def show_hough_transform(image):
h, theta, d = hough_line(canny(image))
fig, ax = plt.subplots(1, 3, figsize=(15, 6))
ax[0].imshow(image, cmap=cm.gray)
ax[0].set_title('Input image')
ax[0].set_axis_off()
ax[1].imshow(
np.log(1 + h),
extent=[np.rad2deg(theta[-1]),
np.rad2deg(theta[0]), d[-1], d[0]],
cmap='gray',
aspect=1 / 20)
ax[1].set_title('Hough transform')
ax[1].set_xlabel('Angles (degrees)')
ax[1].set_ylabel('Distance (pixels)')
ax[2].imshow(image, cmap=cm.gray)
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - image.shape[1] * np.cos(angle)) / np.sin(angle)
ax[2].plot((0, image.shape[1]), (y0, y1), '-r')
ax[2].set_xlim((0, image.shape[1]))
ax[2].set_ylim((image.shape[0], 0))
ax[2].set_axis_off()
ax[2].set_title('Detected lines')
plt.tight_layout()
return hough_line_peaks(h, theta, d)
def search_lines(blob,
angle_range,
npoints=1000,
min_distance=100,
min_angle=300,
threshold=None):
thetas = np.linspace(np.deg2rad(angle_range[0]),
np.deg2rad(angle_range[1]), npoints)
hspace, angles, distances = hough_line(blob, thetas)
if threshold is not None:
accum, angles, dists = hough_line_peaks(hspace,
angles,
distances,
min_distance=min_distance,
min_angle=min_angle,
threshold=threshold *
np.max(hspace))
else:
accum, angles, dists = hough_line_peaks(hspace,
angles,
distances,
min_distance=min_distance,
min_angle=min_angle)
return accum, angles, dists
def test_hough_line_peaks_dist():
img = np.zeros((100, 100), dtype=np.bool_)
img[:, 30] = True
img[:, 40] = True
hspace, angles, dists = tf.hough_line(img)
assert len(tf.hough_line_peaks(hspace, angles, dists, min_distance=5)[0]) == 2
assert len(tf.hough_line_peaks(hspace, angles, dists, min_distance=15)[0]) == 1
def test_hough_line_peaks_dist():
img = np.zeros((100, 100), dtype=np.bool_)
img[:, 30] = True
img[:, 40] = True
hspace, angles, dists = tf.hough_line(img)
with expected_warnings(['`background`']):
assert len(tf.hough_line_peaks(hspace, angles, dists,
min_distance=5)[0]) == 2
assert len(tf.hough_line_peaks(hspace, angles, dists,
min_distance=15)[0]) == 1
def test_hough_line_peaks_dist():
img = np.zeros((100, 100), dtype=np.bool_)
img[:, 30] = True
img[:, 40] = True
hspace, angles, dists = tf.hough_line(img)
with expected_warnings(['`background`']):
assert len(
tf.hough_line_peaks(hspace, angles, dists, min_distance=5)[0]) == 2
assert len(
tf.hough_line_peaks(hspace, angles, dists,
min_distance=15)[0]) == 1
def skew_correction(img, edges, threshold=None, _save=False, _save_path="./"):
# get HoughLinesMap
h, theta, d = hough_line(edges)
h_p, theta_p, d_p = None, None, None
# get peaks of the HoughLinesMap
if threshold is None:
h_p, theta_p, d_p = hough_line_peaks(h, theta, d)
else:
h_p, theta_p, d_p = hough_line_peaks(h, theta, d, threshold=threshold)
# get the histogram of the values of the angles
hist, bins = np.histogram(theta_p, 360)
# the most probable angle is complementary to the skew
c_skew = np.rad2deg(
(bins[np.argmax(hist)] + bins[np.argmax(hist) + 1]) / 2)
skew = 90 - c_skew if c_skew > 0 else -c_skew - 90
# create a rotation matrix and rotate the image around its center
matrix = cv2.getRotationMatrix2D((img.shape[1] / 2, img.shape[0] / 2),
-skew, 1)
dst = cv2.warpAffine(img, matrix, (img.shape[1], img.shape[0]),
cv2.INTER_NEAREST)
if _save:
_path_name = os.path.join(_save_path, "_skew_hough")
if not os.path.isdir(_path_name):
os.mkdir(_path_name)
hough_lines_all = img.copy()
hough_lines_true = img.copy()
diag = np.sqrt(np.power(img.shape[0], 2) + np.power(img.shape[1], 2))
for _, angle, dist in zip(*(h_p, theta_p, d_p)):
a, b = np.cos(angle), np.sin(angle)
x0, y0 = a * dist, b * dist
x1 = int(x0 + diag * (-b))
y1 = int(y0 + diag * (a))
x2 = int(x0 - diag * (-b))
y2 = int(y0 - diag * (a))
cv2.line(hough_lines_all, (x1, y1), (x2, y2), (0, 255, 0), 4)
if bins[np.argmax(hist)] <= angle <= bins[np.argmax(hist) + 1]:
cv2.line(hough_lines_true, (x1, y1), (x2, y2), (0, 255, 0), 4)
cv2.imwrite(os.path.join(_path_name, "pre_skew.jpg"), img)
cv2.imwrite(os.path.join(_path_name, "post_skew.jpg"), dst)
cv2.imwrite(os.path.join(_path_name, "hough_lines_all.jpg"),
hough_lines_all)
cv2.imwrite(os.path.join(_path_name, "hough_lines_true.jpg"),
hough_lines_true)
return dst, -np.deg2rad(skew)
def calculate_rotation(image):
# sometimes we snag corners, by cropping the left and right 10% of the image we focus only on the
# vertical bars formed by the structure
height, width = image.shape
crop = int(width * 0.1)
cropped_image = image[:, crop: width - crop]
# Find edges that have a strong vertical direction
vertical_edges = sobel_v(cropped_image)
# Separate out the areas where there is a large amount of vertically-oriented stuff
segmentation = segment_edge_areas(vertical_edges)
# Draw a line that follows the center of the segments at each point, which should be roughly vertical
# We should expect this to give us four approximately-vertical lines, possibly with many gaps in
# each line
skeletons = skeletonize(segmentation)
# Use the Hough transform to get the closest lines that approximate those four lines
hough = transform.hough_line(skeletons, np.arange(-constants.FIFTEEN_DEGREES_IN_RADIANS,
constants.FIFTEEN_DEGREES_IN_RADIANS,
0.0001))
# Create a list of the angles (in radians) of all of the lines the Hough transform produced, with 0.0
# being completely vertical
# These angles correspond to the angles of the four sides of the channels, which we need to
# correct for
angles = [angle for _, angle, dist in zip(*transform.hough_line_peaks(*hough))]
if not angles:
raise ValueError("Image rotation could not be calculated. Check the images to see if they're weird.")
else:
# Get the average angle and convert it to degrees
offset = sum(angles) / len(angles) * 180.0 / math.pi
if offset > constants.ACCEPTABLE_SKEW_THRESHOLD:
log.warn("Image is heavily skewed. Check that the images are valid.")
return offset
def rotate_pic(pic):
"""
rotate_pic finds the offset angle by measuring the grid with a Hough line transform. It then de-rotates by this amount
"""
# for speed, only look at a 500x500 slice in the center of the image
centercut = np.copy(
pic[int(pic.shape[0] / 2.) - 500:int(pic.shape[0] / 2.) + 500,
int(pic.shape[1] / 2.) - 500:int(pic.shape[1] / 2.) + 500, 0])
# find the grid in the image by looking for brighter regions
centermask = (centercut < 1) * (centercut > 0.8)
h, theta, d = tf.hough_line(centermask,
theta=np.linspace(np.pi / 2 - 0.2,
np.pi / 2 + 0.2, 1000))
_, angles, dist = tf.hough_line_peaks(h, theta, d)
# plt.figure()
# plt.imshow(centermask)
# for ii in range(len(angles)):
# rho = dist[ii]
# theta = angles[ii]
#
# y0 = (rho - 0 * np.cos(theta)) / np.sin(theta)
# y1 = (rho - centermask.shape[1] * np.cos(theta)) / np.sin(theta)
#
# plt.plot([0,centermask.shape[0]],[y0,y1],'r-')
angle = np.mean(angles * 180 / np.pi - 90)
corrected_pic = tf.rotate(pic, angle, cval=1)
return corrected_pic, angle
def test_hough_line_peaks_zero_input():
# Test to make sure empty input doesn't cause a failure
img = np.zeros((100, 100), dtype='uint8')
theta = np.linspace(0, np.pi, 100)
hspace, angles, dists = transform.hough_line(img, theta)
h, a, d = transform.hough_line_peaks(hspace, angles, dists)
assert_equal(a, np.array([]))
def compute_hough_transform(line_det_map1, line_det_map2, re_focus_angle=True):
# focus theta for cuneiform horizontal lines
theta_range = np.linspace(np.deg2rad(83), np.deg2rad(97), 50)
# theta_range = np.linspace(np.deg2rad(-90) ,np.deg2rad(90), 180) # normal range
# Classic straight-line Hough transform (usually angles from -90 to +90)
h, theta, d = hough_line(line_det_map1, theta=theta_range)
# debug
# plt.imshow(np.log(1 + h), extent=[np.rad2deg(theta[-1]), np.rad2deg(theta[0]), d[-1], d[0]], cmap='gray', aspect=1/1.5)
# plt.show()
# focus angle and re-run
if re_focus_angle:
# get peaks
accum, angles, dists = hough_line_peaks(h,
theta,
d,
min_distance=1,
min_angle=16,
num_peaks=50)
# get median angle
m_angle = np.median(np.rad2deg(angles))
# modify theta
theta_range = np.linspace(np.deg2rad(m_angle - 2),
np.deg2rad(m_angle + 2), 50)
theta_range2 = np.linspace(np.deg2rad(m_angle - 3),
np.deg2rad(m_angle + 3), 50)
# Classic straight-line Hough transform (usually angles from -90 to +90)
h, theta, d = hough_line(line_det_map2, theta=theta_range)
return h, theta, d, theta_range, theta_range2
def jet_detect(img, calibratemean, calibratestd, x):
mean, std = image_stats(img)
# compare mean & calibratemean
if (mean < calibratemean * 0.8) or (mean > calibratemean * 1.2):
print('no jet')
for c in range(x):
try:
# binary = (img / (mean + 2 * std * 0.90 ** c)).astype(np.uint8)
# binary = cv2.adaptiveThreshold(img, 1, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv.THRESH_BINARY, 50, 2)
# binary = cv2.bitwise_not(imagem)
# lines = cv2.HoughLines(binary, 1, np.radians(0.25), 30)
# rho, theta = lines[0][0]
binary = canny(img,
sigma=2,
use_quantiles=True,
low_threshold=0.9,
high_threshold=0.99)
fig, (ax1, ax2) = plt.subplots(nrows=1,
ncols=2,
figsize=(8, 3),
sharex=True,
sharey=True)
ax1.imshow(img, cmap=plt.cm.gray)
ax2.imshow(binary, cmap=plt.cm.gray)
plt.show()
h, theta, d = hough_line(binary)
accum, angles, ds = hough_line_peaks(h, theta, d, min_distance=2)
print(accum)
print(angles)
print(ds)
# if (theta > math.radians(45)) and (theta < math.radians(135)):
# print('invalid jet')
# if (get_jet_width(img, rho, theta) * pxsize > 0.1):
# print('invalid jet')
# for rho, theta in lines[0]:
# jetValid = true
# if (theta > math.radians(70)):
# jetValid = False
# width = get_jet_width(binary, rho, theta)
# if (width > [x]):
# jetValid = False
# if (jetValid == False):
# reject jet
except Exception:
print(c)
continue
else:
# show binary image
# cv2.imshow('binary', binary)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# return rho, theta
# return 0, 0
return angles[0], ds[0]
raise ValueError('unable to detect jet')
def get_orientation(image, debug=False):
bin_image = (image > threshold_li(image)) * 1
dilated = scipy.ndimage.morphology.binary_dilation(bin_image,
iterations=30)
labeled, num_regions = mh.label(dilated)
sizes = mh.labeled.labeled_size(labeled)
mh.labeled.labeled_size(labeled)
if len(sizes) > 2:
too_small = np.where(sizes < np.flip(np.sort(sizes))[1])
labeled = mh.labeled.remove_regions(labeled, too_small)
labeled = (labeled / np.max(np.unique(labeled))).astype(np.int)
skeleton = skeletonize(labeled)
h, theta, d = hough_line(skeleton)
origin = np.array((0, skeleton.shape[1]))
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0, y1 = (dist - origin * np.cos(angle)) / np.sin(angle)
gradient = (y0 - y1) / origin[1]
if debug == False:
return math.degrees(np.arctan(gradient))
elif debug == True:
f, ax = plt.subplots(figsize=(40, 20))
plt.imshow(bin_image)
plt.show()
f, ax = plt.subplots(figsize=(40, 20))
plt.imshow(dilated)
plt.show()
f, ax = plt.subplots(figsize=(40, 20))
plt.imshow(skeleton)
plt.show()
return math.degrees(np.arctan(gradient))
def calculate(self, image: np.ndarray, disk_size: int=9,
mean_threshold: int=100, min_object_size: int=750) -> float:
# Find edges that have a strong vertical direction
vertical_edges = sobel_v(image)
# Separate out the areas where there is a large amount of vertically-oriented stuff
segmentation = self._segment_edge_areas(vertical_edges, disk_size, mean_threshold, min_object_size)
# Draw a line that follows the center of the segments at each point, which should be roughly vertical
# We should expect this to give us four approximately-vertical lines, possibly with many gaps in
# each line
skeletons = skeletonize(segmentation)
# Use the Hough transform to get the closest lines that approximate those four lines
hough = transform.hough_line(skeletons, np.arange(-constants.FIFTEEN_DEGREES_IN_RADIANS,
constants.FIFTEEN_DEGREES_IN_RADIANS,
0.0001))
# Create a list of the angles (in radians) of all of the lines the Hough transform produced, with 0.0
# being completely vertical
# These angles correspond to the angles of the four sides of the channels, which we need to
# correct for
angles = [angle for _, angle, dist in zip(*transform.hough_line_peaks(*hough))]
if not angles:
raise ValueError("Image rotation could not be calculated. Check the images to see if they're weird.")
