def locate_thumbnail(thumbnail_filename, source_filename, display=False, save_visualization=False,
save_reconstruction=False, reconstruction_format="jpg"):
thumbnail_basename, thumbnail_image = open_image(thumbnail_filename)
source_basename, source_image = open_image(source_filename)
logging.info("Attempting to locate %s within %s", thumbnail_filename, source_filename)
kp_pairs = match_images(thumbnail_image, source_image)
if len(kp_pairs) >= 4:
title = "Found %d matches" % len(kp_pairs)
logging.info(title)
H, mask = find_homography(kp_pairs)
new_thumbnail, corners, rotation = reconstruct_thumbnail(thumbnail_image, source_image, kp_pairs, H)
print(json.dumps({
"master": {
"source": source_filename,
"dimensions": {
"height": source_image.shape[0],
"width": source_image.shape[1],
}
},
"thumbnail": {
"source": thumbnail_filename,
"dimensions": {
"height": thumbnail_image.shape[0],
"width": thumbnail_image.shape[1],
}
},
"bounding_box": {
"height": corners[0][1] - corners[0][0],
"width": corners[1][1] - corners[1][0],
"x": corners[1][0],
"y": corners[0][0],
},
"rotation_degrees": rotation
}))
if save_reconstruction:
new_filename = "%s.reconstructed.%s" % (thumbnail_basename, reconstruction_format)
cv2.imwrite(new_filename, new_thumbnail)
logging.info("Saved reconstructed thumbnail %s", new_filename)
else:
logging.warning("Found only %d matches; skipping reconstruction", len(kp_pairs))
new_thumbnail = corners = H = mask = None
if display or save_visualization:
vis_image = visualize_matches(source_image, thumbnail_image, new_thumbnail, corners, kp_pairs, mask)
if save_visualization:
vis_filename = "%s.visualized%s" % os.path.splitext(thumbnail_filename)
cv2.imwrite(vis_filename, vis_image)
logging.info("Saved match visualization %s", vis_filename)
if display:
cv2.imshow(title, vis_image)
cv2.waitKey()
cv2.destroyAllWindows()
def display_images(extractor, files):
window_name = "Controls"
images = []
for f in files:
print("Loading %s" % f)
try:
images.append(open_image(f))
except Exception as exc:
print(exc, file=sys.stderr)
continue
def update_display(*args):
extractor.canny_threshold = cv2.getTrackbarPos("Canny Threshold",
window_name)
extractor.erosion_element = cv2.getTrackbarPos("Erosion Element",
window_name)
extractor.erosion_size = cv2.getTrackbarPos("Erosion Size",
window_name)
extractor.dilation_element = cv2.getTrackbarPos(
"Dilation Element", window_name)
extractor.dilation_size = cv2.getTrackbarPos("Dilation Size",
window_name)
# TODO: tame configuration hideousness:
labels = [
"Canny Threshold: %s" % extractor.canny_threshold,
"Erosion Element: %s" %
FigureExtractor.MORPH_TYPE_KEYS[extractor.erosion_element],
"Erosion Size: %s" % extractor.erosion_size,
"Dilation Element: %s" %
FigureExtractor.MORPH_TYPE_KEYS[extractor.dilation_element],
"Dilation Size: %s" % extractor.dilation_size
]
labels_img = numpy.zeros((30 * (len(labels) + 1), 600, 3), numpy.uint8)
for i, label in enumerate(labels, 1):
cv2.putText(labels_img, label, (0, i * 30),
cv2.FONT_HERSHEY_SIMPLEX, 1.0, (192, 192, 192))
cv2.imshow("Controls", labels_img)
print("Settings:\n\t", "\n\t".join(labels))
print()
for name, image in images:
filtered_image = extractor.filter_image(image)
contours, hierarchy = extractor.find_contours(filtered_image)
# The filtered image will be heavily processed down to 1-bit depth. We'll convert it to RGB
# so we can display the effects of the filters with full-color overlays for detected figures:
output = cv2.cvtColor(filtered_image, cv2.COLOR_GRAY2RGB)
print("Processing %s" % name)
for bbox in extractor.get_bounding_boxes_from_contours(
contours, filtered_image):
print("\tExtract: %s" % bbox)
output[bbox.image_slice] = image[bbox.image_slice]
cv2.polylines(output,
bbox.poly,
True, (32, 192, 32),
thickness=3)
cv2.drawContours(output,
contours,
bbox.contour_index, (32, 192, 32),
hierarchy=hierarchy,
maxLevel=0)
cv2.rectangle(output, (bbox.x1, bbox.y1), (bbox.x2, bbox.y2),
color=(32, 192, 192))
cv2.imshow(name, output)
cv2.namedWindow(window_name)
cv2.resizeWindow(window_name, 600, 340)
cv2.createTrackbar("Canny Threshold", window_name,
extractor.canny_threshold, 255, update_display)
cv2.createTrackbar("Erosion Element", window_name,
extractor.erosion_element,
len(extractor.MORPH_TYPES) - 1, update_display)
cv2.createTrackbar("Erosion Size", window_name, extractor.erosion_size, 64,
update_display)
cv2.createTrackbar("Dilation Element", window_name,
extractor.dilation_element,
len(extractor.MORPH_TYPES) - 1, update_display)
cv2.createTrackbar("Dilation Size", window_name, extractor.dilation_size,
64, update_display)
update_display()
if args.interactive:
while cv2.waitKey() not in (13, 27):
