def edges(cls):
from scipy import ndimage, misc
import numpy as np
from skimage import feature
col = Image.open("f990.jpg")
gray = col.convert('L')
# Let numpy do the heavy lifting for converting pixels to pure black or white
bw = np.asarray(gray).copy()
# Pixel range is 0...255, 256/2 = 128
bw[bw < 245] = 0 # Black
bw[bw >= 245] = 255 # White
bw[bw == 0] = 254
bw[bw == 255] = 0
im = bw
im = ndimage.gaussian_filter(im, 1)
edges2 = feature.canny(im, sigma=2)
labels, numobjects =ndimage.label(im)
slices = ndimage.find_objects(labels)
print('\n'.join(map(str, slices)))
misc.imsave('f990_sob.jpg', im)
return
#im = misc.imread('f990.jpg')
#im = ndimage.gaussian_filter(im, 8)
sx = ndimage.sobel(im, axis=0, mode='constant')
sy = ndimage.sobel(im, axis=1, mode='constant')
sob = np.hypot(sx, sy)
misc.imsave('f990_sob.jpg', edges2)
def calc_orient(image):
#image = cv2.imread('test_window2.png', cv2.CV_LOAD_IMAGE_GRAYSCALE)
#image = resize.resize_img(image)
#dbl_image = image.astype(float)
#cv2.imshow('orginal', image)
smoothed_im = cv2.GaussianBlur(image,(3,3), 0) # gaussian blur using 3x3 window
dx = ndimage.sobel(smoothed_im, 0) # horizontal sobel gradient
dy = ndimage.sobel(smoothed_im, 1) # vertical sobel gradient
print dy
theta_mx = np.array([[]])
for y in range(0,len(dy)):
for x in range(0,len(dx)):
theta_mx[x][y] = math.atan2(dy[y],dx[x])
#theta = # in radians
print theta_mx
#cv2.imshow('dx',dx)
#cv2.waitKey()
lines_im = draw_lines2.draw_lines(image, theta, 1)
#print lines_im
return lines_im
#print dx.shape[0],dx.shape[1]
#print dx
#print 'hello',dy
#calc_orient()
def test_multiple_modes():
# Test that the filters with multiple mode cababilities for different
# dimensions give the same result as applying a single mode.
arr = np.array([[1., 0., 0.],
[1., 1., 0.],
[0., 0., 0.]])
mode1 = 'reflect'
mode2 = ['reflect', 'reflect']
assert_equal(sndi.gaussian_filter(arr, 1, mode=mode1),
sndi.gaussian_filter(arr, 1, mode=mode2))
assert_equal(sndi.prewitt(arr, mode=mode1),
sndi.prewitt(arr, mode=mode2))
assert_equal(sndi.sobel(arr, mode=mode1),
sndi.sobel(arr, mode=mode2))
assert_equal(sndi.laplace(arr, mode=mode1),
sndi.laplace(arr, mode=mode2))
assert_equal(sndi.gaussian_laplace(arr, 1, mode=mode1),
sndi.gaussian_laplace(arr, 1, mode=mode2))
assert_equal(sndi.maximum_filter(arr, size=5, mode=mode1),
sndi.maximum_filter(arr, size=5, mode=mode2))
assert_equal(sndi.minimum_filter(arr, size=5, mode=mode1),
sndi.minimum_filter(arr, size=5, mode=mode2))
assert_equal(sndi.gaussian_gradient_magnitude(arr, 1, mode=mode1),
sndi.gaussian_gradient_magnitude(arr, 1, mode=mode2))
assert_equal(sndi.uniform_filter(arr, 5, mode=mode1),
sndi.uniform_filter(arr, 5, mode=mode2))
def image_multi_features(img, maxPixel, num_features,imageSize):
# X is the feature vector with one row of features per image
# consisting of the pixel values a, num_featuresnd our metric
mask = img > img.mean()
label_im, nb_labels = ndimage.label(mask)
X=np.zeros(num_features, dtype=float)
# image=ndimage.median_filter(img, 3)
X[imageSize]=img.max()
X[imageSize+1]=img.min()
X[imageSize+2]=nb_labels
extra_sizes=[4,8,12,16,20,24]
image = resize(img, (maxPixel, maxPixel))
granulo = granulometry(image,sizes=extra_sizes)
sx = ndimage.sobel(image, axis=0, mode='constant')
sy = ndimage.sobel(image, axis=1, mode='constant')
sob = np.hypot(sx, sy)
#edges=canny(image,3,0.3,0.2)
# Store the rescaled image pixels
X[0:imageSize] = np.reshape(sob,(1, imageSize))
for i in range(len(extra_sizes)):
X[imageSize+3+i]=granulo[i]
return X
def _compute_derivatives(image, mode='constant', cval=0):
"""Compute derivatives the way that scikit-image does it (for comparison).
This method is fully copied from the repository."""
imgy = ndimage.sobel(image, axis=0, mode=mode, cval=cval)
imgx = ndimage.sobel(image, axis=1, mode=mode, cval=cval)
return imgx, imgy
def canny():
image = cv2.imread('1_1.png', cv2.CV_LOAD_IMAGE_GRAYSCALE)
#image = misc.lena()
# apply gaussian blur 3x3 sigma 0.5
smoothed_im = cv2.GaussianBlur(image,(3,3), 0)
# calculate gradients with sobel filter
grad_x = ndimage.sobel(image, 0)
grad_y = ndimage.sobel(image, 1)
# calculate magnitude
grad_mag = numpy.sqrt(grad_x**2+grad_y**2)
# calculating theta
grad_angle = numpy.arctan2(grad_y, grad_x)
# quantize angles 0 to 3
quantized_angle = numpy.around(3 * (grad_angle + numpy.pi) / (numpy.pi * 2))
# non-maximal suppression
# quantize magnitude into 4 directions
#NE = ndimage.maximum_filter(grad_mag, footprint=_NE)
#cv2.imshow('grad_x',grad_x)
#cv2.imshow('grad_y',grad_y)
#cv2.imshow('grad_mag',grad_angle)
cv2.waitKey()
def showAllImages():
fig, current = plt.figure(), 0
for file_name in glob.glob("*.png"):
image = img.imread(file_name)
R, G, B = image[:,:,0], image[:,:,1], image[:,:,2]
greyscale = 0.21 * R + 0.72 * G + 0.07 * B
threshold = 210.0 / 256
greyscale[greyscale > threshold] = 1
greyscale[greyscale <=threshold] = 0
dx = ndimage.sobel(greyscale, 0) # horizontal derivative
dy = ndimage.sobel(greyscale, 1) # vertical derivative
mag = np.hypot(dx, dy) # magnitude
y_min, x_min, y_max, x_max = findMinMax(mag)
mag = mag[x_min:x_max+1, y_min:y_max+1]
# x, y = mag.shape
# ration = float(x) / y
# mag = imresize(mag, (50, int(50 / ration)))
#
# mag[mag > threshold] = 1
# mag[mag <=threshold] = 0
current += 1
if current == 7:
fig.add_subplot(1, 1, 1).imshow(mag, cmap=cm.Greys_r)
print guessFigure(mag)
break
plt.tight_layout()
plt.show()
def compute_harris_response(image, eps=1e-6):
""" compute the Harris corner detector response function
for each pixel in the image"""
# derivatives
image = ndimage.gaussian_filter(image, 1)
imx = ndimage.sobel(image, axis=0, mode='constant')
imy = ndimage.sobel(image, axis=1, mode='constant')
Wxx = ndimage.gaussian_filter(imx * imx, 1.5, mode='constant')
Wxy = ndimage.gaussian_filter(imx * imy, 1.5, mode='constant')
Wyy = ndimage.gaussian_filter(imy * imy, 1.5, mode='constant')
# determinant and trace
Wdet = Wxx * Wyy - Wxy ** 2
Wtr = Wxx + Wyy
harris = Wdet / (Wtr + eps)
# Non maximum filter of size 3
harris_max = ndimage.maximum_filter(harris, 3, mode='constant')
harris *= harris == harris_max
# Remove the image corners
harris[:3] = 0
harris[-3:] = 0
harris[:, :3] = 0
harris[:, -3:] = 0
return harris
def slope_aspect(array, pix_size, scale):
dzdx = ndimage.sobel(array, axis=1)/(8.*pix_size)
dzdy = ndimage.sobel(array, axis=0)/(8.*pix_size)
hyp = numpy.hypot(dzdx,dzdy)
slp = numexpr.evaluate("arctan(hyp * scale)")
asp = numexpr.evaluate("arctan2(dzdy, -dzdx)")
return slp, asp
def _compute_derivatives(image, mode="constant", cval=0):
"""Compute derivatives in x and y direction using the Sobel operator.
