def matches(im1, im2, matches, dist=None, options={}):
""" show a figure with lines joining the accepted matches in im1 and im2
input: im1,im2 (images as arrays), locs1,locs2 (location of features),
matchscores (as output from 'match').
"""
scale = options.get("scale", 1)
filename = options.get("filename", None)
max_dist = options.get("max_dist", 100)
line_width = options.get("line_width", 0.8)
size = options.get("size", (12, 8))
separation = options.get("separation", 20)
if scale != 1:
s = numpy.array([scale, scale])
im1_size = numpy.array(im1.shape[1::-1] * s, dtype=numpy.uint16)
im2_size = numpy.array(im2.shape[1::-1] * s, dtype=numpy.uint16)
if scale < 0.5:
im1_scaled = imaging.get_thumbnail(im1, size=im1_size)
im2_scaled = imaging.get_thumbnail(im2, size=im2_size)
else:
im1_scaled = imaging.open_img(im1, size=im1_size)
im2_scaled = imaging.open_img(im2, size=im2_size)
im3 = append_images(im1_scaled, im2_scaled, separation)
matches = [(m[0] * s, m[1] * s) for m in matches]
else:
im3 = append_images(im1, im2, separation)
im1_scaled = im1
# Create figure
fig = pylab.figure(frameon=False, figsize=size)
ax = pylab.Axes(fig, [0., 0., 1., 1.])
ax.set_axis_off()
fig.add_axes(ax)
# Display image
ax.imshow(im3)
# Get colors
if dist != None and len(dist) == len(matches):
cs = [
getRedGreen(numpy.log(d + 1) / numpy.log(max_dist)) for d in dist
]
else:
cs = ['#00aaff' for m in matches]
# Plot all lines
offset_x = im1_scaled.shape[1]
for i, ((x1, y1), (x2, y2)) in enumerate(matches):
ax.plot([x1, x2 + offset_x + separation], [y1, y2],
color=cs[i],
lw=line_width)
pylab.xlim(0, im3.shape[1])
pylab.ylim(im3.shape[0], 0)
if filename != None:
fig.savefig(filename, bbox_inches='tight', dpi=72)
def visualize_log(log, im1, im2, stop_at=None, scale=None, size=(14, 8)):
# Prepare images
if scale == None:
scale = min(1.0, 600 / float(im1.shape[1]))
s = numpy.array([scale, scale])
im1_size = numpy.array(im1.shape[1::-1] * s, dtype=numpy.uint16)
im2_size = numpy.array(im2.shape[1::-1] * s, dtype=numpy.uint16)
im1_scaled = imaging.get_thumbnail(im1, size=im1_size)
im2_scaled = imaging.get_thumbnail(im2, size=im2_size)
separation = 20
offset_x = im1_size[0] + separation
im3 = append_images(im1_scaled, im2_scaled, separation)
def translate_point(point, image="im1"):
x, y = numpy.array(point) * s
if image == "im1":
return x, y
else:
return (x + offset_x, y)
# Create figure
fig = pylab.figure(frameon=False, figsize=size)
ax = pylab.Axes(fig, [0., 0., 1., 1.])
ax.set_axis_off()
fig.add_axes(ax)
# Display image
ax.imshow(im3)
# Plot log data
for i, d in enumerate(log):
# Enough?
if stop_at != None and i > stop_at:
break
# Plot target grid
((x_min, x_max), (y_min, y_max)) = d["target_grid"]
margin = d["margin"]
# Add square
ax.add_patch(
pylab.Rectangle(translate_point((x_min + margin, y_min + margin),
"im2"),
(x_max - x_min - 2 * margin) * scale,
(y_max - y_min - 2 * margin) * scale,
fill=False))
# Add circle
ax.add_patch(
pylab.Circle(translate_point(d["query_pos"], "im1"),
d["radius"] * scale,
fill=False,
linewidth=0.5))
# Add matches
for pos_q, pos_t in d["matches"]:
x1, x2 = numpy.array([pos_q[0], pos_t[0]]) * s
y1, y2 = numpy.array([pos_q[1], pos_t[1]]) * s
ax.plot([x1, x2 + offset_x], [y1, y2], color="#2595e3", lw=1)
# Limit figure to area around im3
pylab.xlim(0, im3.shape[1])
pylab.ylim(im3.shape[0], 0)
def matches(im1, im2, matches, dist = None, options = {}) :
""" show a figure with lines joining the accepted matches in im1 and im2
input: im1,im2 (images as arrays), locs1,locs2 (location of features),
matchscores (as output from 'match').
