def test_homography_torture(self):
"""Test homography estimation accuracy across many inputs.
This "torture test" for homography estimation exercises the
homography() function across several input images and a broad range of
perspective distortions, computing a final "score" that is the median
difference between the estimated and true homography.
"""
# Initialize the random module to a known state.
random.seed(42)
differences = []
for filename in ("houses_left.png", "camera_hand.png", "boots.png",
"clouds.png", "foam.png"):
image = cv2.imread(os.path.join("test_data/torture", filename),
cv2.CV_LOAD_IMAGE_GRAYSCALE)
rows, cols = image.shape
for i in range(10):
H_expected = self._random_homography(rows, cols)
image_warped = cv2.warpPerspective(image, H_expected,
(int(cols * 1.5),
int(rows * 1.5)))
# Uncomment to visualize.
# cv2.imshow("warp", image_warped)
# cv2.waitKey(0)
H_actual = pano_stitcher.homography(image_warped, image)
H_difference = numpy.absolute(H_expected - H_actual)
H_difference_magnitude = numpy.linalg.norm(H_difference)
differences.append(H_difference_magnitude)
print "Difference for", filename, "homography", i, ":",
print H_difference_magnitude
print "Median difference: ", sorted(differences)[len(differences) / 2]
def test_homography(self):
"""Checks that a known homography is recovered accurately."""
# Load the left_houses image.
houses_left = cv2.imread("test_data/houses_left.png",
cv2.CV_LOAD_IMAGE_GRAYSCALE)
rows, cols = houses_left.shape
# Warp with a known homography.
H_expected = self._known_homography(rows, cols)
houses_left_warped = cv2.warpPerspective(houses_left, H_expected,
(cols, rows))
# Compute the homography with the library.
H_actual = pano_stitcher.homography(houses_left_warped, houses_left)
# The two should be nearly equal.
H_difference = numpy.absolute(H_expected - H_actual)
H_difference_magnitude = numpy.linalg.norm(H_difference)
print "Expected homography:"
print H_expected
print "Actual homography:"
print H_actual
print "Difference:"
print H_difference
print "Magnitude of difference:", H_difference_magnitude
max_difference_magnitude = 0.5
self.assertLessEqual(H_difference_magnitude, max_difference_magnitude)
def main():
cards_1 = cv2.imread("cards_1.jpg")
cards_2 = cv2.imread("cards_2.jpg")
cards_3 = cv2.imread("cards_3.jpg")
H_1 = pano_stitcher.homography(cards_2, cards_1)
H_2 = pano_stitcher.homography(cards_2, cards_2)
H_3 = pano_stitcher.homography(cards_2, cards_3)
cards_1_warped, cards_1_origin = pano_stitcher.warp_image(cards_1, H_1)
cards_2_warped, cards_2_origin = pano_stitcher.warp_image(cards_2, H_2)
cards_3_warped, cards_3_origin = pano_stitcher.warp_image(cards_3, H_3)
images = (cards_1_warped, cards_3_warped, cards_2_warped)
origins = (cards_1_origin, cards_3_origin, cards_2_origin)
pano = pano_stitcher.create_mosaic(images, origins)
cv2.imwrite("pano.png", pano)
def create_pano1():
gray_left = cv2.imread("my_panos/images/left.png",
cv2.CV_LOAD_IMAGE_GRAYSCALE)
left = cv2.imread("my_panos/images/left.png") # , -1)
gray_middle = cv2.imread("my_panos/images/middle.png",
cv2.CV_LOAD_IMAGE_GRAYSCALE)
middle = cv2.imread("my_panos/images/middle.png")
# ***middle = cv2.cvtColor(middle, cv2.COLOR_BGR2BGRA)
gray_right = cv2.imread("my_panos/images/right.png",
cv2.CV_LOAD_IMAGE_GRAYSCALE)
right = cv2.imread("my_panos/images/right.png") # , -1)
# rows, cols = left_image.shape
# cv2.namedWindow('image', cv2.WINDOW_NORMAL)
# cv2.imshow("image", left_image)
# cv2.waitKey(0)
# compute homography for left image
homography1 = pano_stitcher.homography(gray_middle, gray_left)
# warp left image
warped_left, origin1 = pano_stitcher.warp_image(left, homography1)
# compute homography for right image
homography2 = pano_stitcher.homography(gray_middle, gray_right)
# warp right image
warped_right, origin2 = pano_stitcher.warp_image(right, homography2)
# print("warped_left: ", warped_left.shape)
# print("middle: ", middle.shape)
# print("warped_right: ", warped_right.shape)
images = (warped_left, warped_right, middle)
origins = (origin1, origin2, (0, 0))
mosaic1 = pano_stitcher.create_mosaic(images, origins)
cv2.imwrite("feetMosaic.png", mosaic1)
Three hardcoded images are used to create a simple
panorama.