Parameters
----------
image : ndarray
Input image.
mode : {'constant', 'reflect', 'wrap', 'nearest', 'mirror'}, optional
How to handle values outside the image borders.
cval : float, optional
Used in conjunction with mode 'constant', the value outside
the image boundaries.
Returns
-------
imx : ndarray
Derivative in x-direction.
imy : ndarray
Derivative in y-direction.
"""
imy = ndimage.sobel(image, axis=0, mode=mode, cval=cval)
imx = ndimage.sobel(image, axis=1, mode=mode, cval=cval)
return imx, imy
def sobel(self, img):
fig = scipy.misc.imread(img).astype('int32')
dx = ndimage.sobel(fig, 0) # horizontal derivative
dy = ndimage.sobel(fig, 1) # vertical derivative
mag = np.hypot(dx, dy) # magnitude
mag *= 255.0 / np.max(mag) # normalize
scipy.misc.imsave('sobel.png', mag)
def gradient_transformation(data):
"""Gradient distance transformation."""
gx = ndimage.sobel(data, 0)
gy = ndimage.sobel(data, 1)
gz = ndimage.sobel(data, 2)
grad = np.sqrt(gx**2 + gy**2 + gz**2)
return grad
def testAll():
fig = plt.figure(figsize=(16, 8))
current = 0
for file_name in glob.glob("img/*.jpg"):
image = img.imread(file_name)
r = imresize(image, (100, 100))
mask1 = ndimage.binary_erosion(coloration(r), structure=np.ones((3, 3)))
greyscale = np.average(image, axis=2)
dx = ndimage.sobel(greyscale, 0) # horizontal derivative
dy = ndimage.sobel(greyscale, 1) # vertical derivative
mask2 = np.hypot(dx, dy) # magnitude
current += 1
fig.add_subplot(3, 3, current).imshow(mask1, cmap=cm.Greys_r, interpolation='none')
labeled_array, num_features = ndimage.label(mask1)
print labeled_array, num_features
current += 1
fig.add_subplot(3, 3, current).imshow(mask2, cmap=cm.Greys_r, interpolation='none')
current += 1
fig.add_subplot(3, 3, current).imshow(r, cmap=cm.Greys_r, interpolation='none')
plt.axis('off')
plt.show()
def detectCircles(img, r, useGradient):
grayimg = rgb2gray(img)
edges = cv2.Canny(img,100,200)
ax[0].imshow(edges, cmap=plt.cm.gray)
ax[0].set_title('after canny image operation')
if useGradient == 0:
accumulator1 = np.zeros(edges.shape)
for (i,j),value in np.ndenumerate(edges):
if value:
for t_idx in np.arange(0,2*math.pi,math.pi/100):
a = int(i - (r * math.cos(t_idx)));
b = int(j + (r * math.sin(t_idx)));
if a>0 and b>0 and a < accumulator1.shape[0] and b < accumulator1.shape[1]:
accumulator1[a, b] += 1
print accumulator1
ax[1].imshow(accumulator1, cmap=plt.cm.gray)
ax[1].set_title('Accumulator array without using gradient')
else:
dx = ndimage.sobel(grayimg, axis=0, mode='constant')
dy = ndimage.sobel(grayimg, axis=1, mode='constant')
accumulator = np.zeros(edges.shape)
for (i,j),value in np.ndenumerate(edges):
if value:
gradient = math.atan(-dx[i,j]/(dy[i,j]+0.00001))
for theta in np.arange(gradient-math.pi/4,gradient+math.pi/4,math.pi/100):
a = int(i - (r * math.cos(theta)));
b = int(j + (r * math.sin(theta)));
if a < accumulator.shape[0] and b < accumulator.shape[1]:
accumulator[a, b] += 1
ax[1].imshow(accumulator, cmap=plt.cm.gray)
ax[1].set_title('Accumulator array with gradient')
print accumulator
return
def calc_orient(image):
#image = cv2.imread('test_window.png', cv2.CV_LOAD_IMAGE_GRAYSCALE)
#image = resize.resize_img(image)
#dbl_image = image.astype(float)
#cv2.imshow('orginal', image)
smoothed_im = cv2.GaussianBlur(image,(3,3), 0) # gaussian blur using 3x3 window
dx = ndimage.sobel(smoothed_im, 0) # horizontal sobel gradient
dy = ndimage.sobel(smoothed_im, 1) # vertical sobel gradient
max_dx = np.max(dx)-np.min(dx)
max_dy = np.max(dy)-np.min(dy)
# print max_dx
#print max_dy
theta = math.atan2(max_dy,max_dx) # in radians
print 'grad',theta
#cv2.imshow('dx',dx)
#cv2.waitKey()
orientation_im = draw_lines.draw_lines(image, theta)
return orientation_im
#print dx.shape[0],dx.shape[1]
#print dx
#print 'hello',dy
#calc_orient()
def get_hist_angle_bins(img):
'''
given a python image, return 2 lists of 16 elements each. 1 list (bin_lowers) has the lower bounds for each bin, the other list (hist_vals)
has the number of pixels in each bin.