"""
scale = options.get("scale", 1)
filename = options.get("filename", None)
max_dist = options.get("max_dist", 100)
line_width = options.get("line_width", 0.8)
size = options.get("size", (12, 8))
separation = options.get("separation", 20)
if scale != 1 :
s = numpy.array([scale, scale])
im1_size = numpy.array(im1.shape[1::-1] * s, dtype=numpy.uint16)
im2_size = numpy.array(im2.shape[1::-1] * s, dtype=numpy.uint16)
if scale < 0.5 :
im1_scaled = imaging.get_thumbnail(im1, size = im1_size)
im2_scaled = imaging.get_thumbnail(im2, size = im2_size)
else :
im1_scaled = imaging.open_img(im1, size = im1_size)
im2_scaled = imaging.open_img(im2, size = im2_size)
im3 = append_images(im1_scaled, im2_scaled, separation)
matches = [(m[0] * s, m[1] * s) for m in matches]
else :
im3 = append_images(im1,im2, separation)
im1_scaled = im1
# Create figure
fig = pylab.figure(frameon=False, figsize=size)
ax = pylab.Axes(fig, [0., 0., 1., 1.])
ax.set_axis_off()
fig.add_axes(ax)
# Display image
ax.imshow(im3)
# Get colors
if dist != None and len(dist) == len(matches) :
cs = [getRedGreen(numpy.log(d+1)/numpy.log(max_dist)) for d in dist]
else :
cs = ['#00aaff' for m in matches]
# Plot all lines
offset_x = im1_scaled.shape[1]
for i,((x1,y1),(x2,y2)) in enumerate(matches) :
ax.plot([x1, x2+offset_x + separation], [y1,y2], color=cs[i], lw=line_width)
pylab.xlim(0,im3.shape[1])
pylab.ylim(im3.shape[0],0)
if filename != None :
fig.savefig(filename, bbox_inches='tight', dpi=72)
def visualize_log(log, im1, im2, stop_at = None, scale = None, size = (14, 8)) :
# Prepare images
if scale == None :
scale = min(1.0, 600 / float(im1.shape[1]))
s = numpy.array([scale, scale])
im1_size = numpy.array(im1.shape[1::-1] * s, dtype=numpy.uint16)
im2_size = numpy.array(im2.shape[1::-1] * s, dtype=numpy.uint16)
im1_scaled = imaging.get_thumbnail(im1, size = im1_size)
im2_scaled = imaging.get_thumbnail(im2, size = im2_size)
separation = 20
offset_x = im1_size[0] + separation
im3 = append_images(im1_scaled, im2_scaled, separation)
def translate_point(point, image = "im1") :
x, y = numpy.array(point) * s
if image == "im1" :
return x, y
else :
return (x + offset_x, y)
# Create figure
fig = pylab.figure(frameon=False, figsize=size)
ax = pylab.Axes(fig, [0., 0., 1., 1.])
ax.set_axis_off()
fig.add_axes(ax)
# Display image
ax.imshow(im3)
# Plot log data
for i, d in enumerate(log) :
