def findBvPeaks(gp1_2d_array):
#bandwidth = 'scott'
kde = stats.kde.gaussian_kde( ( gp1_2d_array.T[0][:], gp1_2d_array.T[1][:] ))#, bw_method = bandwidth )
div = 128j
# Regular grid to evaluate kde upon
x_flat = np.r_[gp1_2d_array[:,0].min():gp1_2d_array[:,0].max():div]
y_flat = np.r_[gp1_2d_array[:,1].min():gp1_2d_array[:,1].max():div]
x,y = np.meshgrid(x_flat,y_flat)
grid_coords = np.append(x.reshape(-1,1),y.reshape(-1,1),axis=1)
z = kde(grid_coords.T)
div_real = 128
z = z.reshape(div_real,div_real)
#plotDistro(kde,(u_min,u_max),(v_min,v_max),title='')
limg = np.arcsinh(z)
limg = limg / limg.max()
low = np.percentile(limg, 0.25)
high = np.percentile(limg, 99.9)
opt_img = skie.exposure.rescale_intensity(limg, in_range=(low,high))
lm = morph.is_local_maximum(limg)
x1, y1 = np.where(lm.T == True)
v = limg[(y1, x1)]
lim = 0.6*high
fp = 13 #smaller odd values are more sensitive to local maxima in image (use roughtly 5 thru 13)
lm = morph.is_local_maximum(limg,footprint=np.ones((fp, fp)))
x1, y1 = np.where(lm.T == True)
v = limg[(y1, x1)]
peaks = [[],[]]
for idx in range(0,len(x1)):
if limg[(y1[idx],x1[idx])] > lim:
peaks[0].append(x1[idx])
peaks[1].append(y1[idx])
fig = plt.figure()
img = plt.imshow(opt_img,origin='lower',interpolation='bicubic')#,cmap=cm.spectral)
ax = fig.add_subplot(111)
ax.scatter(peaks[0],peaks[1], s=100, facecolor='none', edgecolor = '#009999')
#plt.savefig(OUTPUT_DATA_DIR + '_peaks.png')
plt.colorbar()
plt.show()
return peaks
def findBivariatePeaks(kde):
# Regular grid to evaluate kde upon
x_flat = np.r_[start:end:div]
y_flat = np.r_[start:end:div]
x, y = np.meshgrid(x_flat, y_flat)
grid_coords = np.append(x.reshape(-1, 1), y.reshape(-1, 1), axis=1)
z = kde(grid_coords.T)
z = z.reshape(div_real, div_real)
#fig = plt.figure()
#i = plt.imshow(z)#,aspect=x_flat.ptp()/y_flat.ptp(),origin='lower')
#plt.show()
limg = np.arcsinh(z)
limg = limg / limg.max()
low = np.percentile(limg, 0.25)
high = np.percentile(limg, 99.9)
opt_img = skie.exposure.rescale_intensity(limg, in_range=(low, high))
lm = morph.is_local_maximum(limg)
x1, y1 = np.where(lm.T == True)
v = limg[(y1, x1)]
lim = 0.6 * high
#x2, y2 = x1[v > lim], y1[x > lim]
peaks = [[], []]
for idx in range(0, len(x1)):
if limg[(y1[idx], x1[idx])] > lim:
peaks[0].append(x1[idx])
peaks[1].append(y1[idx])
#print peaks
#print x_flat[peaks[0]]
#print y_flat[peaks[1]]
num_peaks = len(peaks[0])
peak_distances = []
max_peak_distance = 0
peak_vels = []
peak_probs = []
for idx in range(0, len(peaks[0][:])):
peak_vels.append((x_flat[peaks[0][idx]], y_flat[peaks[1][idx]]))
peak_probs.append(
kde((x_flat[peaks[0][idx]], y_flat[peaks[1][idx]]))[0])
if len(peaks[0]) > 1:
for idx in range(0, len(peaks[0])):
for idx2 in range(0, len(peaks[1])):
peak_distances.append(math.sqrt(math.pow(x_flat[peaks[0][idx]]-x_flat[peaks[0][idx2]],2) \
+ math.pow(y_flat[peaks[1][idx]] - y_flat[peaks[1][idx2]],2)))
max_peak_distance = max(peak_distances)
print " %%max peak dist%% = " + str(max_peak_distance)
return peak_vels, peak_probs, max_peak_distance