else:
# Get the average angle and convert it to degrees
offset = sum(angles) / len(angles) * 180.0 / math.pi
if offset > constants.ACCEPTABLE_SKEW_THRESHOLD:
log.warn("Image is heavily skewed. Check that the images are valid.")
return offset
def detect_vertical_lines(data,
max_deviation_from_vertical=1,
row_begin=100,
row_end=-100,
min_distance=50):
if "cropped_original" in data:
image = data["cropped_original"]
else:
image = data["original"]
image = np.gradient(image)[1] #Vertical component of gradient
image = image[row_begin:row_end, :]
h, theta, d = hough_line(image) #hough transform
h[:, :90 -
max_deviation_from_vertical] = 0 #Mask out areas that relate to non vertical lines
h[:, 90 + max_deviation_from_vertical:] = 0
h = np.log(1 + h) #log scale for better visibility
fx = []
for _, angle, dist in zip(
*hough_line_peaks(h, theta, d, min_distance=min_distance)):
y0 = (dist) / np.sin(angle) #y for x = 0
y1 = (dist - image.shape[1] * np.cos(angle)) / np.sin(
angle) #y for x = image.shape[1]
intercept = y0
slope = (y1 - y0) / float(image.shape[1])
fx.append([slope, intercept])
data["houghlines"] = fx
#ax.plot(x, intercept+x*slope, '-r')
return data
def hough_transform(self, err=np.pi * 1 / 12, threshold=0.49, ks=3):
# -------------------- scikit-image hough line transform --------------------
theta = np.linspace(-np.pi * 1 / 4, np.pi * 3 / 4, 180)
h, theta, distance = hough_line(self.edges_img, theta)
hits, phi, rho = hough_line_peaks(h,
theta,
distance,
min_distance=10,
min_angle=50,
threshold=threshold * h.max(),
num_peaks=np.inf)
lines = list(zip(phi, rho))
# -------------------- OpenCV hough line transform --------------------
# lines = cv2.HoughLines(self.edges_img.astype(np.uint8), 1, np.pi / 180, threshold).reshape(-1, 2)
# lines = list(map(lambda x: tuple(x), lines[:, ::-1].tolist()))
# -------------------- discriminate between vertical and horizontal lines --------------------
self.lines = dict(v=[], h=[], o=[])
for ix, (phi, _) in enumerate(lines):
if abs(phi) < err or abs(phi - np.pi) < err:
# vertical
self.lines['v'].append(lines[ix])
elif abs(phi - np.pi / 2) < err:
# horizontal
self.lines['h'].append(lines[ix])
else:
# irrelevant lines
self.lines['o'].append(lines[ix])
return self.lines
def hough_transform(binary):
try:
h, angles, d = hough_line(binary)
res = hough_line_peaks(h,
angles,
d,
min_distance=1,
threshold=int(binary.shape[0] / 3))
except Exception:
raise ValueError('ERROR hough: not a jet')
fig, axes = plt.subplots(1, 2)
axes[0].imshow(binary)
axes[1].imshow(binary)
valid = []
for _, theta, dist in zip(*res):
jetValid = True
if (theta < np.radians(-45)) or (theta > np.radians(45)):
print('ERROR angle: not a jet')
jetValid = False
yint = dist / np.sin(theta)
xint = np.tan(theta) * yint
if (dist < 0) or (xint > binary.shape[1]):
print('ERROR xint: not a jet')
jetValid = False
if (jetValid):
y0 = (dist - 0 * np.cos(theta)) / np.sin(theta)
y1 = (dist - binary.shape[1] * np.cos(theta)) / np.sin(theta)
axes[1].plot((0, binary.shape[1]), (y0, y1), 'r')
valid.append([theta, dist])
axes[1].set_xlim((0, binary.shape[1]))
axes[1].set_ylim((binary.shape[0], 0))
return valid
def y_coordinate_extractor(img_arr):
middle_y_coor = 0.5 * img_arr.shape[0]
image = rgb2gray(img_arr)
image = np.asarray(image > 0.5, dtype=np.float32)
edges = filters.sobel_h(image)
h, theta, d = hough_line(edges)
upper_y = img_arr.shape[0]
lower_y = 0
thres_upp = img_arr.shape[0]
thres_low = img_arr.shape[0]
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - image.shape[1] * np.cos(angle)) / np.sin(angle)
avg = (y0 + y1) / 2
if avg - middle_y_coor < thres_low and avg - middle_y_coor > 0:
lower_y = avg
thres_low = avg - middle_y_coor
if middle_y_coor - avg < thres_upp and middle_y_coor - avg > 0:
upper_y = avg
thres_upp = middle_y_coor - avg
return upper_y, lower_y
def filter_errors(image_folder, image_list, threshold=0.5, plot=False):
"""
This function filters errenous images out of an the image_list. It is based
on the notion that erroneous images show slightly of horizontal bands over
the image. Hough line detection is used to detect these lines. The erroneous
images are separated from the correct images by thresholding the standard
deviation over the line anlges.
Parameters
----------
image_folder : string
path to folder where images are located
image_list : list
list of image filenames (strings)
threshold : float
threshold on standard deviation of angles, under this threshold the
images are considered erroneous. The default is 0.5.
plot : boolean, optional
if true the detected lines are plotted. Only for 1 image.
Returns
-------
errors : list
list with the erroneous image filenames (strings)
fig : matplotlib Figure, only if plot = True
figure with the lines detected in the image
Tested
------
bad_images = ['40_a.npy', '163_a.npy', '163_b.npy', '269_a.npy', '313_b.npy', '122_b.npy', '277_a.npy', '328_b.npy', '311_a.npy', '184_b.npy', '398_b.npy', '158_a.npy', '356_b.npy']
good_images = ['240_a.npy', '451_b.npy', '558_b.npy','617_a.npy', '61_b.npy','267_b.npy','490_b.npy','243_b.npy', '685_b.npy','696_a.npy']
good_images = ['318_a.npy', '527_b.npy', '100_a.npy', '100_b.npy', '101_a.npy', '101_b.npy', '102_a.npy', '102_b.npy', '103_a.npy', '103_b.npy', '104_a.npy', '104_b.npy', '105_a.npy','105_b.npy', '107_a.npy']
"""
errors = []
for image in image_list:
im = np.load((os.path.join(image_folder, image)))[:, :, 0]
# edge filter using canny algorithm
edges = canny(im, sigma=6)
# detect lines using hough transform
test_angles = np.linspace(np.pi / 4, 7 * np.pi / 4 / 2, 360 / 4)
h, theta, d = hough_line(edges, theta=test_angles)
accum, angle, dist = hough_line_peaks(h, theta, d)
if plot:
assert len(image_list) == 1, "plot can only be used for 1 image"
fig = plot_detectedlines(im, h, theta, d)
# filter based on standard deviation of angles
std = np.std(angle)
print(std)
if std < threshold:
errors.extend([image])
if plot:
return errors, fig
else:
return errors
def manipExtract(image, thetaInit, method='standard'):
if np.issubdtype(image.dtype, 'bool'):
edge = image
else:
edge = canny_cv(image)
rows, cols = image.shape
h, theta, d = hough_line(edge,
theta=np.arange(thetaInit - .3, thetaInit + .3,
.01))
try:
_, angle, dist = hough_line_peaks(h,
theta,
d,
min_distance=1,
num_peaks=1)
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
except IndexError: # i think this error is being thrown if the manipulator is not found?
y0 = np.NaN
y1 = np.NaN
angle = np.NaN
dist = np.NaN
return y0, y1, angle, dist
def test_hough_line_peaks_zero_input():
# Test to make sure empty input doesn't cause a failure
img = np.zeros((100, 100), dtype='uint8')
theta = np.linspace(0, np.pi, 100)
hspace, angles, dists = transform.hough_line(img, theta)
h, a, d = transform.hough_line_peaks(hspace, angles, dists)
assert_equal(a, np.array([]))
def show_hough_transform(image, filename):
low_dark = (0, 0, 0)
high_dark = (50, 50, 50)
only_dark = cv2.inRange(img, low_dark, high_dark)
h, theta, d = hough_line(canny(only_dark)) # вычисляем преобразование Хаффа от границ изображения
fig, ax = plt.subplots(1, 3, figsize=(15, 6))
ax[0].imshow(only_dark, cmap=cm.gray)
ax[0].set_title('Input image')
ax[0].set_axis_off()
ax[1].imshow(np.log(1 + h),
extent=[np.rad2deg(theta[-1]), np.rad2deg(theta[0]), d[-1], d[0]],
cmap='gray', aspect=1/20)
ax[1].set_title('Hough transform')
ax[1].set_xlabel('Angles (degrees)')
ax[1].set_ylabel('Distance (pixels)')
ax[2].imshow(image, cmap=cm.gray)
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - image.shape[1] * np.cos(angle)) / np.sin(angle)
ax[2].plot((0, image.shape[1]), (y0, y1), '-r')
ax[2].set_xlim((0, image.shape[1]))
ax[2].set_ylim((image.shape[0], 0))
ax[2].set_axis_off()
ax[2].set_title('Detected lines')
plt.tight_layout()
plt.savefig(filename, bbox_inches='tight')
plt.close(fig)
def deskew(image, edge=True):
"""Rotate image so that lines will be horizontally aligned.