'''
im = numpy.array(img)
im = numpy.resize(im,(8,8))#reshape array to model 8x8 cell
sx = ndimage.sobel(im, axis=0, mode = 'constant')#apply sobel operator in x-direction
sy = ndimage.sobel(im, axis=1, mode = 'constant')#apply sobel operator in y-direction
sx = list(numpy.array(sx).reshape(-1,))#reshape sobel output into 1-d list for easy manipulation
sy = list(numpy.array(sx).reshape(-1,))
sobel = []
for index in range(0,64):
x = float(sx[index])
y = float(sy[index])
if x != 0:#Avoid a divide-by-zero error
angle = math.degrees((math.atan2(y,x))) #does this give the angle we're looking for? trying to get the full 360-degree range
else:
angle = math.degrees((math.atan2(y,1)))#what should the value be if x is 0?
if angle<0:
angle +=360
sobel.append(angle)
hist, bin_edges = numpy.histogram(sobel, bins = 16)
bin_lowers = list(numpy.array(bin_edges).reshape(-1,))#unnecessary because i've already reshaped the data?
bin_lowers.pop()#gets rid of the high side of the highest bin
hist_vals = list(numpy.array(hist).reshape(-1,))#also unnecessary?
return bin_lowers, hist_vals
def load_contourlabel(self, idx):
"""
Load label image as 1 x height x width integer array of label indices.
The leading singleton dimension is required by the loss.
"""
import scipy.io
mat = scipy.io.loadmat('{}/cls/{}.mat'.format(self.sbdd_dir, idx))
label = mat['GTcls'][0]['Segmentation'][0].astype(np.uint8)
from scipy import ndimage
edge_horizont = ndimage.sobel(label, 0)
edge_vertical = ndimage.sobel(label, 1)
magnitude = np.hypot(edge_horizont, edge_vertical)
contour_label = np.zeros(label.shape,dtype=np.uint8)
contour_mask = np.zeros(label.shape,dtype=np.uint8)
label[label==0]=254
label[label<254]=1
contour_mask[magnitude>0] = 255
kernel = np.ones((5,5),np.uint8)
import cv2
#contour_mask = cv2.dilate(contour_mask,kernel,iterations = 1)
#contour_mask = cv2.erode(contour_mask,kernel,iterations = 1)
contour_label[label==1]=1
contour_label[label==254]=0
contour_label[contour_mask==255]=2
contour_label = contour_label[np.newaxis, ...]
return contour_label
def sobel_gradient(Im):
Imy=nd.sobel(Im, axis=0)
Imx=nd.sobel(Im, axis=1)
ImMag=(Imx**2 +Imy**2)**0.5
return ImMag, Imx, Imy
def canny(image, high_threshold, low_threshold):
grad_x = ndimage.sobel(image, 0)
grad_y = ndimage.sobel(image, 1)
grad_mag = numpy.sqrt(grad_x**2+grad_y**2)
grad_angle = numpy.arctan2(grad_y, grad_x)
# next, scale the angles in the range [0, 3] and then round to quantize
quantized_angle = numpy.around(3 * (grad_angle + numpy.pi) / (numpy.pi * 2))
# Non-maximal suppression: an edge pixel is only good if its magnitude is
# greater than its neighbors normal to the edge direction. We quantize
# edge direction into four angles, so we only need to look at four
# sets of neighbors
NE = ndimage.maximum_filter(grad_mag, footprint=_NE)
W = ndimage.maximum_filter(grad_mag, footprint=_W)
NW = ndimage.maximum_filter(grad_mag, footprint=_NW)
N = ndimage.maximum_filter(grad_mag, footprint=_N)
thinned = (((grad_mag > W) & (quantized_angle == _N_d )) |
((grad_mag > N) & (quantized_angle == _W_d )) |
((grad_mag > NW) & (quantized_angle == _NE_d)) |
((grad_mag > NE) & (quantized_angle == _NW_d)) )
thinned_grad = thinned * grad_mag
# Now, hysteresis thresholding: find seeds above a high threshold, then
# expand out until we go below the low threshold
high = thinned_grad > high_threshold
low = thinned_grad > low_threshold
canny_edges = ndimage.binary_dilation(high, structure=numpy.ones((3,3)), iterations=-1, mask=low)
return grad_mag, thinned_grad, canny_edges
def make_sobel(self):
"""Find edges using the Sobel filter"""
luminance = self.arrays['l']
sx = ndimage.sobel(luminance, axis=0, mode='nearest')
sy = ndimage.sobel(luminance, axis=1, mode='nearest')
sob = numpy.hypot(sx, sy)
return sx, sy, sob
def structure_tensor(im, radius):
bi = ndimage.gaussian_filter(im.astype(np.float32), 1)
imy = ndimage.sobel(bi, axis=0)
imx = ndimage.sobel(bi, axis=1)
Axx = ndimage.gaussian_filter(imx ** 2, radius)
Axy = ndimage.gaussian_filter(imx * imy, radius)
Ayy = ndimage.gaussian_filter(imy ** 2, radius)
return Axx, Axy, Ayy
def compute_energy(im):
im = rgb2gray(im)
im = im.astype('int32')
dx = ndimage.sobel(im, 0)
dy = ndimage.sobel(im, 1)
mag = numpy.hypot(dx,dy)
mag *= 255.0/numpy.max(mag)
return mag
def detect_edges(self):
sx = ndimage.sobel(self.image_raw.image, axis=0, mode='constant')
sy = ndimage.sobel(self.image_raw.image, axis=1, mode='constant')
image_edges = np.hypot(sx, sy)
t = image_edges.mean()*1.4
image_edges[image_edges<t]=1
image_edges[image_edges>=t]=0
self.image_bw.image = image_edges
def apply_sobel_filter(image):
""" Apply sobel filter on an image, return the filtered object.
