def oxford():
mem = Memory(cachedir='.')
image1 = mean(imread('keble_a.jpg'), 2)[::-1].astype(np.float)
image2 = mean(imread('keble_b.jpg'), 2)[::-1].astype(np.float)
image3 = mean(imread('keble_c.jpg'), 2)[::-1].astype(np.float)
image1 = image1
# image2 = image1 + 10
image2 = image2
# image2 = mean(imread('keble_b.jpg'), 2)[::-1]
#FIXME we assume that image1.shape = image2.shape
# the image is rotated
coords1 = mem.cache(detect_harris_detector)(image1, threshold=.995)
key_points1 = utils.create_frames_from_harris_points(coords1)
# the image is rotated
coords2 = mem.cache(detect_harris_detector)(image2, threshold=.995)
key_points2 = utils.create_frames_from_harris_points(coords2)
coords3 = mem.cache(detect_harris_detector)(image3, threshold=.999)
key_points3 = utils.create_frames_from_harris_points(coords3)
# Rearrange the keypoints to be close
if 0:
import hungarian
dist = euclidean_distances(key_points1.T, key_points2.T)
ordering = mem.cache(hungarian.hungarian)(dist)[:, 1]
key_points2 = key_points2[:, ordering]
# Get sift descriptors
f1, d1 = mem.cache(vl_sift)(np.array(image1, 'f', order='F'))
#frames=key_points1,
#orientations=False)
f1, d1 = f1.transpose(), d1.transpose()
# Get sift descriptors
f2, d2 = mem.cache(vl_sift)(np.array(image2, 'f', order='F'))
#frames=key_points2,
#orientations=False)
f2, d2 = f2.transpose(), d2.transpose()
# f3, d3 = mem.cache(vl_sift)(np.array(image3, 'f', order='F'),
# frames=key_points3,
# orientations=False)
# f3, d3 = f3.transpose(), d3.transpose()
#
matched_desc, matches_d = match_descriptors(d1, d2, f1, f2)
matched_desc = np.array(matched_desc)
matches_d = np.array(matches_d)
match_image = show_matched_desc(image1.copy(), image2.copy(), matched_desc.copy())
return match_image
image1 = show_sift_desc(image1, f1)
image2 = show_sift_desc(image2, f2)
image3 = show_sift_desc(image3, f3)
# points12 = np.array([[340, 38, 51, 23],
# [359, 340, 62, 337],
# [691, 286, 395, 286],
# [655, 116, 367, 128]])
points12 = np.array([[38, 340, 23, 51],
[340, 359, 337, 62],
[286, 691, 286, 395],
[116, 655, 128, 367]])
em12 = stitchLR(image1, image2, points12)
#return em12
points32 = np.array([[621, 18, 323, 35],
[323, 44, 11, 22],
[435, 398, 125, 400],
[349, 360, 34, 363],
[653, 336, 344, 340]])
points23 = np.array([[35, 323, 18, 621],
[22, 11, 44, 323],
[400, 125, 398, 435],
[363, 34, 360, 349],
[340, 344, 336, 653]])
em23 = stitchRL(image2, image3, points23)
em = np.zeros((image2.shape[0], image2.shape[1] + 1000))
em[:, :image2.shape[1] + 500] = em12
em[:, 500:] = em23
imsave( "oxford.eps", em)
return em
def breteuil():
#if __name__ == "__main__":
mem = Memory(cachedir='.')