:param image: ndarray, 2D binarized image
:return: ndarray, 2D deskewed image
"""
rows, cols = image.shape
image2 = canny(image, 5) if edge else image
plt.imshow(image2, cmap='gray')
plt.show()
plt.imshow(image, cmap='gray')
skew = 0
_, angles, dists = hough_line_peaks(*hough_line(image2))
for angle, dist in zip(angles, dists):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
plt.plot((0, cols), (y0, y1), '-r')
skew += 90 + np.rad2deg(angle) if angle < 0 else np.rad2deg(angle) - 90
skew /= angles.shape[0] if angles.shape[0] != 0 else 1
print(skew)
plt.xlim([0, cols])
plt.ylim([rows, 0])
plt.show()
return rotate(image,
skew,
resize=True,
mode='constant',
cval=255,
preserve_range=True).astype(np.uint8)
def estimate_corner_lines(corner_region, corner_type, image_size):
'''
Estimate the line parameters for the edges of the given marker corner
Inputs:
- corner_region: regionprops for the given corner of the marker
- corner_type: either 'BL' for bottom-left or 'TR' for top-right
- image_size: tuple of original camera image's size
Return:
A list of length 2 holding the line parameters of horizontal and vertical edges of the marker
'''
corner_miny, corner_minx, corner_maxy, corner_maxx = corner_region.bbox
# Pad the corner image to match the size of the entire image
corner_image = util.pad(corner_region.intensity_image,
((corner_miny, image_size[0] - corner_maxy),
(corner_minx, image_size[1] - corner_maxx)),
mode='constant',
constant_values=0)
# Perform edge detection and Hough line transform
corner_edges = feature.canny(corner_image, sigma=0)
corner_hough = transform.hough_line(corner_edges,
theta=np.linspace(
-np.pi / 2, np.pi / 2, 360))
corner_hough_peaks = list(
zip(*transform.hough_line_peaks(
*corner_hough, min_angle=45, num_peaks=2)))
corner_lines = []
def is_horizontal(peak):
return abs(np.rad2deg(peak[1])) > 75
def is_vertical(peak):
return abs(np.rad2deg(peak[1])) <= 15
# Categorized the detected lines as vertical or horizontal
horizontal_peaks = list(filter(is_horizontal, corner_hough_peaks))
vertical_peaks = list(filter(is_vertical, corner_hough_peaks))
# Add the first estimated horizontal line
if horizontal_peaks:
corner_lines.append(horizontal_peaks[0])
else:
# Create a horizontal line from the bottom edge (if BL type) or top edge (if TR type)
angle = np.pi / 2
dist = corner_maxy if corner_type == 'BL' else corner_miny
corner_lines.append((0, angle, dist))
# Add the first estimated vertical line
if vertical_peaks:
corner_lines.append(vertical_peaks[0])
else:
# Create a vertical line from the left edge (if BL type) or right edge (if TR type)
angle = 0.001
dist = corner_minx if corner_type == 'BL' else corner_maxx
corner_lines.append((0, angle, dist))
return corner_lines
def is_arrow_check(img_edges):
# Find lines using hough transform
hspace, theta, dists = transform.hough_line(img_edges)
# Extract lines
lines = transform.hough_line_peaks(hspace, theta, dists, num_peaks=4)
# Convert lines to list of lines
lines = np.array(list(zip(*lines)), dtype=[('hspace', np.uint64), ('angle', np.float), ('dist', np.float)])
if len(lines) != 4: return False
lines = list(lines)
dists = []
while lines:
found = False
for i in range(1, len(lines)):
if angle_difference(lines[0], lines[i]) < 0.1:
# Found a line parallel to lines[0]
dists.append(abs(lines[0][2] - lines[i][2]))
del lines[i]
del lines[0]
found = True
break
if not found: break
if len(dists) == 2 and abs(12.0 - dists[0]) < 5.0 and abs(12.0 - dists[1]) < 5.0:
return True
return False
def get_limits(self, max_width_proportion=0.45):
"""Busca dos lineas paralelas que correspondan al bajalenguas
Args:
max_width_proportion (float): Maxima proporcion del alto de la
imagen (considerada apaisada) que puede abarcar el ancho del
bajalenguas.
Returns:
status (str): Una descripcion del resultado de la busqueda.
limits ([angles, dists]): Contiene dos numpy arrays. El primero
contiene dos angulos y el segundo dos distancias. Cada par de
angulo-distancia define la recta de arriba y de abajo del bajalenguas.
"""
max_angle_diff = 5. / 180 * np.pi
im_width = np.amin(self.curr_im_lowres_g.shape)
min_dist = int(1. / 6 * im_width)
sigma = 3
edges = canny(self.curr_im_lowres_g, sigma)
while np.mean(edges) < 0.01:
sigma = (sigma - 0.1)
if sigma < 0:
break
edges = canny(self.curr_im_lowres_g, sigma)
self.edges = edges
h, theta, d = hough_line(edges)
params = hough_line_peaks(h,
theta,
d,
num_peaks=6,
min_distance=min_dist)
self.params = params
# Normalizo al ancho de la imagen
dists = params[2] / im_width
angles = params[1]
dangles = pairwise_distances(angles[:, None])
dangles = np.dstack((dangles, np.abs(dangles - np.pi)))
dangles = np.amin(dangles, 2)
np.fill_diagonal(dangles, np.inf)
i, j = np.unravel_index(np.argmin(dangles), dangles.shape)
angles = np.array([angles[i], angles[j]])
dists = np.array([dists[i], dists[j]])
# Ordeno los bordes para que el de arriba quede primero
norm_dist = np.sign(angles) * dists
sort_idx = np.argsort(norm_dist)
if i == j:
status = 'Sin bajalenguas'
limits = None
elif dangles[i, j] > max_angle_diff:
status = 'Sin bajalenguas - mal paralelismo'
limits = None
elif abs(np.diff(norm_dist)) > max_width_proportion:
status = 'Sin bajalenguas - mal ratio'
limits = None
elif abs(angles[0]) < 20. / 180 * np.pi:
status = 'Sin bajalenguas - mala inclinacion'
limits = None
else:
status = 'Con bajalenguas'
limits = [angles[sort_idx], dists[sort_idx]]
return status, limits
def _determine_rotation_offset(image):
"""
Finds rotational skew so that the sides of the central trench are (nearly) perfectly vertical.
:param image: raw image data in a 2D (i.e. grayscale) numpy array
:type image: np.array()
"""
segmentation = create_vertical_segments(image)
# Draw a line that follows the center of the segments at each point, which should be roughly vertical
# We should expect this to give us four approximately-vertical lines, possibly with many gaps in each line
skeletons = skeletonize(segmentation)
# Use the Hough transform to get the closest lines that approximate those four lines
hough = transform.hough_line(
skeletons,
np.arange(-Constants.FIFTEEN_DEGREES_IN_RADIANS,
Constants.FIFTEEN_DEGREES_IN_RADIANS, 0.0001))
# Create a list of the angles (in radians) of all of the lines the Hough transform produced, with 0.0 being
# completely vertical
# These angles correspond to the angles of the four sides of the channels, which we need to correct for
angles = [
angle for _, angle, dist in zip(*transform.hough_line_peaks(*hough))
]
if not angles:
log.warn(
"Image skew could not be calculated. The image is probably invalid."
)
return 0.0
else:
# Get the average angle and convert it to degrees
offset = sum(angles) / len(angles) * 180.0 / math.pi
if offset > Constants.ACCEPTABLE_SKEW_THRESHOLD:
log.warn(
"Image is heavily skewed. Check that the images are valid.")
return offset
def removeChessboard(img):
# Get the major lines in the image
edges, dilatedEdges, (h, theta, d) = findLines(img)
# Create image with ones to fill inn lines
lines = np.ones(img.shape[:2])
# Add lines to image as zeroes
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - img.shape[1] * np.cos(angle)) / np.sin(angle)
x, y = line(int(y1), 0, int(y0), img.shape[1] - 1)
x = np.clip(x, 0, img.shape[0] - 1)
y = np.clip(y, 0, img.shape[1] - 1)
lines[x, y] = 0
# Remove border edges from image with all edges
w = 4
edges = np.pad(edges[w:img.shape[0] - w, w:img.shape[1] - w],
w,
mode='constant')
# Erode the lines bigger, such that they cover the original lines
lines = erosion(lines, square(13))
# Remove major lines and close shape paths
removedChessboard = closing(edges * lines, square(8))
return removedChessboard
def align_hough(input_Image):
#Converte a imagem do padrão RGB para uma imagem binária
monImage = color.rgb2gray(input_Image)
monImage = feature.canny(monImage)
houghTransform, angles, d = transform.hough_line(monImage,
theta=np.linspace(
-np.pi / 2, np.pi / 2,
180))
hTP, angles, d = transform.hough_line_peaks(houghTransform, angles, d)
angles = np.rad2deg(angles) + 90
angles = np.round(angles, 0)
angles = angles.astype(int)
angle = np.bincount(angles).argmax()
if (angle > 90 and angle < 180):
angle = angle + 180
transformedImage = transform.rotate(input_Image, angle, resize=True)
#Pós processamento para remoção das bordas resultantes da rotação
transformedImage = border_remove(transformedImage)
return transformedImage, angles[0]
def classic_hough_line(img):
edges = canny(img, sigma=10, low_threshold=0.05, high_threshold=0.10)
# Classic straight-line Hough transform
h, theta, d = hough_line(edges, )
# Generating figure 1
fig, axes = plt.subplots(1, 3, figsize=(15, 6))
ax = axes.ravel()
ax[0].imshow(img, cmap=cm.gray)
ax[0].set_title('Input image')
ax[0].set_axis_off()
ax[1].imshow(
np.log(1 + h),
extent=[np.rad2deg(theta[-1]),
np.rad2deg(theta[0]), d[-1], d[0]],
cmap=cm.gray,
aspect=1 / 1.5)
ax[1].set_title('Hough transform')
ax[1].set_xlabel('Angles (degrees)')
ax[1].set_ylabel('Distance (pixels)')
ax[1].axis('image')
ax[2].imshow(img, cmap=cm.gray)
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - img.shape[1] * np.cos(angle)) / np.sin(angle)
ax[2].plot((0, img.shape[1]), (y0, y1), '-r')
ax[2].set_xlim((0, img.shape[1]))
ax[2].set_ylim((img.shape[0], 0))
ax[2].set_axis_off()
ax[2].set_title('Detected lines')
plt.tight_layout()
plt.show()
def houghTransform(s, sThreshold):
"""
DESCRIPTION:
This function computes the hough transform, for determining the shock front.
see: https://scikit-image.org/docs/dev/auto_examples/edges/plot_line_hough_transform.html
INPUT:
s - the 2D density field ln(\rho/\rho_0)
sThreshold - the threshold for the field. Densities above this threshold
will be used to identify the shock front
OUTPUT:
x - x coordinate of the Hough transform
y - y coordinate of the Hough transform
xMin - x coordinate of the window around the shock front
yMin - y coordinate of the window around the shock front
sMask - the mask, s.max() * sThreshold (for checking)
"""
print("Calculating the Hough Transform.")