This routine roughly follows the solution provided in:
http://stackoverflow.com/questions/7185655/applying-the-sobel-filter-using-scipy
"""
dx = ndimage.sobel(image, 0) # horizontal derivative
dy = ndimage.sobel(image, 1) # vertical derivative
mag = numpy.hypot(dx, dy) # magnitude
return mag
def sobelFunction(s):
im = scipy.misc.imread(s)
im = im.astype('int32')
imagename = s[:-4]
dx = ndimage.sobel(im, 0) # horizontal derivative
dy = ndimage.sobel(im, 1) # vertical derivative
mag = numpy.hypot(dx, dy) # magnitude
mag *= 255.0 / numpy.max(mag) # normalize (Q&D)
scipy.misc.imsave(imagename+'SobelOutput.jpg', mag)
def TestsobHypot(rec):
x=ndimage.sobel(abs(rec)**2,axis=0, mode='constant')
y=ndimage.sobel(abs(rec)**2,axis=1, mode='constant')
hype = np.hypot(x,y)
hype = hype.mean(axis=0).mean(axis=0)
index = hype.argmax()
return index
def add_basic_transforms(self):
rgb = [_Transform(lambda im, k=i: im[..., k]) for i in range(3)]
laplace = [_Transform(lambda im, k=i: ndimage.filters.laplace(im)[..., k]) for i in range(3)]
grad = [_Transform(lambda im, k=i: np.arctan2(ndimage.sobel(im[..., k], 0), ndimage.sobel(im[..., k], 1)))
for i in range(3)] + [_Transform(
lambda im, k=i: np.hypot(ndimage.sobel(im[..., k], 0), ndimage.sobel(im[..., k], 1))) for i in range(3)]
# svm = Svm()
# svm.load('svm_skin.pkl')
# skin = [_Transform(lambda im: svm.classify_vec(im))]
self.transform_lists += [rgb, laplace, grad]
def ExtractEdge(imdata):
dx = ndimage.sobel(imdata.getDataBuffer(), 0)
dy = ndimage.sobel(imdata.getDataBuffer(), 1)
mag = np.hypot(dx, dy)
mag = mag.astype("int32")
mag = 255 * mag / np.max(np.max(mag))
imdata.setDataBuffer(mag)
return imdata
def energy_image_2 (im):
gray_img = rgb2gray(im)
double_img = im2double(gray_img)
dx = ndimage.sobel(im, 0) # horizontal derivative
dy = ndimage.sobel(im, 1) # vertical derivative
mag = numpy.hypot(dx, dy) # magnitude
mag *= 255.0 / numpy.max(mag) # normalize (Q&D)
scipy.misc.imsave('sobel.jpg', mag)
#gradient_img = ndimage.filters.sobel(double_img, axis=-1, output=None, mode='reflect', cval=0.0)
return im2double(gradient_img)
def saveEdges(binary, name):
"""
Creates an image where you only see the edges of the particles.
"""
dilatedForSobel = binary.astype(np.int)
dilatedForSobel[binary] = 255
dx = ndimage.sobel(dilatedForSobel, 0) # horizontal derivative
dy = ndimage.sobel(dilatedForSobel, 1) # vertical derivative
mag = np.hypot(dx, dy) # magnitude
cv2.imwrite(name+"_"+_NAME_OF_EDGES_IMAGE, mag)
def sobel(file):
sx = ndimage.sobel(file, axis=0, mode='constant')
sy = ndimage.sobel(file, axis=1, mode='constant')
sob = np.hypot(sx, sy)
#es warn 235 und 250
sob = split_histogram(sob, 50000, 80000)
sx2 = ndimage.sobel(sob, axis=0, mode='constant')
sy2 = ndimage.sobel(sob, axis=1, mode='constant')
sob2 = np.hypot(sx2, sy2)
#sob2 = split_histogram(sob2, 0, 200)
return sob2
def process(img):
# processes (resize, color, crop, etc) the image to be sent
# to trained model for evaluation
crop = img[tah:h, 0:(w // 3)] # crops
resized = cv2.resize(crop, (28, 28)) # resizes image to 28*28 pixels
resized = ndimage.gaussian_filter(resized,
0.25) #gassian filter with sigma = 0.25
ho = ndimage.sobel(resized, 0) #horizontal sobel edge detection
v = ndimage.sobel(resized, 1) # vertical sobel edge detection
inverted = np.hypot(ho, v) # combine both edge detection images
return inverted #return processed image
def apply_sobel_2d(img):
"""
Takes in the data array from a nii file and applies the Sobel gradient operator on it. Sobel gradient
directions are binned from 0-8.
Parameters:
NUMPY.NDARRAY representing the greyscale voxel values
Returns:
NUMPY.NDARRAY of the binned Sobel direction for each voxel
NUMPY.NDARRAY of the Sobel magnitude for each voxel
"""
start = time.time()
sobel_threshold = 2
#matrices of sobel gradients in x and y directions
dx = ndimage.sobel(img, 0)
dy = ndimage.sobel(img, 1)
shape = dx.shape
sobel_magnitudes = numpy.ndarray(shape=shape)
sobel_directions = numpy.ndarray(shape=shape)
#for every pixel
for x in range(shape[0]):
for y in range(shape[1]):
coord = tuple((x, y))
mx = dx.item(coord)
my = dy.item(coord)
#calculate gradient magnitude
mag = math.sqrt((mx**2) + (my**2))
sobel_magnitudes.itemset(coord, mag)
#if mag is high enough, calculate gradient bin and save it
if mag > sobel_threshold:
bin = bin_sobel_2d(mx, my)
sobel_directions.itemset(coord, bin)
else:
sobel_directions.itemset(coord, 0)
# print("TIME: " + str(time.time()-start))
return sobel_directions, sobel_magnitudes
def main():
if not os.path.exists('results'):
os.makedirs('results')
for filename in os.listdir("data"):
for method in ["SSD", "NCC"]:
print(filename)
###################################################
#### using naive method on smaller .jpg images ####
###################################################
if filename.endswith(".jpg"):
b, g, r = split_image("data/" + filename)
# align the green and red color channels to the blue color channel
ag, g_shift = align(g, b)
ar, r_shift = align(r, b)
# stack the color channels
im_out = np.dstack([ar, ag, b])
# save the image
skio.imsave("results/" + str.split(filename, ".")[0] + "_" + method + "_g_" + str(g_shift) + \
"_r_" + str(r_shift) + ".jpg", im_out)
####################################################
