to the brain
brain_mask = markers == largest_component
brain_out = img.copy()
#In a copy of the original image, clear those pixels that don't correspond to the brain
brain_out[brain_mask == False] = (0, 0, 0)
img = cv2.imread(filename)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(gray, 0, 255,
cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
#canny-edge-detection
auto = auto_canny(thresh)
#cv2.imshow("canny",auto)
cv2.waitKey(0)
resu = auto + thresh
#cv2.imshow("res=canny+thresh",resu)
cv2.waitKey(0)
#cv2.imshow("invert res",cv2.bitwise_not(resu))
cv2.waitKey(0)
# noise removal
kernel = np.ones((3, 3), np.uint8)
opening = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel, iterations=2)
# sure background area
sure_bg = cv2.dilate(opening, kernel, iterations=3)
# Finding sure foreground area
dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 5)
ret, sure_fg = cv2.threshold(dist_transform, 0.7 * dist_transform.max(),
255, 0)
# Finding unknown region
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_bg, sure_fg)
# Marker labelling
ret, markers = cv2.connectedComponents(sure_fg)
# Add one to all labels so that sure background is not 0, but 1
markers = markers + 1
# Now, mark the region of unknown with zero
markers[unknown == 255] = 0
markers = cv2.watershed(img, markers)
img[markers == -1] = [255, 0, 0]
im1 = cv2.cvtColor(img, cv2.COLOR_HSV2RGB)
ShowImage('Watershed segmented image', im1, 'gray')
cv2.imwrite('./static/watershed.jpg', im1,
[int(cv2.IMWRITE_JPEG_QUALITY), 100])
brain_mask = np.uint8(brain_mask)
kernel = np.ones((8, 8), np.uint8)
closing = cv2.morphologyEx(brain_mask, cv2.MORPH_CLOSE, kernel)
#ShowImage('Closing', closing, 'gray')
brain_out = img.copy()
#In a copy of the original image, clear those pixels that don't correspond to the brain
brain_out[closing == False] = (0, 0, 0)
#img_path="./static/WhatsApp_Image_2020-10-14_at_19.35.58.jpeg"
#image_detect(img_path)
dx, dy = 0, 0
new_rect = (rect[0] + x - dx, rect[1] + y - dy, rect[2] + 2 * dx,
rect[3] + 2 * dy)
new_rects.append(new_rect)
segment_rects = new_rects
return segment_rects, np.dstack(3 * [display])
if __name__ == "__main__":
image = cv2.imread("../../data/drawer.jpg")
scaled = 1.0
image = cv2.resize(
image, (int(image.shape[1] * scaled), int(image.shape[0] * scaled)))
test_grabcut = False
if test_grabcut:
seeds = [[800, 400], [820, 740], [830, 840], [630, 240], [560, 270]]
rects, display = segment_grabcut(image, seeds=seeds)
display = np.array(display[:, :, 0])
display = cv2.distanceTransform(display, cv2.cv.CV_DIST_L2, 5)
cv2.imshow("disp", (display).astype(np.uint8))
else:
rects, display = segment_edges(image,
window=None,
resize=(5000, 5000),
variance_threshold=100,
size_filter=1)
cv2.imshow("disp", (display).astype(np.uint8))
while cv2.waitKey(0) != 27:
pass
ret, thresh = cv2.threshold(gray, 0, 255,
cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
#remocao de ruido
kernel = np.ones((2, 2), np.uint8)
closing = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel, iterations=2)
# background certo do frame
sure_bg = cv2.dilate(closing, kernel, iterations=3)