# Create a mask on s for detecting the ionisation front
sMask = s > s.max() * sThreshold
# run a hough transform on the masked density data
dtheta = 0.05 # the d\theta around a vertical line approximation for the ion. front.
# create a sample of test angles close to a vertical line
tested_angles = np.linspace(np.pi + dtheta, np.pi - dtheta, 90)
# run the Hough transform
origin = np.array((0, sMask.shape[1])) # define the origin coordinate
h, theta, d = hough_line(sMask,
theta=tested_angles) # define the H. transform
# pick the most dominant line from the H. transform
hspace, angle, dist = hough_line_peaks(h, theta,
d) # extract param. values
y0, y1 = (dist[0] - origin * np.cos(angle[0])) / np.sin(angle[0])
# pick out the cooridinates in the density field from the line and create a window
x, y = sampleHough(origin, [y0, y1])
xMin = x - windowSize
xMax = x + windowSize
# Just taking a single value at the for the window
xMin = xMin[0]
xMax = xMax[0]
# Make sure the window size isn't larger than the actual data domain
if xMin < 0:
xMin = 0
elif xMin > sMask.shape[0] - 1:
xMin = sMask.shape[0] - 1
# and for xMax too
if xMax < 0:
xMax = 0
elif xMax > sMask.shape[0] - 1:
xMax = sMask.shape[0] - 1
return x, y, xMin, xMax, sMask
def removeChessboard(img):
# Get the major lines in the image
edges, dilatedEdges, (h, theta, d) = findLines(img)
# Create image with ones to fill inn lines
lines = np.ones(img.shape[:2])
# Add lines to image as zeroes
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - img.shape[1] * np.cos(angle)) / np.sin(angle)
x, y = line(int(y1), 0, int(y0), img.shape[1] - 1)
x = np.clip(x, 0, img.shape[0] - 1)
y = np.clip(y, 0, img.shape[1] - 1)
lines[x, y] = 0
# Remove border edges from image with all edges
w = 4
edges = np.pad(edges[w:img.shape[0] - w, w:img.shape[1] - w], w, mode='constant')
# Erode the lines bigger, such that they cover the original lines
lines = erosion(lines, square(13))
# Remove major lines and close shape paths
removedChessboard = closing(edges * lines, square(8))
return removedChessboard
def alignRodCellvertically(
Image,
ifbinary=False,
thres=0.4, #thres value for binary image
plotting=False, # plotting the detected rod-direction
inverse=False):
"""
automatically align rod-shaped cells vertically
"""
#image = util.ReadImg(Imagename).astype(np.uint8)
#grey = cv2.cvtColor(rod,cv2.COLOR_RGB2GRAY)
if ifbinary:
binaryimage = Image
else:
if inverse:
dummy, binaryimage = cv2.threshold(
cv2.cvtColor(Image, cv2.COLOR_BGR2GRAY), 255 * thres, 255,
cv2.THRESH_BINARY_INV)
else:
dummy, binaryimage = cv2.threshold(
cv2.cvtColor(Image, cv2.COLOR_BGR2GRAY), 255 * thres, 255,
cv2.THRESH_BINARY)
# detect the direction of rod using the hough line detection
# note, the angle unit is radian (1 radian = 57.2958 degree)
Hspace, Angle, Dist = transform.hough_line(binaryimage)
# find the peaks in the hough space
peaks = transform.hough_line_peaks(Hspace, Angle, Dist)
if len(peaks[0]) > 1:
raise RuntimeError(
'More than 1 directions exist, plz use a image that contains only 1 direction'
)
angle = peaks[1][0]
dist = peaks[2][0]
# now rotate the image based one the angle
radianTodegree = 57.29
rotatedImage = imutils.rotate(Image, angle * radianTodegree)
if plotting:
origin = np.array((0, binaryimage.shape[1]))
line = (dist - origin * np.cos(angle)) / np.sin(angle)
plt.subplot(1, 2, 1)
plt.imshow(Image)
plt.plot(origin, line, 'r-')
plt.ylim([0, binaryimage.shape[1]])
plt.xticks([])
plt.yticks([])
plt.title('detected directions')
plt.subplot(1, 2, 2)
plt.imshow(rotatedImage)
plt.xticks([])
plt.yticks([])
plt.title('align vertically')
return rotatedImage
def identify_based_on_lines(img_edges):
# Find lines using hough transform
hspace, theta, dists = transform.hough_line(img_edges)
# Extract lines
lines = transform.hough_line_peaks(hspace, theta, dists, num_peaks=10)
# Convert lines to list of lines
lines = np.array(list(zip(*lines)),
dtype=[('hspace', np.uint64), ('angle', np.float),
('dist', np.float)])
shape_likeness = np.zeros(Shape.count)
for i in range(1, Shape.count):
shape = Shape.iterator[i]
if shape == Shape.CAKE:
hval = 0.0
if len(lines) >= 3:
hval = abs(((lines[0][0] + lines[1][0]) * 0.5) - lines[2][0])
if hval >= 5.0:
shape_likeness[i] = 0.0
else:
shape_likeness[i] = 10.0
else:
shape_likeness[i] = abs(shape.line_count - len(lines))
return shape_likeness
def test_hough_line_peaks_single_angle():
# Regression test for gh-4814
# This code snippet used to raise an IndexError
img = np.random.random((100, 100))
tested_angles = np.array([np.pi / 2])
h, theta, d = transform.hough_line(img, theta=tested_angles)
accum, angles, dists = transform.hough_line_peaks(h, theta, d, threshold=2)
def Hough(image, filter='Canny', sigma=3, show=False):
'''
Read in an image and get the peak Hough line
Parameters:
image: The image to be read
filter: Set whether to use Canny (default) or Sobel edge detection
sigma: Sensitivity for Canny
show: boolean to show the plot Hough plot
Return:
max_peak: The horizon line corresponding to the max peak
edges: the edges
'''
img_gray = color.rgb2gray(image) # Grayscale
if (filter == 'Canny'):
# Edge detection
edges = feature.canny(img_gray, sigma=3) # Canny
if (filter == 'Sobel'):
edges = filters.sobel(img_gray) # Use Sobel
# Hough transformation
h, theta, d = hough_line(edges)
# Get the max_peak
max_peak = hough_line_peaks(h, theta, d, num_peaks=1)
if(show):
fig = plt.figure()
plt.imshow(np.log(1 + h), extent=[np.rad2deg(theta[-1]), np.rad2deg(theta[0]), d[-1], d[0]],
cmap=plt.cm.gray, aspect=1 / 1.5)
return max_peak, edges
def test_hough_line_peaks_num():
img = np.zeros((100, 100), dtype=np.bool_)
img[:, 30] = True
img[:, 40] = True
hspace, angles, dists = transform.hough_line(img)
assert len(transform.hough_line_peaks(hspace, angles, dists,
min_distance=0, min_angle=0,
num_peaks=1)[0]) == 1
def align_with_boarder(image, sigma=1):
edges = ft.canny(image, sigma=sigma)
# edges = abs(fil.sobel_v(image))
h, theta, d = tf.hough_line(edges)
a, rot_angle, c = tf.hough_line_peaks(h, theta, d, min_distance=0)
image = rotate(image, np.rad2deg(rot_angle[0]))
return image
def test_hough_line_peaks():
img = np.zeros((100, 150), dtype=int)
rr, cc = line(60, 130, 80, 10)
img[rr, cc] = 1
out, angles, d = tf.hough_line(img)
out, theta, dist = tf.hough_line_peaks(out, angles, d)
assert_equal(len(dist), 1)
assert_almost_equal(dist[0], 80.723, 1)
assert_almost_equal(theta[0], 1.41, 1)
def test_ideal_tfr(self):
"""Test if the ideal TFR can be found using the instantaneous frequency
laws."""
_, iflaw1 = fmlin(128, 0.0, 0.2)
_, iflaw2 = fmlin(128, 0.3, 0.5)
iflaws = np.c_[iflaw1, iflaw2].T
tfr, _, _ = pproc.ideal_tfr(iflaws)
tfr[tfr == 1] = 255
tfr = tfr.astype(np.uint8)
hspace, angles, dists = hough_line(tfr)
for x in hough_line_peaks(hspace, angles, dists):
self.assertEqual(len(x), 2)
def houghSides(bottle, edges, threshold, left):
h, theta, d = hough_line(edges)
accum = zip(*hough_line_peaks(h, theta, d))
sortedAccum = sorted(accum, key=getKey, reverse=True)
sortedAccumR = [sa for sa in sortedAccum if sa[2] > 200]
sortedAccumL = [sa for sa in sortedAccum if sa[2] <= 200]
if left:
hpeak, angle, dist = sortedAccumL[0]
else:
hpeak, angle, dist in sortedAccumR[0]
return (1, dist, angle)
def test_hough_line_peaks_ordered():
# Regression test per PR #1421
testim = np.zeros((256, 64), dtype=np.bool)
testim[50:100, 20] = True
testim[85:200, 25] = True
testim[15:35, 50] = True
testim[1:-1, 58] = True
hough_space, angles, dists = tf.hough_line(testim)
hspace, _, _ = tf.hough_line_peaks(hough_space, angles, dists)
assert hspace[0] > hspace[1]
def check_door(image, Pw_corners, Pi_corners, door_edges,
required_matching_ratio=0.7, verbose=0):
"""Check if door is closed."""