#### using pyramid method on larger .tif images ####
####################################################
elif filename.endswith(".tif"):
b, g, r = split_image("data/" + filename)
# apply sobel edge filter (improvement is neglibile for all images other than emir.tiff)
b_s = np.abs(sobel(b))
g_s = np.abs(sobel(g))
r_s = np.abs(sobel(r))
_, g_shift = pyramid(g_s, b_s)
_, r_shift = pyramid(r_s, b_s)
# apply displacement shift to original color channels
ag = np.roll(g, g_shift, (0, 1))
ar = np.roll(r, r_shift, (0, 1))
# stack the color channels
im_out = np.dstack([ar, ag, b])
# save the image
skio.imsave("results/" + str.split(filename, ".")[0] + "_" + method + "_g_" + str(g_shift) + \
"_r_" + str(r_shift) + ".jpg", im_out)
def gen_data(self, row, size=30):
gray = self.gray_padded_image
binar = self.bin_padded_image
self.my_dict = {'gray_pixel_value': [], 'mean_gray_pixel_value': [], 'bin_percentage_colored': [], 'sobel_gradient': [], 'label': []}
for r in range(30, 220):
row_idx = r
for c in range(230, 255):
col_idx = c
self.plot_frame(gray, row_idx, col_idx, size)
self.plot_frame(binar, row_idx, col_idx, size)
plt.show()
gray_window = self.gray_padded_image[row_idx:row_idx+size, col_idx:col_idx+size]
bin_window = self.bin_padded_image[row_idx:row_idx+size, col_idx:col_idx+size]
gray_pixel_value = gray_window[15, 15]
bin_pixel_value = bin_window[15, 15]
mean_pixel_value = np.mean(gray_window)
gray_window_sobel = ndimage.sobel(gray_window, axis=0)
colored_percentage = np.count_nonzero(bin_window==0)/(30**2)
above_area = np.mean(gray_window_sobel[8:16, 15])
below_area = np.mean(gray_window_sobel[16:23, 15])
sobel_gradient = above_area - -below_area
if bin_pixel_value == 0.0:
label = self.get_label()
self.update_dict(gray_pixel_value, mean_pixel_value, colored_percentage, sobel_gradient, label)
self.save_imgs(label, gray_window, bin_window, row_idx, col_idx)
print(label)
self.df = pd.DataFrame.from_dict(self.my_dict)
self.add_label_history()
self.save_csv(row_idx)
def plot_frame(self, zoom, row_index, col_index, size):
masked_window = np.random.random((zoom.shape[0],zoom.shape[1]))
masked_window[row_index:row_index+size, col_index:col_index+size] = 1
masked_window = np.ma.masked_where(masked_window != 1, masked_window)
masked_pixel = np.random.random((zoom[row_index:row_index+size, col_index:col_index+size].shape[0],zoom[row_index:row_index+size, col_index:col_index+size].shape[1]))
masked_pixel[15,15] = 1
masked_pixel = np.ma.masked_where(masked_pixel != 1, masked_pixel)
masked_pixel1 = np.random.random((zoom[row_index:row_index+size, col_index:col_index+size].shape[0],zoom[row_index:row_index+size, col_index:col_index+size].shape[1]))
masked_pixel1[8:23, 15] = 1
masked_pixel1 = np.ma.masked_where(masked_pixel1 != 1, masked_pixel1)
window = zoom[row_index:row_index+size, col_index:col_index+size]
colored_percentage = np.count_nonzero(window==0)/(30**2)
pixel_value = window[15, 15]
window_sobel = ndimage.sobel(window, axis=0)
above_area = np.mean(window_sobel[8:16, 15])
below_area = np.mean(window_sobel[16:23, 15])
sobel_value = window_sobel[15, 15]
# Overlay the two images
fig, ax = plt.subplots(1, 3)
ax.ravel()
ax[0].imshow(zoom, cmap='gray')
ax[0].imshow(masked_window, cmap='prism', interpolation='none')
# ax[0].imshow(masked_pixel, cmap=cm.jet, interpolation='none')
ax[0].set_title('Colored: {}'.format(round(colored_percentage, 2)))
ax[1].imshow(window, cmap='gray')
ax[1].imshow(masked_pixel, cmap='prism', interpolation='none')
ax[1].set_title('Pixel Value: {}'.format(pixel_value))
ax[2].imshow(window_sobel)
ax[2].imshow(masked_pixel1, cmap='jet')
ax[2].imshow(masked_pixel, cmap='prism', interpolation='none')
def showDrawing(event):
threshold = thresholdScale.get()
startTime = time.time()
imageArray = np.array(dataSet[imageScale.get(), :, :])
dimension = imageArray[0].size
imageArray = (imageArray - imageArray.min()) / (
imageArray.max() - imageArray.min()) * threshold / 255
maskArray = np.array(
file.get(list(file.items())[0][0])[imageScale.get(), ...])
maskArray = sobel(maskArray)
maskArray = np.sqrt(maskArray**2)
minmaxLabel.configure(text="Min = " + str(imageArray.min()) + " Max = " +
str(imageArray.max()))
maskArray = (maskArray - maskArray.min()) / (maskArray.max() -
maskArray.min())
if not applyMask.get():
finalArray = np.rint(imageArray * 255)
else:
imageArray = np.rint(imageArray * 255)
rgbArray = np.zeros((dimension, dimension, 3), 'uint8')
rgbArray[...,
0] = imageArray + maskArray * (255 - thresholdScale.get())
rgbArray[..., 1] = imageArray
rgbArray[..., 2] = imageArray
finalArray = rgbArray
img = Image.fromarray(finalArray)
img = img.resize((800, 800))
img = ImageTk.PhotoImage(master=window, image=img)
window.dontDeleteMePlease = img
canvas.create_image(pixelWidth, pixelHeight, image=img)
print(time.time() - startTime)
def sobel(img_array):
'''
image: image to convert with Sobel filter
'''
col = np.zeros((img_array.shape))
for i in range(img_array.shape[0]):
# print(i)
img = img_array[i]
dx = ndimage.sobel(img, 0) # horizontal derivative
dy = ndimage.sobel(img, 1) # vertical derivative
mag = np.hypot(dx, dy) # magnitude, equivalent to sqrt(dx**2 + dy**2)
mag *= 255.0 / np.max(mag) # normalize (Q&D)
col[i] = mag
return col
def harris_feature(im, region_size=5, to_return='harris', scale=0.05):
"""
Harris-motivated feature detection on a d-dimensional image.