def contour_features(im): """features 19- 24; contours of the image: number of contours in the image, mean of distance to background, and local maxima:pixel ratio""" contours=[] x_size=int(im.size[1]/float(10)) i=im.resize((x_size,int((im.size[0]/float(im.size[1])*x_size)))) x=i.getdata() z=np.array(x) if z.size/(len(z))==1: if z.size%3 !=0: z= z[0:len(z)-len(z)%3] z= z.reshape(-1,3) #print z.resize(z.size/float(3),3) #print z #contour mapping contour_array=measure.find_contours(z,0.5) len_contours=contours.__len__() contours.append(len_contours) # 19-21: distance features: use ndimage.distance to compute "distance" to background of each pixel. Then find mean among x, y, and z values (r,g,b) distance=ndimage.distance_transform_edt(z) yvaluemean= sum(distance[:,0])/len(distance[:,0]) xvaluemean= sum(distance[:,1])/len(distance[:,1]) zvaluemean= sum(distance[:,2])/len(distance[:,2]) contours.append(yvaluemean) contours.append(xvaluemean) contours.append(zvaluemean) #22-24: local_max features: find number of local maxima among x,y,z values (rgb) compared to the number of pixels local_max=is_local_maximum(distance,z,np.ones((3,3))) xmaxsum, ymaxsum, zmaxsum = [0,0,0] for a in (local_max[:,0]): if a ==False: xmaxsum+=1 contours.append(xmaxsum/float(len(local_max[:,0]))) for a in (local_max[:,1]): if a ==False: ymaxsum+=1 contours.append(ymaxsum/float(len(local_max[:,0]))) for a in (local_max[:,2]): if a ==False: zmaxsum+=1 contours.append(zmaxsum/float(len(local_max[:,0]))) ###25: object count filt_img=ndimage.gaussian_filter(z,10) T=70 labeled,nr_objects=ndimage.label(filt_img>T) contours.append(nr_objects) return contours
from skimage.morphology import watershed, is_local_maximum
import matplotlib.pyplot as plt
from scipy import ndimage
# Generate an initial image with two overlapping circles
x, y = np.indices((80, 80))
x1, y1, x2, y2 = 28, 28, 44, 52
r1, r2 = 16, 20
mask_circle1 = (x - x1) ** 2 + (y - y1) ** 2 < r1 ** 2
mask_circle2 = (x - x2) ** 2 + (y - y2) ** 2 < r2 ** 2
image = np.logical_or(mask_circle1, mask_circle2)
# Now we want to separate the two objects in image
# Generate the markers as local maxima of the distance
# to the background
distance = ndimage.distance_transform_edt(image)
local_maxi = is_local_maximum(distance, image, np.ones((3, 3)))
markers = ndimage.label(local_maxi)[0]
labels = watershed(-distance, markers, mask=image)
plt.figure(figsize=(9, 3.5))
plt.subplot(131)
plt.imshow(image, cmap='gray', interpolation='nearest')
plt.axis('off')
plt.subplot(132)
plt.imshow(-distance, interpolation='nearest')
plt.axis('off')
plt.subplot(133)
plt.imshow(labels, cmap='spectral', interpolation='nearest')
plt.axis('off')
plt.subplots_adjust(hspace=0.01, wspace=0.01, top=1, bottom=0, left=0,
bkg = func(grid_x, grid_y)
bkg = bkg / np.max(bkg)
# Creating points
clean = np.zeros((200, 400))
clean[(x, y)] += 5
clean = ndimage.gaussian_filter(clean, 3)
clean = clean / np.max(clean)
# Combining both the non-uniform background
# and points
fimg = bkg + clean
fimg = fimg / np.max(fimg)
# Calculating local maxima
lm1 = morph.is_local_maximum(fimg)
x1, y1 = np.where(lm1.T == True)
# Creating figure to show local maximum detection
# rate success
fig = plt.figure(figsize=(8, 4))
ax = fig.add_subplot(111)
ax.imshow(fimg)
ax.scatter(x1, y1, s=100, facecolor='none', edgecolor='#009999')
ax.set_xlim(0, 400)
ax.set_ylim(0, 200)
ax.xaxis.set_visible(False)
ax.yaxis.set_visible(False)
fig.savefig('scikits_412_ex1.png', bbox_inches='tight')
import matplotlib.pyplot as plt
from skimage.morphology import watershed, is_local_maximum
# Generate an initial image with two overlapping circles
x, y = np.indices((80, 80))
x1, y1, x2, y2 = 28, 28, 44, 52
r1, r2 = 16, 20
mask_circle1 = (x - x1)**2 + (y - y1)**2 < r1**2
mask_circle2 = (x - x2)**2 + (y - y2)**2 < r2**2
image = np.logical_or(mask_circle1, mask_circle2)
# Now we want to separate the two objects in image
# Generate the markers as local maxima of the distance to the background
from scipy import ndimage