results = {}
image_sobel, image_edges = detect_edges(image)
# Detect lines with Hough transform
hough_accumulator, angles, dists = hough_line(image_edges)
hspace, angles, dists = hough_line_peaks(
hough_accumulator, angles, dists, threshold=150.0)
# Estimate camera transformation by minimizing the distance between
# calibration points
params = optimize_transform(camera_params, Pw_corners, Pi_corners)
if verbose >= 1:
print("Parameters: %s" % np.round(params, 3))
cam2world = transform_from(matrix_from_euler_xyz(params[:3]), params[3:6])
kappa = params[-1]
W2I = partial(world2image, cam2world=cam2world, kappa=kappa,
**camera_params)
# Get edge pixels in vicinity of lines
Pi_line_points = check_edge_is_on_line(image_edges, angles, dists)
if len(Pi_line_points) == 0:
if verbose >= 1:
print("No lines detected, assume that door is closed")
door_closed = True
else:
# Check how good the edges of the door projected to the image match
# detected edge pixels that correspond to lines
matchings = [check_line_is_edge(edge, Pi_line_points, cam2world, kappa,
camera_params) for edge in door_edges]
results["door_edges_in_image"] = [m[0] for m in matchings]
ratios = np.array([m[1] for m in matchings])
if verbose >= 1:
print(("Matching ratios: " + ", ".join(["%.2f"] * len(ratios)))
% tuple(100 * ratios))
door_closed = np.any(ratios > required_matching_ratio)
results["cam2world"] = cam2world
results["Pi_line_points"] = Pi_line_points
results["image_sobel"] = image_sobel
results["image_edges"] = image_edges
results["lines"] = (angles, dists)
return door_closed, W2I, results
def test_hough_line_peaks_angle():
img = np.zeros((100, 100), dtype=np.bool_)
img[:, 0] = True
img[0, :] = True
hspace, angles, dists = tf.hough_line(img)
assert len(tf.hough_line_peaks(hspace, angles, dists, min_angle=45)[0]) == 2
assert len(tf.hough_line_peaks(hspace, angles, dists, min_angle=90)[0]) == 1
theta = np.linspace(0, np.pi, 100)
hspace, angles, dists = tf.hough_line(img, theta)
assert len(tf.hough_line_peaks(hspace, angles, dists, min_angle=45)[0]) == 2
assert len(tf.hough_line_peaks(hspace, angles, dists, min_angle=90)[0]) == 1
theta = np.linspace(np.pi / 3, 4. / 3 * np.pi, 100)
hspace, angles, dists = tf.hough_line(img, theta)
assert len(tf.hough_line_peaks(hspace, angles, dists, min_angle=45)[0]) == 2
assert len(tf.hough_line_peaks(hspace, angles, dists, min_angle=90)[0]) == 1
def test_example():
# Construct toydata
image = np.zeros((100, 100))
idx = np.arange(25, 75)
image[idx[::-1], idx] = 255
image[idx, idx] = 255
# Classic straight-line Hough transform
h, theta, d = hough_line(image)
# plot
plt.figure(figsize=(8, 4))
plt.subplot(131)
plt.imshow(image, cmap=plt.cm.gray)
plt.title('Input image')
plt.subplot(132)
plt.imshow(np.log(1 + h),
extent=[np.rad2deg(theta[-1]), np.rad2deg(theta[0]),
d[-1], d[0]],
cmap=plt.cm.gray, aspect=1/1.5)
plt.title('Hough transform')
plt.xlabel('Angles (degrees)')
plt.ylabel('Distance (pixels)')
plt.subplot(133)
plt.imshow(image, cmap=plt.cm.gray)
rows, cols = image.shape
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
plt.plot((0, cols), (y0, y1), '-r')
plt.axis((0, cols, rows, 0))
plt.title('Detected lines')
def satellites(self,maskflag=8,det_nthresh=3.0,ngrowth=1,min_length=200,conv_sigma=2.0,angle_accuracy=1.0,min_distance=40,min_angle=5,threshold=None,num_peaks=np.inf,touch_edge=False,thresh_absolute=True,nmax=100):
"""
maskflag: flag for satellite trails
det_nthresh: relative threshold for detecting satellites
ngrowth: dilation iterations for detections
min_length: minimun lenght for satellites
conv_sigma: convolution kernel to convolve the raw image
angle_accuracy: the accurancy of caculating the theta angle of straight lines
min_distance,min_angle,threshold,num_peaks: keywords in hough_line_peaks (skimage)
touch_edge: need satellites touch the image edge
thresh_abolute: threshold is absolute not relative
nmax: maximum lines to be detected
"""
start=time.time()
maskflag=maskflag | self.bad_flag
if self.sky is None or self.sky_rms is None:
raise AttributeError("get sky and skyrms first!")
# convolve data with gaussian profile
data_conv=self.data.copy()
if self.mask is not None:
data_conv[self.mask]=self.sky
if conv_sigma is not None:
data_conv=snd.filters.gaussian_filter(data_conv,sigma=conv_sigma)
# detect signals
gmask=data_conv - self.sky > self.sky_rms*det_nthresh
# growth satellite trails by dilation operation
if ngrowth >0:
struct=np.ones((3,3)).astype('int')
gmask=snd.morphology.binary_dilation(gmask,struct,iterations=ngrowth)
if self.mask is not None:
mask=gmask | (self.mask)
else:
mask=gmask.copy()
#fits.writeto('gmask.fits',gmask.astype('uint8'),clobber=True)
#fits.writeto('mask.fits',mask.astype('uint8'),clobber=True)
# remove objects that are obviously not satellites
labels,_=snd.label(mask)
slices=snd.find_objects(labels)
flags=np.zeros_like(self.data).astype('int')
nr,nc=self.data.shape
slcs=[]
labs=[]
for i,slc in enumerate(slices):
ilabel=labels[slc]
imask=mask[slc]
igmask=gmask[slc]
tmpmask=ilabel == i+1
slc0,slc1=slc
# remove objects not touch edge using raw data
if touch_edge and (not (slc0.start==0 or slc0.stop==nr or slc1.start==0 or slc1.stop==nc)):
imask[tmpmask]=0
igmask[tmpmask]=0
ilabel[tmpmask]=0
continue
# remove objects that are small
size=ilabel.shape
if np.max(size) < min_length:
imask[tmpmask]=0
igmask[tmpmask]=0
ilabel[tmpmask]=0
continue
# remove objects not touch edge using masked data
if touch_edge:
tmpdata=igmask.copy()
tmpdata[~tmpmask]=0
if not tmpdata.any(): continue
tmpslc0,tmpslc1=snd.find_objects(tmpdata)[0]
r0=slc0.start+tmpslc0.start; c0=slc1.start+tmpslc1.start
r1=r0+tmpslc0.stop-tmpslc0.start;
c1=c0+tmpslc1.stop-tmpslc1.start;
if not (r0==0 or r1==nr or c0==0 or c1==nc):
imask[tmpmask]=0
igmask[tmpmask]=0
ilabel[tmpmask]=0
continue
slcs.append(slc)
labs.append(i+1)
# no labels
if len(slcs) == 0:
print('No Satellite Found!')
return [],[],[]
# find satellites by using hough transform
t=np.linspace(-np.pi/2.0,np.pi/2.0,np.fix(180/angle_accuracy))
h,theta,d=hough_line(mask,theta=t)
print('hough max:',h.max())
thres=None
if threshold is not None:
if thresh_absolute:
thres=threshold
else:
thres=threshold*h.max()
else:
if thresh_absolute:
thres=3*min_length
_,angles,dists = hough_line_peaks(h,theta,d, min_distance=min_distance, min_angle=min_angle, threshold=thres, num_peaks=num_peaks)
print(angles,dists)
# no labels
if len(angles) > nmax:
print('Too much lines, not to detect satellite!')
return [],[],[]
# refine the lines
smask=np.zeros_like(self.data).astype('bool')
newdists=[] #np.zeros_like(dists)
newangles=[] #np.zeros_like(dists)
widths=[] #np.ones_like(angles)
# angle is y+ -> left is plus while y+ -> right is minus
for i,(angle,dist) in enumerate(zip(angles,dists)):
if np.abs(angle) >= np.pi/4.0:
imask=mask.copy()
data=data_conv.copy()
angle1=angle
else:
imask=mask.transpose()
data=data_conv.transpose()
if angle >=0:
angle1=np.pi/2.0-angle
else:
angle1=-np.pi/2.0-angle
dist=-dist
shape=data.shape
lmask=np.zeros(shape).astype('bool')
xx=np.arange(shape[1])
yy=(dist-xx*np.cos(angle1))/np.sin(angle1)
intyy=np.round(yy).astype('int')
for j in np.arange(2*min_distance+1):
jsub=j-min_distance
tmpyy=intyy+jsub
index=np.where((tmpyy >=0) & (tmpyy < shape[0]))[0]
if len(index) == 0: continue
tmpxx=xx[index]; tmpyy=tmpyy[index]
lmask[tmpyy,tmpxx]=imask[tmpyy,tmpxx]
if touch_edge:
tmpmask1=lmask[0,:] | lmask[-1,:]
tmpmask2=lmask[:,0] | lmask[:,-1]
if (not tmpmask1.any()) and (not tmpmask2.any()): continue
#fits.writeto('lmask.fits',lmask.astype('int16'),clobber=True)
width=lmask.sum(axis=0)
tmpmask=width>0
if not tmpmask.any(): continue
mwidth=np.ceil(np.median(width[tmpmask])).astype('int')
if mwidth % 2 ==0: mwidth+=1
if mwidth >= min_distance: continue
#print('satellite widths:',mwidth)
mdata=data.copy()
mdata[~lmask]=-np.inf
indmax=np.argmax(mdata,axis=0)
tmpmask=(width > 0)
if tmpmask.sum() == 0: continue
yy1=indmax[tmpmask];xx1=xx[tmpmask]
model_line=linear_model.RANSACRegressor(linear_model.LinearRegression())
model_line.residual_threshold=min_distance
model_line.fit(xx1.reshape((len(xx1),1)),yy1)
"""
y=kx+b
d=x*cos(t)+y*sin(t)
y=-1/tan(t)*x+d/sin(t)
k=-1/tan(t)
b=d/sin(t)
"""
k=model_line.estimator_.coef_[0]
b=model_line.estimator_.intercept_
if k==0:
angle1=np.pi/2.0
else:
angle1=np.arctan(-1/k)
mdist1=b*np.sin(angle1)
tmpangle=angle1
tmpdist=mdist1
if np.abs(angle) < np.pi/4.0:
if angle1 >=0:
tmpangle=np.pi/2.0-angle1
else:
tmpangle=-np.pi/2.0-angle1
tmpdist=-mdist1
newdists.append(tmpdist)
newangles.append(tmpangle)
widths.append(mwidth)
lmask=np.zeros(shape).astype('bool')
xx=np.arange(shape[1])
yy=(mdist1-xx*np.cos(angle1))/np.sin(angle1)
intyy=np.round(yy).astype('int')
tflag=0
for j in np.arange(mwidth):
jsub=j-mwidth//2
tmpyy=intyy+jsub
index=np.where((tmpyy >0) & (tmpyy < shape[0]))[0]
if len(index) == 0: continue
tmpxx=xx[index]; tmpyy=tmpyy[index]
lmask[tmpyy,tmpxx]=imask[tmpyy,tmpxx]
if touch_edge:
tmpmask1=lmask[0,:] | lmask[-1,:]
tmpmask2=lmask[:,0] | lmask[:,-1]
if (not tmpmask1.any()) and (not tmpmask2.any()): continue
if np.abs(angle) >= np.pi/4.0:
smask=smask | lmask
#mask=mask & (~lmask)
else:
smask=smask | lmask.transpose()
#mask=mask & (~lmask.transpose())
flags=smask*maskflag
if self.flags is not None:
self.flags = np.bitwise_or(self.flags,flags)
else:
self.flags=flags
if self.mask is not None:
self.mask = self.mask | (flags > 0)
else:
self.mask = flags > 0
if self.weight is not None:
tmpmask=flags>0
self.weight[tmpmask]=0
if len(newdists)==0:
print('no satellite found!')