Parameters
---------
im
region_size
to_return : {'harris','matrix','trace-determinant'}
"""
ndim = im.ndim
#1. Gradient of image
grads = [nd.sobel(im, axis=i) for i in range(ndim)]
#2. Corner response matrix
matrix = np.zeros((ndim, ndim) + im.shape)
for a in range(ndim):
for b in range(ndim):
matrix[a, b] = nd.filters.gaussian_filter(grads[a] * grads[b],
region_size)
if to_return == 'matrix':
return matrix
#3. Trace, determinant
trc = np.trace(matrix, axis1=0, axis2=1)
det = np.linalg.det(matrix.T).T
if to_return == 'trace-determinant':
return trc, det
else:
#4. Harris detector:
harris = det - scale * trc * trc
return harris
def generate_features3D(image, sigma):
# generate range of sigmas
sigmas = range(sigma[0], sigma[1] + 1)
f_values = image.flatten()
f_sobel = scp.sobel(image).flatten()
f_gauss = np.zeros([len(image.flatten()), len(sigmas)])
f_dog = np.zeros([len(image.flatten()), len(sigmas) - 1])
idx = 0
for s in range(sigma[0], sigma[1] + 1):
# consider only Re part for gabor filter
f_gauss[:, idx] = scp.gaussian_filter(image, s).flatten()
if (idx != 0):
f_dog[:, idx - 1] = f_gauss[:, idx] - f_gauss[:, idx - 1]
idx += 1
f_max = scp.maximum_filter(image, sigma[0]).flatten()
f_median = scp.median_filter(image, sigma[0]).flatten() # run median only with the minimal sigma
f_laplacian = scp.laplace(image).flatten()
# full set of features
f_set = np.vstack([f_values, f_max,
f_median, f_sobel,
f_gauss.T, f_dog.T,
f_laplacian]).T
return f_set
def demo(image_path, model_path, device):
import os
import numpy as np
from matplotlib import pyplot as plt
from scipy.ndimage import sobel
from PIL import Image
import torch
from dextr.model import DextrModel
if device is None:
if torch.cuda.is_available():
device = 'cuda:0'
else:
device = 'cpu'
torch_device = torch.device(device)
# Load model
if model_path is not None:
# A model path was provided; load it
dextr_model = torch.load(model_path, map_location=torch_device)
else:
# No model path; download (if necessary) and load a pre-trained model
dextr_model = DextrModel.pascalvoc_resunet101().to(torch_device)
dextr_model.eval()
# Load image
if image_path is None:
image_path = os.path.join('images', 'giraffes_1.jpg')
image = Image.open(image_path)
img_arr = np.array(image) / 255.0
plt.ion()
plt.axis('off')
plt.imshow(img_arr)
plt.title('Click four extreme points of the object\nHit enter when done')
def overlay(image, colour, mask, alpha):
alpha_img = mask * alpha
return image * (1 - alpha_img[:, :, None]
) + colour[None, None, :] * alpha_img[:, :, None]
while True:
# Get points from user
extreme_points = np.array(plt.ginput(4, timeout=0))
if len(extreme_points) < 4:
# Less than four points; quit
break
# Predict masks (points come from matplotlib in [x,y] order; this must be flipped)
masks = dextr_model.predict([image], extreme_points[None, :, ::-1])
mask_bin = masks[0] >= 0.5
edges = sobel(mask_bin.astype(float)) != 0
img_arr = overlay(img_arr, np.array([0.0, 1.0, 0.0]), mask_bin, 0.3)
img_arr = overlay(img_arr, np.array([1.0, 1.0, 0.0]), edges, 0.3)
plt.imshow(img_arr)
plt.plot(extreme_points[:, 0], extreme_points[:, 1], 'gx')
def main(file1, file2):
mat = cv2.imread(file1)
nmat = cv2.imread(file2)
blurmat = cv2.GaussianBlur(mat, (5, 5), 3)
nblurmat = cv2.GaussianBlur(nmat, (25, 25), 50)
sx = ndimage.sobel(mat, axis=0, mode='constant')
sy = ndimage.sobel(mat, axis=1, mode='constant')
sobel = np.hypot(sx, sy).mean()
print(sobel)
sx = ndimage.sobel(nmat, axis=0, mode='constant')
sy = ndimage.sobel(nmat, axis=1, mode='constant')
sobel = np.hypot(sx, sy).mean()
print(sobel)
cv2.imshow("", blurmat)
cv2.imshow("n", nblurmat)
cv2.waitKey(0)
def featArray(data, timesObs):
freqs1 = n.linspace(100, 200, n.shape(data)[1])
sh = n.shape(data)
CW1mean = n.zeros_like(data)
for i in range(CW1mean.shape[1]):
CW1 = cwt(n.abs(data[:, i]), haar, n.arange(1, 10, 1))
CW1 = n.ma.masked_where(CW1 == 0, CW1)
CW1mean[:, i] = n.ma.mean(n.abs(CW1), 0)
CT1mean = n.zeros_like(data)
for j in range(CW1mean.shape[0]):
CT1 = cwt(data[j, :], signal.morlet, n.arange(1, 3, 1))
CT1 = n.ma.masked_where(CT1 == 0, CT1)
CT1mean[j, :] = n.mean(n.abs(CT1), 0)
processed = ndimage.sobel(n.abs(data))
X1 = n.zeros((sh[0] * sh[1], 5))
X1[:, 0] = (n.real(data)).reshape(sh[0] * sh[1])
X1[:, 1] = (n.imag(data)).reshape(sh[0] * sh[1])
#X1[:,2] = n.abs(CW1mean).reshape(sh[0]*sh[1])
X1[:, 2] = n.abs(CT1mean).reshape(sh[0] * sh[1])
# X1[:,2] = n.log10(n.abs(processed)).reshape(sh[0]*sh[1])
X1[:, 3] = (n.array([timesObs] * sh[1])).reshape(sh[0] * sh[1])
X1[:, 4] = (n.array([freqs1] * sh[0])).reshape(sh[0] * sh[1])
X1 = n.nan_to_num(X1)
# for m in range(n.shape(X1)[1]):
# X1[:,m] = X1[:,m]/X1[:,m].max()
X1 = normalize(X1, norm='l2', axis=1)
X1 = n.nan_to_num(X1)
return X1
def _compute_derivatives(image, mode='constant', cval=0):
"""Compute derivatives in axis directions using the Sobel operator.
Parameters
----------
image : ndarray
Input image.
mode : {'constant', 'reflect', 'wrap', 'nearest', 'mirror'}, optional
How to handle values outside the image borders.
cval : float, optional
Used in conjunction with mode 'constant', the value outside
the image boundaries.
Returns
-------
derivatives : list of ndarray
Derivatives in each axis direction.
"""
derivatives = [
ndi.sobel(image, axis=i, mode=mode, cval=cval)
for i in range(image.ndim)
]
return derivatives
def func_write_gradmag_img(imf, fout): im_dy = ndimage.sobel(imf, 0) im_dx = ndimage.sobel(imf, 1) mmin = np.minimum(np.min(im_dy), np.min(im_dx)) im_dy -= mmin im_dx -= mmin mmax = np.maximum(np.max(im_dy), np.max(im_dx)) im_dy = (im_dy / mmax)*255.0 im_dx = (im_dx / mmax)*255.0 im_dmag = np.sqrt(im_dx**2 + im_dy**2) im_dmag -= np.min(im_dmag) im_dmag = (im_dmag / np.max(im_dmag) )*255 im_d = np.stack((im_dmag, im_dx, im_dy), axis=2) im_d = im_d.astype(np.uint8) plt.imsave(fout, im_d)
def createPoints_sober(image1):
Ix = ndimage.sobel(image1, 0) # pre-calculate the derivatives