"""
import cv2
import numpy as np
import pano_stitcher as ps
# Open all of the images
b1 = cv2.imread('my_panos/src/part1.jpg')
b2 = cv2.imread('my_panos/src/part2.jpg')
b3 = cv2.imread('my_panos/src/part3.jpg')
# Find a homography to warp image 1
# onto image 2, warp it
b1_homog = ps.homography(b2, b1)
b1_warped, b1_origins = ps.warp_image(b1, b1_homog)
print 'origins for warped b1:', b1_origins
# Set b2 to be at the origin
b2_origins = (0, 0)
# Find a homography to warp image 3
# onto image 2, warp it
b3_homog = ps.homography(b2, b3)
b3_warped, b3_origins = ps.warp_image(b3, b3_homog)
print 'origins for warped b3:', b3_origins
# Convert b2 to a 4-channel image
b2 = cv2.cvtColor(b2, cv2.COLOR_BGR2BGRA)
import cv2
import numpy as np
import pano_stitcher as ps
# Load source images
p1 = cv2.imread('my_panos/src/part1.jpg')
p2 = cv2.imread('my_panos/src/part2.jpg')
p3 = cv2.imread('my_panos/src/part3.jpg')
# Warp first image by the homography mapping
# the first image to the second image
p1_homography = ps.homography(p2, p1)
p1_warped, p1_origin = ps.warp_image(p1, p1_homography)
# Warp third image by the homography mapping
# the third image to the second image
p3_homography = ps.homography(p2, p3)
p3_warped, p3_origin = ps.warp_image(p3, p3_homography)
# Add alpha channel to second image
blue, green, red = cv2.split(p2)
alpha = np.zeros(green.shape, dtype=np.uint8)
alpha.fill(255)
p2 = cv2.merge([blue, green, red, alpha])
# Composite warped images and image in target plane
pano = ps.create_mosaic(
[p1_warped, p2, p3_warped], [p1_origin, (0, 0), p3_origin])
cv2.imwrite('my_panos/pano.jpg', pano)
import cv2
import numpy as np
import pano_stitcher as ps
# Load source images
p1 = cv2.imread('my_panos/src/part1.jpg')
p2 = cv2.imread('my_panos/src/part2.jpg')
p3 = cv2.imread('my_panos/src/part3.jpg')
# Warp first image by the homography mapping
# the first image to the second image
p1_homography = ps.homography(p2, p1)
p1_warped, p1_origin = ps.warp_image(p1, p1_homography)
# Warp third image by the homography mapping
# the third image to the second image
p3_homography = ps.homography(p2, p3)
p3_warped, p3_origin = ps.warp_image(p3, p3_homography)
# Add alpha channel to second image
blue, green, red = cv2.split(p2)
alpha = np.zeros(green.shape, dtype=np.uint8)
alpha.fill(255)
p2 = cv2.merge([blue, green, red, alpha])
# Composite warped images and image in target plane
pano = ps.create_mosaic([p1_warped, p2, p3_warped],
[p1_origin, (0, 0), p3_origin])
cv2.imwrite('my_panos/pano.jpg', pano)
import numpy
left = cv2.imread("3.jpg", -1)
right = cv2.imread("5.jpg", -1)
right2 = cv2.imread("6.jpg", -1)
middle = cv2.imread("4.jpg", -1)
left1 = cv2.imread("2.jpg", -1)
left2 = cv2.imread("1.jpg", -1)
leftImages = [left2, left1, left]
rightImages = [right2, right]
while len(leftImages) > 1:
print "Left Image Loop"
image = leftImages.pop(0)
h**o = pano_stitcher.homography(leftImages[0], image)
warped, origin = pano_stitcher.warp_image(image, h**o)
leftImages[0] = pano_stitcher.create_mosaic(
[warped, cv2.cvtColor(leftImages[0], cv2.COLOR_BGR2BGRA)],
[origin, (0, 0)])
leftImages[0] = cv2.cvtColor(leftImages[0], cv2.COLOR_BGRA2BGR)
left_pano = leftImages[0]
cv2.imshow("sup", left_pano)
cv2.waitKey()
while len(rightImages) > 1:
print "Right Image Loop"
image = rightImages.pop(0)
h**o = pano_stitcher.homography(rightImages[0], image)
warped, origin = pano_stitcher.warp_image(image, h**o)
Three hardcoded images are used to create a simple
panorama.