print(newdists,newangles,widths)
print('satellite time: ',time.time()-start)
return newdists,newangles,widths
def interest_region(image, plot_image = 0):
# constant
th_x = 0.5
th_y = 0.5
lid_thickness = 680
bot_thickness = 230
side_thickness = 80
rotation_limit = 2
angle_u_lim = (90.0 + rotation_limit) / 180.0 * np.pi
angle_d_lim = (90.0 - rotation_limit) / 180.0 * np.pi
#Grayscale Conversion and Edge Detection
if(len(image.shape)>2):
image = rgb2gray(image)
img_edge = canny(image, sigma = 3)
#Image Rotation
h, theta, d = hough_line(img_edge)
h_peak, theta_peak, d_peak = hough_line_peaks(h, theta, d)
theta_rotation = theta[np.where(np.absolute(theta_peak)<=angle_u_lim)]
theta_rotation = theta_rotation[np.where(np.absolute(theta_rotation)>=angle_d_lim)]
if(theta_rotation.size):
rotate_angle = np.pi/2-np.absolute(np.mean(theta_rotation))
img_rotate = rotate(img_edge,rotate_angle*180)
#rectangular region selection
index = np.where(img_rotate>0)
[hy,b] = np.histogram(index[0],100)
by = (b[0:(len(b)-1)]+b[1:len(b)])/2
[hx,b] = np.histogram(index[1],100)
bx = (b[0:(len(b)-1)]+b[1:len(b)])/2
temp = by[np.where(hy>=th_y*np.mean(hy))]
if(len(temp)>0):
bottom = np.amin(temp)
top = np.amax(temp)
else:
bottom = 450
top = 1500
temp = bx[np.where(hx>=th_x*np.mean(hx))]
if(len(temp)>0):
left = np.amin(temp)+lid_thickness
right = np.amax(temp)-bot_thickness
if (right <= left):
left = np.amin(temp)+int(lid_thickness/2)
right = np.amax(temp)+int(lid_thickness/2)
else:
left = 1700
right = 3600-bot_thickness
bottom = bottom + side_thickness
top = top - side_thickness
#print [bottom,top,left,right];
image_rotation = rotate(image,rotate_angle*180)
interest_image = image_rotation[bottom:top,left:right];
if(plot_image == 1):
fig, ax = plt.subplots(2,3)
ax[0,0].imshow(image,cmap=plt.cm.gray)
ax[0,0].set_title('Original Image')
ax[0,1].imshow(img_edge,cmap=plt.cm.gray)
ax[0,1].set_title('Canny Endge Detection')
ax[0,2].imshow(image, cmap=plt.cm.gray)
rows, cols = image.shape
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
ax[0,2].plot((0, cols), (y0, y1),'-r')
ax[0,2].axis((0, cols, rows, 0))
ax[0,2].set_title('Detected lines')
ax[0,2].set_axis_off()
ax[1,0].imshow(img_rotate,cmap=plt.cm.gray)
ax[1,0].set_title('Image Edge after Rotation')
ax[1,1].scatter(index[1],index[0])
ax[1,1].set_aspect('equal')
ax[1,1].set_title('Pixel Axises Histogram')
divider = make_axes_locatable(ax[1,1])
axHistx = divider.append_axes("top", 1.2, pad=0.1, sharex=ax[1,1])
axHisty = divider.append_axes("right", 1.2, pad=0.1, sharey=ax[1,1])
# make some labels invisible
plt.setp(axHistx.get_xticklabels() + axHisty.get_yticklabels(),visible=False)
# now determine nice limits by hand:
axHistx.hist(index[1], bins=bx)
axHistx.axhline(y=th_y*np.mean(hy),c="red",linewidth=2,zorder=0)
axHisty.hist(index[0], bins=by, orientation='horizontal')
axHisty.axvline(x=th_x*np.mean(hx),c="red",linewidth=2,zorder=0)
#axHistx.axis["bottom"].major_ticklabels.set_visible(False)
for tl in axHistx.get_xticklabels():
tl.set_visible(False)
axHistx.set_yticks([0, 400, 800, 1200])
#axHisty.axis["left"].major_ticklabels.set_visible(False)
for tl in axHisty.get_yticklabels():
tl.set_visible(False)
axHisty.set_xticks([0, 400, 800, 1200])
ax[1,2].imshow(interest_image,cmap=plt.cm.gray)
ax[1,2].set_title('Edge Detection')
plt.show()
#"""
return [interest_image,[bottom,top,left,right],rotate_angle*180]
h, theta, d = hough_line(image)
fig, ax = plt.subplots(1, 3, figsize=(6, 3.5))
# Original image
ax[0].imshow(image, cmap=plt.cm.gray)
ax[0].axis('off')
# Display Hough transform
ax[1].imshow(np.log(1 + h),
extent=[np.rad2deg(theta[-1]), np.rad2deg(theta[0]),
d[-1], d[0]],
cmap=plt.cm.gray, aspect=1 / 1.5)
ax[1].axis('off')
# Original image with lines detected
ax[2].imshow(image, cmap=plt.cm.gray)
rows, cols = image.shape
for _, angle, dist in zip(*hough_line_peaks(h, theta, d, min_angle=5,
min_distance=5)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
ax[2].plot((0, cols), (y0, y1), '-r')
ax[2].axis((0, cols, rows, 0))
ax[2].axis('off')
# Save a nice figure
fig.tight_layout()
fig.subplots_adjust(wspace=0, left=0, right=1, top=1, bottom=0)
fig.savefig('./hough_lines.png', dpi=300)
def main():
if len(sys.argv) != 2:
print "ERROR! Correct usage is:"
print "\tpython test_hough_transform.py [gps_point_collection.dat]"
return
GRID_SIZE = 500
results = np.zeros((GRID_SIZE, GRID_SIZE), np.float)
# Load GPS points
with open(sys.argv[1], "rb") as fin:
point_collection = cPickle.load(fin)
for pt in point_collection:
y_ind = math.floor((pt[0] - const.RANGE_SW[0]) / (const.RANGE_NE[0] -const.RANGE_SW[0]) * GRID_SIZE)
x_ind = math.floor((pt[1] - const.RANGE_NE[1]) / (const.RANGE_SW[1] -const.RANGE_NE[1]) * GRID_SIZE)
results[x_ind, y_ind] += 1.0
if results[x_ind, y_ind] >= 64:
results[x_ind, y_ind] = 63
results /= np.amax(results)
thresholded_results = np.zeros((GRID_SIZE, GRID_SIZE), np.bool)
THRESHOLD = 0.02
for i in range(0, GRID_SIZE):
for j in range(0, GRID_SIZE):
if results[i,j] >= THRESHOLD:
thresholded_results[i,j] = 1
else:
thresholded_results[i,j] = 0
box_size = 50
x_ind = random.randint(box_size, GRID_SIZE-box_size)
y_ind = random.randint(box_size, GRID_SIZE-box_size)
# x_ind = 217
# y_ind = 175
test_img = thresholded_results[(x_ind-box_size):(x_ind+box_size),\
(y_ind-box_size):(y_ind+box_size)]
h, theta, d = hough_line(test_img)
# fig = plt.figure(figsize=(30,16))
# ax = fig.add_subplot(121, aspect='equal')
# ax.imshow(test_img, cmap=plt.cm.gray)
#
# ax = fig.add_subplot(122)
# img = skeletonize(test_img)
# ax.imshow(img, cmap=plt.cm.gray)
# plt.show()
# fig = plt.figure(figsize=(30,16))
# ax = fig.add_subplot(131, aspect='equal')
# ax.imshow(test_img, cmap=plt.cm.gray)
#
# ax = fig.add_subplot(132)
# ax.imshow(np.log(1+h),
# extent=[np.rad2deg(theta[-1]), np.rad2deg(theta[0]), d[-1], d[0]],
# cmap=plt.cm.gray, aspect=1/1.5)
# ax = fig.add_subplot(133)
# ax.imshow(test_img, cmap=plt.cm.gray)
# rows, cols = test_img.shape
# for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
# y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
# y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
# ax.plot((0, cols), (y0, y1), '-r')
# ax.set_xlim([0, cols])
# ax.set_ylim([rows, 0])
# plt.show()
print "point at: ", x_ind, y_ind
coords = [(y_ind-box_size, x_ind-box_size), (y_ind+box_size, x_ind-box_size),\
(y_ind+box_size, x_ind+box_size), (y_ind-box_size, x_ind+box_size), \
(y_ind-box_size, x_ind-box_size)]
bound_box = Polygon(coords)
patch = PolygonPatch(bound_box, fc='none', ec='red')
fig = plt.figure(figsize=(30,16))
rows, cols = thresholded_results.shape
ax = fig.add_subplot(121, aspect='equal')
ax.imshow(thresholded_results, cmap=plt.cm.gray)
ax.add_patch(patch)
ax.set_xlim([0, cols])
ax.set_ylim([rows, 0])
ax = fig.add_subplot(122)
ax.imshow(test_img, cmap=plt.cm.gray)
for _, angle, dist in zip(*hough_line_peaks(h, theta, d, min_distance=8)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
ax.plot((0, cols), (y0, y1), '-r')
rows, cols = test_img.shape
ax.set_xlim([0, cols])
ax.set_ylim([rows, 0])
plt.show()
def _find_edges(self, image):
"""
Takes as input an RGB chessboard image and computes the Hough lines
and intersections between them corresponding to the edges of the 64
squares. The function returns a tuple whose first element is a
numpy array with shape (N,2) where N is the number of
intersections between the Hough lines. The second element is a list
of tuples (intensity, abs(theta), abs(r)) describing the significant
Hough lines (in polar coordinates).
"""
image_canny = self.compute_canny_image(image)
h, theta, d = hough_line(image_canny)
'''
Small trick: we expect to get something like 11 horizontal lines and
another 11 vertical lines: indeed the chessboard is made of 8 rows and
8 columns, i.e. 9 lines, plus 2 lines for the edges of the chessboard.
So, to help hough_line_peaks() let's set the min. distance expected
between two lines to the size of the image divided by 11.
TODO: better way? Might cause problems if the board does not fit the
whole frame...
'''
min_distance = int(floor(image.shape[1] / 11))
hough_peaks = hough_line_peaks(h, theta, d, min_distance=min_distance,
threshold=40) # TODO adjust threshold
# separate horizontal from vertical lines
vertical_lines = []
horizontal_lines = []
'''
In Hough space, straight lines are parametrized by coefficients (r,θ):
r = x.cos(θ) + y.sin(θ) ; (r,θ) are the polar coordinates of
the point on the line which lies closest to the origin, with r being
either positive or negative, and -π/2 ≤ θ ≤ π/2 (with π/2 ≈ 1.57)
A horizontal line is therefore one with θ=±π/2, and a vertical line is
one with θ=0.
Also, two lines are parallel iif they have the same θ (mod π/2).
Finally, in scikit-image, keep in mind that the orientation is :
O----->
| (x axis)
|
V (y axis)
(where the origin is the top-left corner of our image)
And as a consequence, positive angles are counted from the X axis toward
the Y axis (i.e. "clockwise").
'''
for intensity, theta, r in zip(*hough_peaks):
if np.fabs(theta) < self.LINE_ANGLE_TOLERANCE:
vertical_lines.append((intensity, theta, r))
elif np.fabs(np.fabs(theta) - math.pi / 2) < self.LINE_ANGLE_TOLERANCE:
horizontal_lines.append((intensity, theta, r))
'''
Only keep the 9 most significant lines of each direction and sort
them by absolute radius, to return edges from top to bottom,
and left to right. We sort by *absolute* radius because horizontal
lines can be parametrized either as 'θ=π/2, r>0' or 'θ=-π/2, r<0'
'''
for i,lines in enumerate((horizontal_lines, vertical_lines)):
lines.sort(reverse=True) # reverse-sort by intensity
del lines[9:] # only keep the 9 first
lines.sort(key=lambda entry: np.fabs(entry[2])) # sort by absolute radius
'''
Now let's find the intersections of these 18 lines.
We are solving the following system:
r1 = x.cos(θ1) + y.sin(θ1) and r2 = x.cos(θ2) + y.sin(θ2) for (x,y)
The non-degenerate solution is:
x = (r1.sin(θ2) - r2.sin(θ1))/D and y = (r2.cos(θ1) - r1.cos(θ2))/D
with D = cos(θ1).sin(θ2) - sin(θ1).cos(θ2) = sin(θ2-θ1)
We can see that this holds only if θ1 ≠ θ2, i.e. only if the lines are
not parallel (which makes sense!).