Iy = ndimage.sobel(image1, 1)
Iz = ndimage.sobel(image1, 2)
Ix2 = Ix * Ix
Iy2 = Iy * Iy
Iz2 = Iz * Iz
Ixy = Ix * Iy
Ixz = Ix * Iz
Iyz = Iy * Iz
k = 0.06
I_feature = (Ix2 * Iy2 * Iz2 + Ixy * Iyz * Ixz + Ixy * Iyz * Ixz -
Ixy * Ixy * Iz2 - Iyz * Iyz * Ix2 -
Ixz * Ixz * Iy2) - k * (Ix2 + Iy2 + Iz2)**3
return I_feature
def sample_points(im, step, mask=None, maxiter=20):
shape = np.asarray(im.shape, dtype=int)
steps = np.ones(im.ndim, dtype=int) * step
## make regular grid
grid = np.zeros(shape)
grid[::steps[0], ::steps[1], ::steps[2]] = 1
if mask is not None:
grid = grid * mask
points0 = np.argwhere(grid)
points = points0
## move points
fim = ndimage.gaussian_filter(im.astype(float), sigma=1.0)
emap = np.sqrt(
np.sum([ndimage.sobel(fim, axis=d)**2 for d in range(im.ndim)],
axis=0))
gradient = np.asarray(np.gradient(emap)).astype(float)
for i in range(np.max(steps) / 2):
axes = i < (steps / 2)
if i >= maxiter: break
dp = gradient[(slice(None), ) + tuple(points.T)].T
dp = dp / np.c_[np.maximum(np.max(np.abs(dp), axis=1), 1e-10)]
dp = (dp + 0.5 * (-1)**(dp < 0)).astype(int)
for axe in np.where(axes)[0]:
points[:, axe] = (points - dp)[:, axe]
points = points[np.all(points > 0, axis=1)
& np.all(points < shape, axis=1)]
return points
def edge_filters(self):
''' Plot five edge-filters (kernels) in grayscale
'''
self.gray = rgb2gray(self.im)
self.edges = {
'Original': self.im,
'Grayscale': self.gray,
'Sobel': ndimage.sobel(self.gray),
'Prewitt': ndimage.prewitt(self.gray),
'Laplacian': ndimage.laplace(self.gray, mode='reflect'),
'LoG': ndimage.gaussian_laplace(self.gray, sigma=1, mode='reflect')
}
fig, axes = plt.subplots(2, 3, figsize=(18, 10))
axs = iter(axes.ravel())
for name, edge in self.edges.items():
ax = next(axs)
ax.imshow(edge, cmap='gray')
ax.set_title(name)
fig.tight_layout()
plt.savefig('.'.join(FNAME.split('.')[:-1]) + '_processed.png')
def suction_cam_callback(image):
assert image.encoding == 'bgra8'
raw = np.fromstring(image.data, dtype=np.uint8).reshape(
(image.height, image.width, 4))
colorized = np.stack((raw[:, :, 2], raw[:, :, 1], raw[:, :, 0]), axis=2)
grayscale = rgb_to_grayscale(raw[:, :, 2], raw[:, :, 1],
raw[:, :, 0]).astype(np.uint8)
blurred = signal.fftconvolve(grayscale, GAUSSIAN_KERNEL, mode='same')
edges = ndimage.sobel(blurred)
dev_mean = np.mean(edges)
dev_bitmap = 1 - (np.abs(edges - dev_mean) > 10 * dev_mean).astype(np.int)
gray_mean = np.mean(grayscale)
gray_bitmap = ((grayscale - gray_mean) > 3 * gray_mean).astype(np.int)
combined_bitmap = dev_bitmap / dev_bitmap.max(
) + 2 * gray_bitmap / gray_bitmap.max()
# saturated = np.minimum(0.97*256, np.maximum(0.03*256, blurred)) # - 0.05*256)/0.9
# b, a = signal.butter(2, 0.001, 'highpass')
# deglared = signal.lfilter(b, a, grayscale)
# deglared = grayscale
# processed = blurred
global SHOWN
if not SHOWN:
SHOWN = True
# plt.imshow(colorized)
plt.imshow(combined_bitmap, cmap='gray')
plt.show()
def sobel_filter_image(self):
self.restore_original()
image = misc.fromimage(self._new_image)
image = image.astype('int32')
dx = ndimage.sobel(image, 1) # horizontal derivative
dy = ndimage.sobel(image, 0) # vertical derivative
mag = np.hypot(dx, dy) # magnitude
mag *= 255.0 / np.max(mag) # normalize (Q&D)
print(222222, mag)
self._new_image = Image.fromarray(mag)
show_preconfigured_hist(self._new_image.getdata())
self._view.update_ui(self._initial_image, self._new_image)
def sobel(self):
"""
Performs the sobel operator in x and y direction and combines the
output. The output is normalized by using contrast_stretch.
"""
gray = ImageUtils.rgb2grey_fixed(self.image_array) / 255
sobel_x = ndimage.sobel(gray, 0)
sobel_y = ndimage.sobel(gray, 1)
sobel_xy = numpy.hypot(sobel_x, sobel_y)
sobel_xy = ImageUtils.normalize_intensity_p298(sobel_xy)
# sobel_xy *= (255 / sobel_xy.max())
sobel_xy = ImageUtils.rgb2grey_fixed(sobel_xy)
self._update_image_data(sobel_xy, "Sobel edge-detection")
def genSimpleFeatures(volume):
return [
ndimage.prewitt(volume),
ndimage.sobel(volume)
] + \
genBlurSharpen(volume, 2.0) + \
genBlurSharpen(volume, 5.0)
def _get_frame_entropy((i, capture, sobelized)):
frame = capture.get_frame(i, True).astype('float')
if sobelized:
frame = ndimage.median_filter(frame, 3)
dx = ndimage.sobel(frame, 0) # horizontal derivative
dy = ndimage.sobel(frame, 1) # vertical derivative
frame = numpy.hypot(dx, dy) # magnitude
frame *= 255.0 / numpy.max(frame) # normalize (Q&D)
histogram = numpy.histogram(frame, bins=256)[0]
histogram_length = sum(histogram)
samples_probability = [float(h) / histogram_length for h in histogram]
entropy = -sum([p * math.log(p, 2) for p in samples_probability if p != 0])
return entropy
def run(self, ips, snap, img, para=None):
imgs = ips.imgs
gradient = np.zeros(imgs.shape, dtype=np.float32)
gradient += ndimg.sobel(imgs, axis=0, output=np.float32)**2
gradient += ndimg.sobel(imgs, axis=1, output=np.float32)**2
gradient += ndimg.sobel(imgs, axis=2, output=np.float32)**2
gradient **= 0.5
msk = np.zeros(imgs.shape, dtype=np.uint8)
msk[imgs > para['thr2']] = 1
msk[imgs < para['thr1']] = 2
#rst = watershed(gradient, msk)
rst = watershed(gradient, msk.astype(np.uint16))
imgs[:] = (rst == 1) * 255
ips.lut = self.buflut
def scrub(self, size=30):
gray = self.gray_image
binar = self.bin_image
visit_list = np.argwhere(gray <= (self.whitespace - 5))
last_3 = [-1, -1, -1]
# self.save_fig(os.path.join(RESULTS_DIRECTORY, '{}_before.png'.format(self.figname)))
for x in visit_list:
i = x[0] - 15
j = x[1] - 15
print(i, j)
# self.plot_frame(gray, i, j, size)
# self.plot_frame(binar, i, j, size)
# plt.show()
gray_window = gray[i:i + size, j:j + size]
bin_window = binar[i:i + size, j:j + size]
gray_window_sobel = ndimage.sobel(gray_window, axis=0)
mean_pixel_value = np.mean(gray_window)
gray_pixel_value = gray_window[15, 15]
colored_percentage = np.count_nonzero(bin_window == 0) / (30**2)