#remocao do background ja encontrado, para deixar apenas o corredor
sure_bg = cv2.absdiff(avg, sure_bg)
sure_bg = cv2.erode(sure_bg, kernel, iterations=5)
#encontrando o plano de frente do video
dist_transform = cv2.distanceTransform(sure_bg, cv2.DIST_L2, 3)
# Threshold
ret, sure_fg = cv2.threshold(dist_transform, 0.1 * dist_transform.max(),
255, 0)
# encontrando a regiao desconhecida da imagem para conectar e realizar o watershed
sure_fg = np.uint8(sure_fg)
unknown = cv2.subtract(sure_bg, sure_fg)
# criacao dos marcadores
ret, markers = cv2.connectedComponents(sure_fg)
markers = markers + 1
markers[unknown == 255] = 0
#aplicando o watershed e pintando o quadro original
def calculatePlantMask(image, toolsize, bilf=[11, 5, 17], morpho_it=[5, 5], debugdir=None):
ExG = calculateExcessGreen(image)
M = ExG.max()
m = ExG.min()
# Scale all values to the range (0, 255)
colorIndex = (255 * (ExG - m) / (M - m)).astype(np.uint8)
# Smooth the image using a bilateral filter
colorIndex = cv2.bilateralFilter(colorIndex, bilf[0], bilf[1], bilf[2])
if debugdir:
cv2.imwrite("%s/03-exgnorm.jpg" % debugdir, colorIndex)
# Calculte the mask using Otsu's method (see
# https://docs.opencv.org/3.0-beta/doc/py_tutorials/py_imgproc/py_thresholding/py_thresholding.html)
th, mask = cv2.threshold(colorIndex, 0, 255, cv2.THRESH_OTSU)
if debugdir:
cv2.imwrite("%s/04-mask1.jpg" % debugdir, mask)
if debugdir:
plt.subplot(1, 5, 1), plt.imshow(image)
plt.title("image"), plt.xticks([]), plt.yticks([])
plt.subplot(1, 5, 2), plt.imshow(ExG, 'gray')
plt.title("ExG"), plt.xticks([]), plt.yticks([])
plt.subplot(1, 5, 3), plt.imshow(colorIndex, 'gray')
plt.title("filtered"), plt.xticks([]), plt.yticks([])
plt.subplot(1, 5, 4), plt.hist(colorIndex.ravel(), 256), plt.axvline(x=th, color="red", linewidth=0.1)
plt.title("histo"), plt.xticks([]), plt.yticks([])
plt.subplot(1, 5, 5), plt.imshow(mask, 'gray')
plt.title("mask"), plt.xticks([]), plt.yticks([])
plt.savefig("%s/05-plot.jpg" % debugdir, dpi=300)
# The kernel is a cross:
# 0 1 0
# 1 1 1
# 0 1 0
kernel = np.ones((3, 3)).astype(np.uint8)
kernel[[0, 0, 2, 2], [0, 2, 2, 0]] = 0
# See https://docs.opencv.org/3.0-beta/doc/py_tutorials/py_imgproc/py_morphological_ops/py_morphological_ops.html
mask = cv2.morphologyEx(mask, cv2.MORPH_OPEN, kernel, iterations=morpho_it[0])
if debugdir:
cv2.imwrite("%s/06-mask2.jpg" % debugdir, mask)
mask = cv2.dilate(mask, kernel=kernel, iterations=morpho_it[1])
if debugdir:
cv2.imwrite("%s/07-mask3.jpg" % debugdir, mask)
# Invert the mask and calculate the distance to the closest black pixel.
# See https://docs.opencv.org/2.4.8/modules/imgproc/doc/miscellaneous_transformations.html#distancetransform
dist = cv2.distanceTransform(255 - (mask.astype(np.uint8)),
cv2.DIST_L2,
cv2.DIST_MASK_PRECISE)
# Turn white all the black pixels that are less than half the
# toolsize away from a white (=plant) pixel
mask = 255 * (1 - (dist > toolsize/2)).astype(np.uint8)
if debugdir:
cv2.imwrite("%s/08-mask.jpg" % debugdir, mask)
return mask
def distance_map_simple(floor_mask,
m_per_pix,
min_robot_width_m,
robot_x_pix,
robot_y_pix,
robot_ang_rad,
disallow_too_narrow=True,
display_on=False,
verbose=False):