'''
intersections = []
for _, theta1, r1 in horizontal_lines:
for _, theta2, r2 in vertical_lines:
# don't attempt to compute the intersection if the lines are parallel
denum = np.sin(theta2 - theta1)
if denum < self.ANGLE_EPSILON:
continue
x_inter = (r1 * np.sin(theta2) - r2 * np.sin(theta1)) / denum
y_inter = (r2 * np.cos(theta1) - r1 * np.cos(theta2)) / denum
# register this intersection iff. it is *inside* the image
if 0 < x_inter < image.shape[1] and 0 < y_inter < image.shape[0]:
intersections.append((x_inter, y_inter))
hough_lines = horizontal_lines + vertical_lines
# TODO we don't check that we get 9x9 intersections...
'''
Since the horizontal and vertical lines have been sorted by increasing
value of |r|, the intersections are now sorted from top-left to bottom-right
corner of the image.
'''
return np.array(intersections), hough_lines
def extract_corner_hough(patch):
""" Extract four corner points using Hough transform
"""
# Find the lines that makes the boundary box using Hough Transform
h, theta, d = hough_line(patch)
# Divide the hough space for searching different horizontal and vertical lines
hcroph = h[:, 0:30]
rows, cols = patch.shape
# Search all vertical lines in the hough parameter space
# It is located in the middle, when theta is near zero
vlf = []
for _, angle, dist in zip(*hough_line_peaks(h, theta, d, min_angle=9)):
x0 = (dist - 0 * np.sin(angle)) / np.cos(angle)
x1 = (dist - rows * np.sin(angle)) / np.cos(angle)
vlf += [((x0, 0), (x1, rows))]
# Likewise previous operation, but with horizontal lines
# Located at the left, when theta is near minus 90 degree
hlf = []
for _, angle, dist in zip(*hough_line_peaks(hcroph, theta, d, min_angle=9)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
hlf += [((0, y0), (cols, y1))]
if DEBUG:
plt.subplot(142)
plt.title('hough line')
plt.imshow(patch, cmap='gray')
for i in xrange(len(vlf)):
plt.plot((vlf[i][0][0], vlf[i][1][0]), (vlf[i][0][1], vlf[i][1][1]), 'r-')
for i in xrange(len(hlf)):
plt.plot((hlf[i][0][0], hlf[i][1][0]), (hlf[i][0][1], hlf[i][1][1]), 'r-')
plt.axis([0, cols, rows, 0])
# Search the rightmost and leftmost vertical lines
vl = []
vl += [search_closest_line(vlf, [(0, 0), (0, rows)])]
vl += [search_closest_line(vlf, [(cols, 0), (cols, rows)])]
# Search the topmost and bottommost horizontal lines
hl = []
hl += [search_closest_line(hlf, [(0, 0), (cols, 0)])]
hl += [search_closest_line(hlf, [(0, rows), (cols, rows)])]
if len(hl) < 2 or len(vl) < 2:
print 'Not enough lines found'
return []
points = [vl[0], vl[-1], hl[0], hl[-1]]
# Check for error
for i in xrange(4):
for j in xrange(i + 1, 4):
if points[i] == points[j]:
print 'Error Line'
return []
# Calculate the intersection between pair of vertical and horizontal lines
# The line is defined by using two points.
p1 = calc_intersection(vl[0][0][0], vl[0][0][1], vl[0][1][0], vl[0][1][1], hl[0][0][0], hl[0][0][1], hl[0][1][0],
hl[0][1][1])
p2 = calc_intersection(vl[0][0][0], vl[0][0][1], vl[0][1][0], vl[0][1][1], hl[1][0][0], hl[1][0][1], hl[1][1][0],
hl[1][1][1])
p3 = calc_intersection(vl[1][0][0], vl[1][0][1], vl[1][1][0], vl[1][1][1], hl[0][0][0], hl[0][0][1], hl[0][1][0],
hl[0][1][1])
p4 = calc_intersection(vl[1][0][0], vl[1][0][1], vl[1][1][0], vl[1][1][1], hl[1][0][0], hl[1][0][1], hl[1][1][0],
hl[1][1][1])
# Find the nearest point for each corner
dim = patch.shape
corners = [(0, 0), (0, dim[0]), (dim[1], dim[0]), (dim[1], 0)]
points = [p1, p2, p3, p4]
dest_points = [[] for x in range(4)]
for i in xrange(4):
dest_points[i] = search_closest_points(corners[i], points)
epsilon = 1e-10
for i in xrange(4):
for j in xrange(i + 1, 4):
if calc_distance(dest_points[i], dest_points[j]) < epsilon:
print 'Error point'
return []
return dest_points
def lines(base, test):
image1 = imread(base, flatten=True)
image2 = imread(test, flatten=True)
# Angles
a1 = []
a2 = []
group1_a1 = []
group2_a1 = []
group3_a1 = []
# # Plot 1 -> Base Case (original)
# print "Plot 1"
h, theta, d = hough_line(image1)
rows, cols = image1.shape
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
degree = degrees(angle)
a1.append(degree)
if degree < -45 or degree > 45:
if degree < -85 or degree > 85:
pos_horizontal = []
neg_horizontal = []
if degree > 0:
pos_horizontal.append(90 - degree)
elif degree <= 0:
neg_horizontal.append(-(degree + 90))
elif degree > -45 and degree < 0:
if degree > -40 and degree < -25:
group2_a1.append(degree)
elif degree > 0 and degree < 45:
if degree > 15 and degree < 40:
group3_a1.append(degree)
if len(pos_horizontal) == 0:
avg_horizontal = reduce(lambda x, y: x + y, neg_horizontal) / float(len(neg_horizontal))
elif len(neg_horizontal) == 0:
avg_horizontal = reduce(lambda x, y: x + y, pos_horizontal) / float(len(pos_horizontal))
else:
avg_neg = reduce(lambda x, y: x + y, neg_horizontal) / float(len(neg_horizontal))
avg_pos = reduce(lambda x, y: x + y, pos_horizontal) / float(len(pos_horizontal))
avg_horizontal = (avg_pos + avg_neg)/2
print "BASE HORIZONTAL: " + str(avg_horizontal)
h1, theta1, d1 = hough_line(image2)
rows, cols = image2.shape
group1_a2 = []
group2_a2 = []
group3_a2 = []
for _, angle, dist in zip(*hough_line_peaks(h1, theta1, d1)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
degree = degrees(angle)
a2.append(degree)
if degree < -45 or degree > 45:
if degree < -85 or degree > 85:
pos_horizontal_user = []
neg_horizontal_user = []
if degree > 0:
pos_horizontal_user.append(90 - degree)
elif degree <= 0:
neg_horizontal_user.append(-(degree + 90))
elif degree > -45 and degree < 0:
if degree > -40 and degree < -25:
group2_a2.append(degree)
elif degree > 0 and degree < 45:
if degree > 15 and degree < 40:
group3_a2.append(degree)
a2.append(degrees(angle))
if len(pos_horizontal_user) == 0:
avg_horizontal_user = reduce(lambda x, y: x + y, neg_horizontal_user) / float(len(neg_horizontal_user))
elif len(neg_horizontal_user) == 0:
avg_horizontal_user = reduce(lambda x, y: x + y, pos_horizontal_user) / float(len(pos_horizontal_user))
else:
avg_neg = reduce(lambda x, y: x + y, neg_horizontal_user) / float(len(neg_horizontal_user))
avg_pos = reduce(lambda x, y: x + y, pos_horizontal_user) / float(len(pos_horizontal_user))
avg_horizontal_user = (avg_pos + avg_neg)/2
avg_down = -1
avg_up =-1
base_down = reduce(lambda x, y: x + y, group2_a1) / float(len(group2_a1))
if len(group2_a2) > 0:
avg_down = reduce(lambda x, y: x + y, group2_a2) / float(len(group2_a2))
base_up = reduce(lambda x, y: x + y, group3_a1) / float(len(group3_a1))
if len(group3_a2) > 0:
avg_up = reduce(lambda x, y: x + y, group3_a2) / float(len(group3_a2))
print "USER HORIZONTAL: " + str(avg_horizontal_user)
print "BASE DOWN: " + str(base_down)
print "USER DOWN: " + str(avg_down)
print "BASE UP: " + str(base_up)
print "USER UP: " + str(avg_up)
base_down = float(base_down)
base_up = float(base_up)
user_down = float(avg_down)
user_up = float(avg_up)
# Calculate score
score = 10
lost_down = abs(base_down - avg_down)
lost_up = abs(base_up - avg_up)
lost_horizontal = abs(avg_horizontal - avg_horizontal_user)
SCORE_FACTOR = 3 # equals max angle of error / 10
score1 = lost_down / SCORE_FACTOR
score2 = lost_up / SCORE_FACTOR
score3 = lost_horizontal / SCORE_FACTOR
neg_points = (score1 + score2 + score3) / 3.0
score -= neg_points
print "SCORE: " + str(score)
return score
def skimage_transform():
#Source: http://scikit-image.org/docs/dev/auto_examples/plot_line_hough_transform.html
image = data.imread('../images/window.jpg', True)
# Classic straight-line Hough transform
h, theta, d = hough_line(image)
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize=(8, 4))
ax1.imshow(image, cmap=plt.cm.gray)
ax1.set_title('Input image')
ax1.set_axis_off()
ax2.imshow(np.log(1 + h),
extent=[np.rad2deg(theta[-1]), np.rad2deg(theta[0]),
d[-1], d[0]],
cmap=plt.cm.gray,
aspect=1/1.5)
ax2.set_title('Hough transform')
ax2.set_xlabel('Angles (degrees)')
ax2.set_ylabel('Distance (pixels)')
ax2.axis('image')
ax3.imshow(image, cmap=plt.cm.gray)
rows, cols = image.shape
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
ax3.plot((0, cols), (y0, y1), '-r')
ax3.axis((0, cols, rows, 0))
ax3.set_title('Detected lines')
ax3.set_axis_off()
# Line finding, using the Probabilistic Hough Transform
image = data.camera()
edges = canny(image,2, 1, 25)
lines = probabilistic_hough_line(edges, threshold=10, line_length=100, line_gap=3)
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize=(8,4), sharex=True, sharey=True)
ax1.imshow(image, cmap=plt.cm.gray)
ax1.set_title('Input image')
ax1.set_axis_off()
ax1.set_adjustable('box-forced')
ax2.imshow(edges, cmap=plt.cm.gray)
ax2.set_title('Canny edges')
ax2.set_axis_off()
ax2.set_adjustable('box-forced')
ax3.imshow(edges * 0)
for line in lines:
p0, p1 = line
ax3.plot((p0[0], p1[0]), (p0[1], p1[1]))
ax3.set_title('Probabilistic Hough')
ax3.set_axis_off()
ax3.set_adjustable('box-forced')
image = data.imread('../images/AMFGP02lg.jpg', True)
edges = canny(image, 2, 1, 25)
lines = probabilistic_hough_line(edges, threshold=10, line_length=100, line_gap=3)
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize=(8,4), sharex=True, sharey=True)
ax1.imshow(image, cmap=plt.cm.gray)
ax1.set_title('Input image')
ax1.set_axis_off()
ax1.set_adjustable('box-forced')
ax2.imshow(edges, cmap=plt.cm.gray)
ax2.set_title('Canny edges')
ax2.set_axis_off()
ax2.set_adjustable('box-forced')
ax3.imshow(edges * 0)
for line in lines:
p0, p1 = line
ax3.plot((p0[0], p1[0]), (p0[1], p1[1]))
ax3.set_title('Probabilistic Hough')
ax3.set_axis_off()
ax3.set_adjustable('box-forced')
image = data.imread('../images/window.jpg', True)
edges = canny(image, 2, 1, 25)
lines = probabilistic_hough_line(edges, threshold=10, line_length=100, line_gap=1)
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, figsize=(8,4), sharex=True, sharey=True)
ax1.imshow(image, cmap=plt.cm.gray)
ax1.set_title('Input image')
ax1.set_axis_off()
ax1.set_adjustable('box-forced')
ax2.imshow(edges, cmap=plt.cm.gray)
ax2.set_title('Canny edges')
ax2.set_axis_off()
ax2.set_adjustable('box-forced')
ax3.imshow(edges * 0)
for line in lines:
p0, p1 = line
ax3.plot((p0[0], p1[0]), (p0[1], p1[1]))
ax3.set_title('Probabilistic Hough')
ax3.set_axis_off()
ax3.set_adjustable('box-forced')
plt.show()
between_fields = logical_and(field_area, logical_not(field_mask))
# Find the roads and separate the fields
# Separate out into smaller blocks
print('Separating Fields in image')
for stride in [100, 200, 400]: # pixels
num_row_strides = int_(between_fields.shape[0]/stride)
num_col_strides = int_(between_fields.shape[1]/stride)
r_stride = int_(between_fields.shape[0]/num_row_strides)
c_stride = int_(between_fields.shape[1]/num_col_strides)
for r in range(num_row_strides+1):
for c in range(num_col_strides+1):
h, theta, d = hough_line(between_fields[r*r_stride:(r+1)*r_stride, c*c_stride:(c+1)*c_stride])
threshold = 0 #0.0005*max(h)
h, theta, d = hough_line_peaks(h, theta, d, min_distance=20, threshold=threshold)
for n in range(len(theta)):
if abs(theta[n]) < 0.1 or abs(theta[n]) > ((pi/2) - 0.1):
draw_hough_line(field_mask[r*r_stride:(r+1)*r_stride, c*c_stride:(c+1)*c_stride], d[n], theta[n])
# do a few small openings
field_mask = binary_opening(field_mask, rectangle(1, 5))
field_mask = binary_opening(field_mask, rectangle(5, 1))
imsave(fname_template.format('segmented_fields', 'png'), field_mask)
# Label fields
field_mask = label(field_mask, 4, 0) + 1
remove_small_objects(field_mask, 100, 1, True)
field_props = regionprops(field_mask)
def lines(base, test):
"""
Reads two images of yoga poses and compares them.