above_area = np.mean(gray_window_sobel[8:16, 15])
below_area = np.mean(gray_window_sobel[16:23, 15])
sobel_gradient = above_area - -below_area
# print('Pix Val: {}, Mean Pix Val: {}, Bin Colored: {}, Sobel Gradient: {}, Last 3: {}'.format(gray_pixel_value, mean_pixel_value, colored_percentage, sobel_gradient, last_3))
prediction = self.predict(gray_pixel_value, mean_pixel_value,
colored_percentage, sobel_gradient,
last_3)
print(prediction)
self.alter_image(i, j, prediction)
last_3.pop()
last_3.insert(0, prediction)
self.save_fig()
def paintLayer(self, ref_img, radius):
S = []
self.cur_nparray = self.Surface2array(self.canvas)
self.ref_nparray = self.Image2array(ref_img)
D = self.img_diff(self.cur_nparray, self.ref_nparray)
grid = int(self.style.grid_size * radius)
width = self.canvas.get_width() // grid * grid
height = self.canvas.get_height() // grid * grid
ref_l = self.Image2array(ref_img.convert(mode='I'))
self.gradient_x = ndimage.sobel(ref_l, 0)
self.gradient_y = ndimage.sobel(ref_l, 1)
cnt = 0
for x in range(0, width, grid):
for y in range(0, height, grid):
M = D[x:x + grid, y:y + grid]
areaError = M.sum() / (grid * grid)
if areaError > self.style.threshold:
cnt += 1
x1, y1 = np.unravel_index(np.argmax(M), M.shape)
s = self.makeSplineStroke(x1 + x, y1 + y, radius)
S.append(s)
random.shuffle(S)
print("radius=%d : stroke %d" % (radius, cnt))
for s in S:
self.context.set_line_width(
max(
self.context.device_to_user_distance(
2 * radius, 2 * radius)))
stroke_color = self.ref_nparray[s[0]] / 255
self.context.set_source_rgb(stroke_color[0], stroke_color[1],
stroke_color[2])
self.context.move_to(s[0][0] / self.canvas.get_width(),
s[0][1] / self.canvas.get_height())
for i in range(1, len(s)):
self.context.line_to(s[i][0] / self.canvas.get_width(),
s[i][1] / self.canvas.get_height())
self.context.move_to(s[i][0] / self.canvas.get_width(),
s[i][1] / self.canvas.get_height())
self.context.close_path()
self.context.stroke()
def single_scale_align_edge(img0, img1, offset):
"""Single scale aligns two images using edge detection
Similar to single_scale_align(), but also runs an edge detection
filter on top of the image before the search starts.
Aligns img1 to img0, which acts like the anchor.
We exhaustively search over a window of possible displacements given
by the offset (e.g. [-15,15] pixels), then score each offset image
using some image matching metric (the sum of squared distances) and
finally choose the displacement with the best score.
Args:
img0: Anchor image to be aligned to
img1: Image to be aligned
offset: List of two offset ranges. First is for green, second
is for red. Each offset range contains a x range first, then a
y range
Returns:
A single aligned img and the alignment array, which contains
alignemtn of green (index 0) and alignement of red (index 1)
"""
min_score = float("inf")
final_offset = []
# Edge detection
edge_sobel_0 = ndimage.sobel(img0, 1) # horizontal derivative
edge_sobel_1 = ndimage.sobel(img1, 1) # horizontal derivative
array0 = np.array(edge_sobel_0)
# imshow(img1)
imshow(concat_n_images([img0, edge_sobel_0, img1, edge_sobel_1]))
for x in range(offset[0][0], offset[0][1]):
for y in range(offset[1][0], offset[1][1]):
roll = np.roll(np.roll(edge_sobel_1, y, axis=0), x, axis=1)
score = np.sum(np.power(array0 - np.array(roll), 2))
if min_score >= score:
min_score = score
final_offset = (x, y)
aligned = np.roll(np.roll(img1, final_offset[1], 0), final_offset[0], 1)
#print("edge align offset", final_offset)
return (aligned, final_offset)
def LucasKanade(It, It1, rect, p0=np.zeros(2)):
# Input:
# It: template image
# It1: Current image
# rect: Current position of the car
# (top left, bot right coordinates)
# p0: Initial movement vector [dp_x0, dp_y0]
# Output:
# p: movement vector [dp_x, dp_y]
# Put your implementation here
dp = np.array([100.0, 100.0])
p = p0
converge = False
while (converge == False):
# Warp the image and compute error
x1, y1, x2, y2 = rect[0], rect[1], rect[2], rect[3]
p_x, p_y = p[0], p[1]
w = shift(It1, (p_y, p_x), It1.dtype)
It_w = w[y1:y2 + 1, x1:x2 + 1]
er = It_w - It
er = er.flatten()
# 3) Compute the gradient and warp them, then demarcate the border
dx = ndimage.sobel(It1, 1) # horizontal derivative
dy = ndimage.sobel(It1, 0) # vertical derivative
w_dx = shift(dx, (p_y, p_x), It1.dtype)
It_dx = w_dx[y1:y2 + 1, x1:x2 + 1]
It_dx = It_dx.flatten()
w_dy = shift(dy, (p_y, p_x), It1.dtype)
It_dy = w_dy[y1:y2 + 1, x1:x2 + 1]
It_dy = It_dy.flatten()
vstack = np.vstack((It_dx, It_dy))
er_dec = np.matmul(vstack, er) # Error of steepest descent
# Compute hessian and find delta p (dp)
l_mag = np.matmul(vstack.T, np.eye(2))
Hessian = np.matmul(l_mag.T, l_mag)
dp = np.matmul(inv(Hessian), er_dec)
p = dp + p
if (np.linalg.norm(dp) < 0.01):
converge = True
return p
def principal_curvature(image):
rows, cols = image.shape
gx = ndimage.sobel(image, axis=0, mode='constant')
gy = ndimage.sobel(image, axis=1, mode='constant')
gxx = ndimage.sobel(gx, axis=0, mode='constant')
gxy = ndimage.sobel(gx, axis=1, mode='constant') # same as gyx
gyy = ndimage.sobel(gy, axis=1, mode='constant')
lamdaplus = np.zeros(image.shape)
for i in range(rows):
for j in range(cols):
lamdaplus[i, j] = lamdafind([gxx[i, j], gxy[i, j], gxy[i, j], gyy[i, j]])
return img_as_float(lamdaplus)
def Wateredge(Water, cellsize, buffer):
# Lets use a sobel filter to find the water edges
sx = ndimage.sobel(Water, axis=0, mode='constant')
sy = ndimage.sobel(Water, axis=1, mode='constant')
sob = np.hypot(sx, sy)
borders = sob >= 1
# Use 2D convolution to create a buffer around the water edges
# first calculate number of cells needed for buffer based on cell size and desired buffer (20m)
buf = int(buffer / cellsize)
kernel = np.ones((buf, buf))
result = np.int64(convolve2d(borders, kernel, mode='same') > 0)
# Delete the water part so only the riperian zone is left
riparian = result - Water
return riparian