distance = ndimage.distance_transform_edt(image)
local_maxi = is_local_maximum(distance, image, np.ones((3, 3)))
markers = ndimage.label(local_maxi)[0]
labels = watershed(-distance, markers, mask=image)
fig, axes = plt.subplots(ncols=3, figsize=(8, 2.7))
ax0, ax1, ax2 = axes
ax0.imshow(image, cmap=plt.cm.gray, interpolation='nearest')
ax1.imshow(-distance, cmap=plt.cm.jet, interpolation='nearest')
ax2.imshow(labels, cmap=plt.cm.spectral, interpolation='nearest')
for ax in axes:
ax.axis('off')
plt.subplots_adjust(hspace=0.01, wspace=0.01, top=1, bottom=0, left=0, right=1)
plt.show()
bkg = func(grid_x, grid_y)
bkg = bkg / np.max(bkg)
# Creating points
clean = np.zeros((200, 400))
clean[(x, y)] += 5
clean = ndimage.gaussian_filter(clean, 3)
clean = clean / np.max(clean)
# Combining both the non-uniform background
# and points
fimg = bkg + clean
fimg = fimg / np.max(fimg)
# Calculating local maxima
lm1 = morph.is_local_maximum(fimg)
x1, y1 = np.where(lm1.T == True)
# Creating figure to show local maximum detection
# rate success
fig = plt.figure(figsize=(8, 4))
ax = fig.add_subplot(111)
ax.imshow(fimg)
ax.scatter(x1, y1, s=100, facecolor="none", edgecolor="#009999")
ax.set_xlim(0, 400)
ax.set_ylim(0, 200)
ax.xaxis.set_visible(False)
ax.yaxis.set_visible(False)
fig.savefig("scikits_412_ex1.pdf", bbox_inches="tight")
def test_target(tar):
'''
if the convexity of the target is <threshold try to do a watershed segmentation
make black image with white polygon
do watershed segmentation
find polygon center
'''
def test(ti):
ctest = ti.convexity > threshold
centtest = self._near_center(*ti.centroid_value)
atest = ma > ti.area > mi
return ctest, centtest, atest
ctest, centtest, atest = test(tar)
if not ctest and (atest and centtest):
# src = self.target_image.get_frame(0)
# self._draw_result(src, tar)
# w, h = self.croppixels
# self.target_image.save('/Users/ross/Sandbox/machine_vision/polygon.jpg', width=w, height=h)
src = self.target_image.get_frame(0)
draw_polygons(src, [tar.poly_points], color=(0, 255, 255))
# make image with polygon
image = zeros(self.croppixels)
points = asarray(tar.poly_points)
points = asarray([(pi.x, pi.y) for pi in points])
rr, cc = polygon(points[:, 0], points[:, 1])
image[cc, rr] = 255
# do watershedding
distance = ndimage.distance_transform_edt(image)
local_maxi = is_local_maximum(distance, image)
markers, ns = ndimage.label(local_maxi)
wsrc = watershed(-distance, markers,
mask=image
)
# find the label with the max area ie max of histogram
def get_limits(values, bins):
ind = argmax(values)
if ind == 0:
bil = bins[ind]
biu = bins[ind + 1]
elif ind == len(bins) - 1:
bil = bins[ind - 1]
biu = bins[ind]
else:
bil = bins[ind - 1]
biu = bins[ind + 1]
return bil, biu, ind
# bins = 3 * number of labels. this allows you to precisely pick the value of the max area
values, bins = histogram(wsrc, bins=ns * 3)
bil, biu, ind = get_limits(values, bins)
if not bil:
values = delete(values, ind)
bins = delete(bins, (ind, ind + 1))
bil, biu, ind = get_limits(values, bins)
# print values
# print bins
# print bil, biu
nimage = ones_like(wsrc, dtype='uint8')
nimage[wsrc < bil] = 0
nimage[wsrc > biu] = 0
img = asMat(nimage)
# locate new polygon from the segmented image
tars = self._locate_targets(img)
if globalv.show_autocenter_debug_image:
do_later(lambda: self.debug_show(image, distance, wsrc, nimage))
if tars:
tar = tars[0]
ctest, centtest, atest = test(tar)