# Enough?
if stop_at != None and i > stop_at :
break
# Plot target grid
((x_min, x_max), (y_min, y_max)) = d["target_grid"]
margin = d["margin"]
# Add square
ax.add_patch(pylab.Rectangle( translate_point((x_min + margin, y_min + margin), "im2"),
(x_max-x_min - 2*margin) * scale, (y_max-y_min - 2*margin) * scale, fill = False))
# Add circle
ax.add_patch(pylab.Circle( translate_point(d["query_pos"], "im1"), d["radius"] * scale, fill = False, linewidth = 0.5))
# Add matches
for pos_q, pos_t in d["matches"] :
x1, x2 = numpy.array([pos_q[0], pos_t[0]]) * s
y1, y2 = numpy.array([pos_q[1], pos_t[1]]) * s
ax.plot([x1, x2+offset_x], [y1,y2], color="#2595e3", lw=1)
# Limit figure to area around im3
pylab.xlim(0,im3.shape[1])
pylab.ylim(im3.shape[0],0)
def match_thumbs(img, cache, thumb_size = (400, 400)) :
# Load target and find descriptors and size
target = get_thumbnail(img, thumb_size)
t_orig_size = get_size(img)
get_features(target)
t_keypoints, t_descriptors = get_features(target)
# Similar for query
q_descriptors = cache.thumb["descriptors"]
q_distances = cache.thumb["distances"]
q_pos = cache.thumb["positions"]
# match thumbnails and find ratio
matcher = cv2.BFMatcher(cv2.NORM_L2, crossCheck=True)
matches = [m for m in matcher.knnMatch(q_descriptors, t_descriptors, k=1) if len(m) > 0]
ratios = numpy.array([m[0].distance / q_distances[m[0].queryIdx] for m in matches])
# Get positions of points and their scaling factor
t_pos = [t_keypoints[m[0].trainIdx].pt for m in matches]
t_ratio_x = t_orig_size[0]/float(target.shape[1])
t_ratio_y = t_orig_size[1]/float(target.shape[0])
t_ratio = numpy.array([t_ratio_x, t_ratio_y])
q_pos = numpy.array([cache.thumb["positions"][m[0].queryIdx] for m in matches])
q_ratio_x = cache.original["size"][0]/float(cache.thumb["size"][0])
q_ratio_y = cache.original["size"][1]/float(cache.thumb["size"][1])
q_ratio = numpy.array([q_ratio_x, q_ratio_y])
# Sort ratios and scale positions
indices = numpy.argsort(ratios)
pos_scaled = numpy.array([(q_p * q_ratio, t_p * t_ratio) for q_p, t_p in zip(q_pos, t_pos)])
return pos_scaled[indices], ratios[indices]
def create_thumbnail(self, path, max_size):
""" Get relevant data for thumbnail """
# Create thumbnail
thumbnail = imaging.get_thumbnail(path, max_size)
# Get thumbnail features
keypoints, descriptors = matchutil.get_features(thumbnail)
# Get nearest neighbor within image (vector with touples of feature points and distances)
matches = matchutil.bf_match(descriptors, descriptors, k=2)
# Distances to nearest neighbor and positions
nn_distances = [r[1].distance for r in matches]
positions = [k.pt for k in keypoints]
# Collect data
self.thumb = {
"descriptors": descriptors,
"positions": positions,
"distances": nn_distances,
"size": thumbnail.shape
}
def create_thumbnail(self, path, max_size) :
""" Get relevant data for thumbnail """
# Create thumbnail
thumbnail = imaging.get_thumbnail(path, max_size)
# Get thumbnail features
keypoints, descriptors = matchutil.get_features(thumbnail)
# Get nearest neighbor within image (vector with touples of feature points and distances)
matches = matchutil.bf_match(descriptors, descriptors, k = 2)
# Distances to nearest neighbor and positions
nn_distances = [r[1].distance for r in matches]
positions = [k.pt for k in keypoints]
# Collect data
self.thumb = {
"descriptors" : descriptors,
"positions" : positions,
"distances" : nn_distances,
"size" : thumbnail.shape
}
def match_thumbs(img, cache, thumb_size=(400, 400)):
# Load target and find descriptors and size
target = get_thumbnail(img, thumb_size)
t_orig_size = get_size(img)
get_features(target)
t_keypoints, t_descriptors = get_features(target)
# Similar for query
q_descriptors = cache.thumb["descriptors"]
q_distances = cache.thumb["distances"]
q_pos = cache.thumb["positions"]
# match thumbnails and find ratio
matcher = cv2.BFMatcher(cv2.NORM_L2, crossCheck=True)
matches = [
m for m in matcher.knnMatch(q_descriptors, t_descriptors, k=1)
if len(m) > 0
]
ratios = numpy.array(
[m[0].distance / q_distances[m[0].queryIdx] for m in matches])
# Get positions of points and their scaling factor
t_pos = [t_keypoints[m[0].trainIdx].pt for m in matches]
t_ratio_x = t_orig_size[0] / float(target.shape[1])
t_ratio_y = t_orig_size[1] / float(target.shape[0])
t_ratio = numpy.array([t_ratio_x, t_ratio_y])
q_pos = numpy.array(
[cache.thumb["positions"][m[0].queryIdx] for m in matches])
q_ratio_x = cache.original["size"][0] / float(cache.thumb["size"][0])
q_ratio_y = cache.original["size"][1] / float(cache.thumb["size"][1])
q_ratio = numpy.array([q_ratio_x, q_ratio_y])
# Sort ratios and scale positions
indices = numpy.argsort(ratios)
pos_scaled = numpy.array([(q_p * q_ratio, t_p * t_ratio)
for q_p, t_p in zip(q_pos, t_pos)])
return pos_scaled[indices], ratios[indices]