if 1:
image1 = mean(imread('breteuil_1.jpg'), 2)[::-1].astype(np.float)
image2 = mean(imread('breteuil_2.jpg'), 2)[::-1].astype(np.float)
image3 = mean(imread('breteuil_3.jpg'), 2)[::-1].astype(np.float)
image1 = image1
# image2 = image1 + 10
image2 = image2
else:
image = np.zeros((300, 400))
image += 30
np.random.seed(0)
image += np.random.random(size=image.shape)
image[125:175, 125:175] = 125
image[100:150, 150:200] = 200
#image = ndimage.gaussian_filter(image, 3)
image1 = image.copy()[50:250, 70:270]
image2 = image.copy()[50:250, 50:250]
# image2 = mean(imread('keble_b.jpg'), 2)[::-1]
#FIXME we assume that image1.shape = image2.shape
# the image is rotated
coords1 = mem.cache(detect_harris_detector)(image1, threshold=.995)
key_points1 = utils.create_frames_from_harris_points(coords1)
# the image is rotated
coords2 = mem.cache(detect_harris_detector)(image2, threshold=.995)
key_points2 = utils.create_frames_from_harris_points(coords2)
coords3 = mem.cache(detect_harris_detector)(image3, threshold=.995)
key_points3 = utils.create_frames_from_harris_points(coords3)
# Rearrange the keypoints to be close
if 0:
import hungarian
dist = euclidean_distances(key_points1.T, key_points2.T)
ordering = mem.cache(hungarian.hungarian)(dist)[:, 1]
key_points2 = key_points2[:, ordering]
# Get sift descriptors
f1, d1 = mem.cache(vl_sift)(np.array(image1, 'f', order='F'),
frames=key_points1,
orientations=False)
f1, d1 = f1.transpose(), d1.transpose()
# Get sift descriptors
f2, d2 = mem.cache(vl_sift)(np.array(image2, 'f', order='F'),
frames=key_points2,
orientations=False)
f2, d2 = f2.transpose(), d2.transpose()
f3, d3 = mem.cache(vl_sift)(np.array(image3, 'f', order='F'),
frames=key_points3,
orientations=False)
f3, d3 = f3.transpose(), d3.transpose()
matched_desc, matches_d = match_descriptors(d1, d2, f1, f2)
matched_desc = np.array(matched_desc)
matches_d = np.array(matches_d)
image1 = show_sift_desc(image1, f1)
image2 = show_sift_desc(image2, f2)
image3 = show_sift_desc(image3, f3)
match_image = show_matched_desc(image1, image2, matched_desc)
points12 = np.array([[386, 143, 157, 136],
[327, 261, 87, 255],
[738, 196, 489, 209],
[584, 246, 364, 251]])
em12 = stitchLR(image1, image2, points12)
points32 = np.array([[431, 178, 65, 180],
[545, 175, 190, 185],
[445, 307, 77, 318],
[720, 258, 347, 268],
[543, 206, 188, 213],
[740, 248, 367, 256]])
points23 = points32.copy()
points23[:, :2] = points32[:, 2:]
points23[:, 2:] = points32[:, :2]
em23 = stitchRL(image2, image3, points23)
em = np.zeros((image2.shape[0], image2.shape[1] + 1000))
em[:, :image2.shape[1] + 500] = em12
em[:, 500:] = em23
imsave("breteuil.png", em)
return em
image += 30
np.random.seed(0)
image += np.random.random(size=image.shape)
image[125:175, 125:175] = 125
image[100:150, 150:200] = 200
#image = ndimage.gaussian_filter(image, 3)
image1 = image.copy()[50:250, 70:270]
image2 = image.copy()[50:250, 50:250]
# image2 = mean(imread('keble_b.jpg'), 2)[::-1]
#FIXME we assume that image1.shape = image2.shape
# the image is rotated
coords1 = mem.cache(detect_harris_detector)(image1, threshold=.995)
key_points1 = utils.create_frames_from_harris_points(coords1)
# the image is rotated
coords2 = mem.cache(detect_harris_detector)(image2, threshold=.995)
key_points2 = utils.create_frames_from_harris_points(coords2)
coords3 = mem.cache(detect_harris_detector)(image3, threshold=.995)
key_points3 = utils.create_frames_from_harris_points(coords3)
# Rearrange the keypoints to be close
if 0:
import hungarian
dist = euclidean_distances(key_points1.T, key_points2.T)
ordering = mem.cache(hungarian.hungarian)(dist)[:, 1]
key_points2 = key_points2[:, ordering]