# ctest = tar.convexity > threshold
# centtest = self.
else:
return None, False
return tar, ctest and atest and centtest
peak_distances = []
for idx in range(0,len(contour_centroids)):
for idx2 in range(0,len(contour_centroids)):
peak_distances.append(math.sqrt(math.pow(contour_centroids[idx][0]-contour_centroids[idx2][0],2) \
+ math.pow(contour_centroids[idx][1] - contour_centroids[idx2][1],2)))
'''
limg = np.arcsinh(z)
limg = limg / limg.max()
low = np.percentile(limg, 0.25)
high = np.percentile(limg, 99.9)
opt_img = skie.exposure.rescale_intensity(limg,
in_range=(low,
high))
lm = morph.is_local_maximum(limg)
x1, y1 = np.where(lm.T == True)
v = limg[(y1, x1)]
lim = 0.6 * high
#x2, y2 = x1[v > lim], y1[x > lim]
peaks = [[], []]
for idx in range(0, len(x1)):
if limg[(y1[idx], x1[idx])] > lim:
peaks[0].append(x1[idx])
peaks[1].append(y1[idx])
print peaks
peak_distances = []
for idx in range(0, len(peaks)):
import skimage.morphology as morph
import skimage.exposure as skie
import matplotlib.pyplot as plt
# Loading astronomy image from an infrared space telescope
img = pyfits.getdata('stellar_cluster.fits')[500:1500, 500:1500]
# Prep file scikit-image environment and plotting
limg = np.arcsinh(img)
limg = limg / limg.max()
low = np.percentile(limg, 0.25)
high = np.percentile(limg, 99.5)
opt_img = skie.exposure.rescale_intensity(limg, in_range=(low, high))
# Calculating local maxima and filtering out noise
lm = morph.is_local_maximum(limg)
x1, y1 = np.where(lm.T == True)
v = limg[(y1, x1)]
lim = 0.5
x2, y2 = x1[v > lim], y1[v > lim]
# Creating figure to show local maximum detection
# rate success
fig = plt.figure(figsize=(8, 4))
fig.subplots_adjust(hspace=0.05, wspace=0.05)
ax1 = fig.add_subplot(121)
ax1.imshow(opt_img)
ax1.set_xlim(0, img.shape[1])
ax1.set_ylim(0, img.shape[0])
ax1.xaxis.set_visible(False)
if seg == 'edge detection':
# perform edge detection. Different edge filters are available in Python, here the Sobel filter is used
edges = sobel(image)
# threshold edges to select only strong contours (NB: intensity values (0-255) are normalized between 0 and 1)
image_segmented = edges > 0.1
# fill potential holes in segmented objects
image_segmented_filled = nd.binary_fill_holes(image_segmented)
#watershed segmentation to split touching objects, if activated in the user's section
if ws == '_ws_':
# distance map is created where the distance of each foreground pixel to the closest background pixel is calculated and used for watershed splitting
distance = nd.distance_transform_edt(image_segmented_filled)
# apply Gaussian filter to the distance map to merge several local maxima into one
distance=nd.gaussian_filter(distance,4)
local_maxi = is_local_maximum(distance, labels=image_segmented_filled, footprint=np.ones((3, 3)))
markers = nd.label(local_maxi)[0]
# return labelled image (array of unique integers per object)
labelled_image = watershed(-distance, markers, mask=image_segmented_filled)
# 2b. WATERSHED SPLIT
#--------------------
# return labelled image (array of unique integers per object)
if ws != '_ws_':
labelled_image, nb_labels = nd.label(image_segmented_filled)
# 3. MEASURE OBJECT PROPERTIES AND CONDUCT SIZE-BASED EXCLUSION
#--------------------------------------------------------------
# measure object size on labelled image