"""
import cv2
import numpy as np
import pano_stitcher as ps
# Open all of the images
b1 = cv2.imread('my_panos/src/part1.jpg')
b2 = cv2.imread('my_panos/src/part2.jpg')
b3 = cv2.imread('my_panos/src/part3.jpg')
# Find a homography to warp image 1
# onto image 2, warp it
b1_homog = ps.homography(b2, b1)
b1_warped, b1_origins = ps.warp_image(b1, b1_homog)
print 'origins for warped b1:', b1_origins
# Set b2 to be at the origin
b2_origins = (0, 0)
# Find a homography to warp image 3
# onto image 2, warp it
b3_homog = ps.homography(b2, b3)
b3_warped, b3_origins = ps.warp_image(b3, b3_homog)
print 'origins for warped b3:', b3_origins
# Convert b2 to a 4-channel image
b2 = cv2.cvtColor(b2, cv2.COLOR_BGR2BGRA)
# Arbitrary index
a = 0
for i in left:
# Show the name of the image
print sys.argv[i]
# Open it and append it to the left images list
left_images.append(cv2.imread(sys.argv[i]))
# If closest to the middle image, warp into that perspective
# Otherwise, warp into the perspective of the last image
# that was warped
if a is 0:
h = ps.homography(middle_image, left_images[a])
else:
h = ps.homography(left_warped_images[a - 1], left_images[a])
warped, origins = ps.warp_image(left_images[a], h)
left_warped_images.append(warped)
# If closest to the middle image, use normal origins
# Otherwise, use the origins + the origins of the
# last image that was warped
if a is 0:
left_images_origins.append(origins)
else:
prev_origins = left_images_origins[a - 1]
new_origins = (origins[0] + prev_origins[0],
origins[1] + prev_origins[1])
left_images_origins.append(new_origins)
import numpy
left = cv2.imread("3.jpg",-1)
right = cv2.imread("5.jpg",-1)
right2 = cv2.imread("6.jpg",-1)
middle = cv2.imread("4.jpg",-1)
left1 = cv2.imread("2.jpg",-1)
left2 = cv2.imread("1.jpg",-1)
leftImages = [left2, left1, left]
rightImages = [right2, right]
while len(leftImages) > 1:
print "Left Image Loop"
image = leftImages.pop(0)
h**o = pano_stitcher.homography(leftImages[0], image)
warped, origin = pano_stitcher.warp_image(image, h**o)
leftImages[0] = pano_stitcher.create_mosaic([warped, cv2.cvtColor(leftImages[0],cv2.COLOR_BGR2BGRA)], [origin, (0,0)])
leftImages[0] = cv2.cvtColor(leftImages[0],cv2.COLOR_BGRA2BGR)
left_pano = leftImages[0]
cv2.imshow("sup", left_pano)
cv2.waitKey()
while len(rightImages) > 1:
print "Right Image Loop"
image = rightImages.pop(0)
h**o = pano_stitcher.homography(rightImages[0], image)
warped, origin = pano_stitcher.warp_image(image, h**o)
rightImages[0] = pano_stitcher.create_mosaic([warped, cv2.cvtColor(rightImages[0],cv2.COLOR_BGR2BGRA)], [origin, (0,0)])
leftImages[0] = cv2.cvtColor(leftImages[0],cv2.COLOR_BGRA2BGR)
rawimages = []
for i in range(leftmost, rightmost+1):
rawimages.append(cv2.imread(imgdir+"3-" + str(i)+".jpg"))
rightmost -= leftmost
middle -= leftmost
leftmost -= leftmost
rawimages[middle] = cv2.cvtColor(rawimages[middle], cv2.COLOR_BGR2BGRA)
warpedimages = []
warpedcorners = []
for i in range(0,middle):
l = rawimages[i]
r = rawimages[i+1]
h = pano_stitcher.homography(r, l)
print("Finished left homography #"+str(i))
warpedimage, warpedcorner = pano_stitcher.warp_image(l, h)
print("Warped image #"+str(i))
warpedimages.append(warpedimage)
warpedcorners.append(warpedcorner)
for i in range(middle,rightmost) :
l = rawimages[i]
r = rawimages[i+1]
h = pano_stitcher.homography(l,r)
print("Finished right homography #"+str(i))
warpedimage, warpedcorner = pano_stitcher.warp_image(r, h)
print("Warped image #"+str(i))
# Arbitrary index
a = 0
for i in left:
# Show the name of the image
print sys.argv[i]
# Open it and append it to the left images list
left_images.append(cv2.imread(sys.argv[i]))
# If closest to the middle image, warp into that perspective
# Otherwise, warp into the perspective of the last image
# that was warped
if a is 0:
h = ps.homography(middle_image, left_images[a])
else:
h = ps.homography(left_warped_images[a - 1], left_images[a])
warped, origins = ps.warp_image(left_images[a], h)
left_warped_images.append(warped)
# If closest to the middle image, use normal origins
# Otherwise, use the origins + the origins of the
# last image that was warped
if a is 0:
left_images_origins.append(origins)
else:
prev_origins = left_images_origins[a - 1]
new_origins = (
origins[0] + prev_origins[0],
origins[1] + prev_origins[1]