# min_robot_width_m : The best case minimum width of the robot in meters when moving forward and backward.
traversable_mask = floor_mask
# model the robot's footprint as being traversable
draw_robot_footprint_rectangle(robot_x_pix, robot_y_pix, robot_ang_rad,
m_per_pix, traversable_mask)
footprint_test_image = np.zeros_like(traversable_mask)
draw_robot_footprint_rectangle(robot_x_pix, robot_y_pix, robot_ang_rad,
m_per_pix, footprint_test_image)
if display_on:
cv2.imshow('robot footprint drawing', footprint_test_image)
cv2.imshow('floor mask after drawing robot footprint',
traversable_mask)
# Optimistic estimate of robot width. Use ceil to account for
# possible quantization. Consider adding a pixel, also.
min_robot_radius_pix = np.ceil((min_robot_width_m / 2.0) / m_per_pix)
# Fill in small non-floor regions (likely noise)
#
# TODO: improve this. For example, fill in isolated pixels that
# are too small to trust as obstacles. Options include a hit or
# miss filter, matched filter, or speckle filter.
fill_in = True
if fill_in:
kernel = np.ones((3, 3), np.uint8)
traversable_mask = cv2.morphologyEx(traversable_mask, cv2.MORPH_CLOSE,
kernel)
if display_on:
cv2.imshow('traversable_mask after filling', traversable_mask)
# ERROR? : The floodfill operation should occur after removing
# filtering candidate robot poses due to the footprint
# radius. Right? Was that too aggressive in the past, so it got
# dropped?
# Select the connected component of the floor on which the robot
# is located.
h, w = traversable_mask.shape
new_traversable_mask = np.zeros((h + 2, w + 2), np.uint8)
#possibly add to floodFill in the future: flags = cv2.FLOODFILL_FIXED_RANGE
cv2.floodFill(traversable_mask, new_traversable_mask,
(robot_x_pix, robot_y_pix), 255)
traversable_mask = 255 * new_traversable_mask[1:-1, 1:-1]
# In previous versions, the traversability mask has treated
# unobserved pixels and observed non-floor pixels
# differently. Such as by treating unobserved pixels
# optimistically as traversable.
# compute distance map: distance from untraversable regions
#
# cv2.DIST_L2 : Euclidean distance
#
# 5 is the mask size : "finds the shortest path to the nearest
# zero pixel consisting of basic shifts: horizontal, vertical,
# diagonal, or knight's move (the latest is available for a 5x5
# mask)" - OpenCV documentation
distance_map = cv2.distanceTransform(traversable_mask, cv2.DIST_L2, 5)
if display_on:
norm_dist_transform = cv2.normalize(distance_map, None, 0, 255,
cv2.NORM_MINMAX, cv2.CV_8U)
cv2.imshow('distance map without threshold for the robot width',
norm_dist_transform)
# Restricts the maximum distance of the distance_map. This will
# favor shorter paths (i.e., straight line paths) when a corridor
# is wide enough instead of moving to the middle of the
# corridor. When the corridor is narrower than the threshold, the
# robot will prefer paths that move it to the center of the
# corridor. However, simple path planning via 4 connected grid and
# Dijkstra's algorithm results in vertical and horizontal motions
# in flat regions rather than point-to-point straight lines.
clip_max_distance = False
if clip_max_distance:
max_distance = 3.0 * min_robot_radius_pix
print('max_distance =', max_distance)
print('np.max(distance_map) =', np.max(distance_map))
# should perform in place clipping
np.clip(distance_map, None, max_distance, distance_map)
print('after clipping np.max(distance_map) =', np.max(distance_map))
if display_on:
norm_dist_transform = cv2.normalize(distance_map, None, 0, 255,
cv2.NORM_MINMAX, cv2.CV_8U)
cv2.imshow('distance map with clipped maximum distance',
norm_dist_transform)
if disallow_too_narrow:
# set parts of the distance transform that represent free space
# less than the robot would require to zero
distance_map[distance_map < min_robot_radius_pix] = 0
if display_on:
norm_dist_transform = cv2.normalize(distance_map, None, 0, 255,
cv2.NORM_MINMAX, cv2.CV_8U)
cv2.imshow('distance map with robot width threshold',
norm_dist_transform)
# traversable_mask is a binary image that estimates where the
# robot can navigate given the robot's current pose and the map,
# but ignoring the robot's radius.
# distance_map is a scalar image that estimates the distance to
# the boundaries of the traversable mask.
return distance_map, traversable_mask