"""
image1 = imread(base, flatten=True)
image2 = imread(test, flatten=True)
# Angles
a1 = []
a2 = []
group1_a1 = []
group2_a1 = []
group3_a1 = []
# Generate figure and 2 axes from matplotlib
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(8, 4))
# # Plot 1 -> Base Case (original)
# print "Plot 1"
h, theta, d = hough_line(image1)
ax1.imshow(image1, cmap=plt.cm.get_cmap('gray'))
rows, cols = image1.shape
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
degree = degrees(angle)
a1.append(degree)
if degree < -45 or degree > 45:
if degree < -85 or degree > 85:
pos_horizontal = []
neg_horizontal = []
if degree > 0:
pos_horizontal.append(90 - degree)
elif degree <= 0:
neg_horizontal.append(-(degree + 90))
ax1.plot((0, cols), (y0, y1), '-r')
elif degree > -45 and degree < 0:
if degree > -40 and degree < -25:
group2_a1.append(degree)
ax1.plot((0, cols), (y0, y1), '-r')
elif degree > 0 and degree < 45:
if degree > 15 and degree < 40:
group3_a1.append(degree)
ax1.plot((0, cols), (y0, y1), '-r')
ax1.axis((0, cols, rows, 0))
ax1.set_title('Detected lines')
ax1.set_axis_off()
if len(pos_horizontal) == 0:
avg_horizontal = reduce(lambda x, y: x + y, neg_horizontal) / float(len(neg_horizontal))
elif len(neg_horizontal) == 0:
avg_horizontal = reduce(lambda x, y: x + y, pos_horizontal) / float(len(pos_horizontal))
else:
avg_neg = reduce(lambda x, y: x + y, neg_horizontal) / float(len(neg_horizontal))
avg_pos = reduce(lambda x, y: x + y, pos_horizontal) / float(len(pos_horizontal))
avg_horizontal = (avg_pos + avg_neg)/2
print "BASE HORIZONTAL: " + str(avg_horizontal)
# print group2_a1
# print group3_a1
# Plot 2 -> Test Case (submission)
# print "\nPlot 2"
h1, theta1, d1 = hough_line(image2)
ax2.imshow(image2, cmap=plt.cm.gray)
rows, cols = image2.shape
group1_a2 = []
group2_a2 = []
group3_a2 = []
for _, angle, dist in zip(*hough_line_peaks(h1, theta1, d1)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - cols * np.cos(angle)) / np.sin(angle)
degree = degrees(angle)
a2.append(degree)
if degree < -45 or degree > 45:
if degree < -85 or degree > 85:
pos_horizontal_user = []
neg_horizontal_user = []
if degree > 0:
pos_horizontal_user.append(90 - degree)
elif degree <= 0:
neg_horizontal_user.append(-(degree + 90))
ax2.plot((0, cols), (y0, y1), '-r')
elif degree > -45 and degree < 0:
if degree > -40 and degree < -25:
group2_a2.append(degree)
ax2.plot((0, cols), (y0, y1), '-r')
elif degree > 0 and degree < 45:
if degree > 15 and degree < 40:
group3_a2.append(degree)
ax2.plot((0, cols), (y0, y1), '-r')
a2.append(degrees(angle))
# ax2.plot((0, cols), (y0, y1), '-r')
if len(pos_horizontal_user) == 0:
avg_horizontal_user = reduce(lambda x, y: x + y, neg_horizontal_user) / float(len(neg_horizontal_user))
elif len(neg_horizontal_user) == 0:
avg_horizontal_user = reduce(lambda x, y: x + y, pos_horizontal_user) / float(len(pos_horizontal_user))
else:
avg_neg = reduce(lambda x, y: x + y, neg_horizontal_user) / float(len(neg_horizontal_user))
avg_pos = reduce(lambda x, y: x + y, pos_horizontal_user) / float(len(pos_horizontal_user))
avg_horizontal_user = (avg_pos + avg_neg)/2
print "USER HORIZONTAL: " + str(avg_horizontal_user)
# print sorted(a2, key=float)
# print group2_a2
# print group3_a2
avg_down = -1
avg_up =-1
base_down = reduce(lambda x, y: x + y, group2_a1) / float(len(group2_a1))
if len(group2_a2) > 0:
avg_down = reduce(lambda x, y: x + y, group2_a2) / float(len(group2_a2))
base_up = reduce(lambda x, y: x + y, group3_a1) / float(len(group3_a1))
if len(group3_a2) > 0:
avg_up = reduce(lambda x, y: x + y, group3_a2) / float(len(group3_a2))
print "BASE DOWN: " + str(base_down)
print "USER DOWN: " + str(avg_down)
print "BASE UP: " + str(base_up)
print "USER UP: " + str(avg_up)
ax2.axis((0, cols, rows, 0))
ax2.set_title('Detected lines')
ax2.set_axis_off()
plt.show()
return a1, a2
ax = axes.ravel()
ax[0].imshow(image, cmap=cm.gray)
ax[0].set_title('Input image')
ax[0].set_axis_off()
ax[1].imshow(np.log(1 + h),
extent=[np.rad2deg(theta[-1]), np.rad2deg(theta[0]), d[-1], d[0]],
cmap=cm.gray, aspect=1/1.5)
ax[1].set_title('Hough transform')
ax[1].set_xlabel('Angles (degrees)')
ax[1].set_ylabel('Distance (pixels)')
ax[1].axis('image')
ax[2].imshow(image, cmap=cm.gray)
for _, angle, dist in zip(*hough_line_peaks(h, theta, d)):
y0 = (dist - 0 * np.cos(angle)) / np.sin(angle)
y1 = (dist - image.shape[1] * np.cos(angle)) / np.sin(angle)
ax[2].plot((0, image.shape[1]), (y0, y1), '-r')
ax[2].set_xlim((0, image.shape[1]))
ax[2].set_ylim((image.shape[0], 0))
ax[2].set_axis_off()
ax[2].set_title('Detected lines')
plt.tight_layout()
plt.show()
# Line finding using the Probabilistic Hough Transform
image = data.camera()
edges = canny(image, 2, 1, 25)
lines = probabilistic_hough_line(edges, threshold=10, line_length=5,
import sys
from scipy.ndimage import io, uniform_filter
from skimage import transform
from skimage import color
from src.image_preprocess.PeakDetector import generate_green_map
__author__ = "Kern"
if len(sys.argv) < 2:
raise ValueError("Usage:", sys.argv[0], " Missing some argument to indicate input files")
path = sys.argv[1]
image_input = io.imread(path)
image = uniform_filter(image_input)
image = generate_green_map(image)
image_grey = color.rgb2gray(image)
hspace, angles, dists = transform.hough_line(image_grey)
hspace, angles, dists = transform.hough_line_peaks(hspace, angles, dists)
print(hspace, angles, dists)
def determine_skew(self, img_file):
img = io.imread(img_file, as_grey=True)
edges = canny(img, sigma=self.sigma)
h, a, d = hough_line(edges)
_, ap, _ = hough_line_peaks(h, a, d, num_peaks=self.num_peaks)
if len(ap) == 0:
return {"Image File": img_file, "Message": "Bad Quality"}
absolute_deviations = [self.calculate_deviation(k) for k in ap]
average_deviation = np.mean(np.rad2deg(absolute_deviations))
ap_deg = [np.rad2deg(x) for x in ap]
bin_0_45 = []
bin_45_90 = []
bin_0_45n = []
bin_45_90n = []
for ang in ap_deg:
deviation_sum = int(90 - ang + average_deviation)
if self.compare_sum(deviation_sum):
bin_45_90.append(ang)
continue
deviation_sum = int(ang + average_deviation)
if self.compare_sum(deviation_sum):
bin_0_45.append(ang)
continue
deviation_sum = int(-ang + average_deviation)
if self.compare_sum(deviation_sum):
bin_0_45n.append(ang)
continue
deviation_sum = int(90 + ang + average_deviation)
if self.compare_sum(deviation_sum):
bin_45_90n.append(ang)
angles = [bin_0_45, bin_45_90, bin_0_45n, bin_45_90n]
lmax = 0
for j in range(len(angles)):
l = len(angles[j])
if l > lmax:
lmax = l
maxi = j
if lmax:
ans_arr = self.get_max_freq_elem(angles[maxi])
ans_res = np.mean(ans_arr)
else:
ans_arr = self.get_max_freq_elem(ap_deg)
ans_res = np.mean(ans_arr)
data = {
"Image File": img_file,
"Average Deviation from pi/4": average_deviation,
"Estimated Angle": ans_res,
"Angle bins": angles}
if self.display_output:
self.display(data)
if self.plot_hough:
self.display_hough(h, a, d)
return data