continue
cv2.destroyAllWindows()
except ImportError:
import pdb
# FIXME: we should have a way to enumerate this from FigureExtractor and feed argparse that way:
param_names = [action.dest for action in extraction_params._group_actions]
params = {k: v for (k, v) in args._get_kwargs() if k in param_names}
try:
extractor = FigureExtractor(**params)
if args.interactive:
display_images(extractor, args.files)
else:
for f in args.files:
try:
base_name, source_image = open_image(f)
except Exception as exc:
print(exc, file=sys.stderr)
continue
output_base = os.path.join(output_dir, base_name)
print("Processing %s" % f)
boxes = []
for i, bbox in enumerate(extractor.find_figures(source_image),
1):
extracted = source_image[bbox.image_slice]
extract_filename = os.path.join(
output_dir, "%s-%d.jpg" % (output_base, i))
def display_images(extractor, files):
window_name = "Controls"
images = []
for f in files:
print "Loading %s" % f
try:
images.append(open_image(f))
except StandardError as exc:
print >>sys.stderr, exc
continue
def update_display(*args):
extractor.canny_threshold = cv2.getTrackbarPos("Canny Threshold", window_name)
extractor.erosion_element = cv2.getTrackbarPos("Erosion Element", window_name)
extractor.erosion_size = cv2.getTrackbarPos("Erosion Size", window_name)
extractor.dilation_element = cv2.getTrackbarPos("Dilation Element", window_name)
extractor.dilation_size = cv2.getTrackbarPos("Dilation Size", window_name)
# TODO: tame configuration hideousness:
labels = ["Canny Threshold: %s" % extractor.canny_threshold,
"Erosion Element: %s" % FigureExtractor.MORPH_TYPE_KEYS[extractor.erosion_element],
"Erosion Size: %s" % extractor.erosion_size,
"Dilation Element: %s" % FigureExtractor.MORPH_TYPE_KEYS[extractor.dilation_element],
"Dilation Size: %s" % extractor.dilation_size]
labels_img = numpy.zeros((30 * (len(labels) + 1), 600, 3), numpy.uint8)
for i, label in enumerate(labels, 1):
cv2.putText(labels_img, label, (0, i * 30), cv2.FONT_HERSHEY_SIMPLEX, 1.0, (192, 192, 192))
cv2.imshow("Controls", labels_img)
print "Settings:\n\t", "\n\t".join(labels)
print
for name, image in images:
filtered_image = extractor.filter_image(image)
contours, hierarchy = extractor.find_contours(filtered_image)
# The filtered image will be heavily processed down to 1-bit depth. We'll convert it to RGB
# so we can display the effects of the filters with full-color overlays for detected figures:
output = cv2.cvtColor(filtered_image, cv2.COLOR_GRAY2RGB)
print "Processing %s" % name
for bbox in extractor.get_bounding_boxes_from_contours(contours, filtered_image):
print "\tExtract: %s" % bbox
output[bbox.image_slice] = image[bbox.image_slice]
cv2.polylines(output, bbox.poly, True, (32, 192, 32), thickness=3)
cv2.drawContours(output, contours, bbox.contour_index, (32, 192, 32), hierarchy=hierarchy, maxLevel=0)
cv2.rectangle(output, (bbox.x1, bbox.y1), (bbox.x2, bbox.y2), color=(32, 192, 192))
cv2.imshow(name, output)
cv2.namedWindow(window_name)
cv2.resizeWindow(window_name, 600, 340)
cv2.createTrackbar("Canny Threshold", window_name, extractor.canny_threshold, 255, update_display)
cv2.createTrackbar("Erosion Element", window_name, extractor.erosion_element, len(extractor.MORPH_TYPES) - 1, update_display)
cv2.createTrackbar("Erosion Size", window_name, extractor.erosion_size, 64, update_display)
cv2.createTrackbar("Dilation Element", window_name, extractor.dilation_element, len(extractor.MORPH_TYPES) - 1, update_display)
cv2.createTrackbar("Dilation Size", window_name, extractor.dilation_size, 64, update_display)
update_display()
if args.interactive:
while cv2.waitKey() not in (13, 27):
continue
cv2.destroyAllWindows()
import pdb
# FIXME: we should have a way to enumerate this from FigureExtractor and feed argparse that way:
param_names = [action.dest for action in extraction_params._group_actions]
params = {k: v for (k, v) in args._get_kwargs() if k in param_names}
try:
extractor = FigureExtractor(**params)
print("MADE EXTRACTOR")
if args.interactive:
display_images(extractor, args.files)
else:
for f in args.files:
try:
base_name, source_image = open_image(f)
except StandardError as exc:
print >>sys.stderr, exc
continue
output_base = os.path.join(output_dir, base_name)
print "Processing %s" % f
boxes = []
for i, bbox in enumerate(extractor.find_figures(source_image), 1):
extracted = source_image[bbox.image_slice]
extract_filename = os.path.join(output_dir, "%s-%d.jpg" % (output_base, i))
print "\tSaving %s" % extract_filename
cv2.imwrite(extract_filename, extracted)
def locate_thumbnail(thumbnail_filename, source_filename, display=False, save_visualization=False,
save_reconstruction=False, reconstruction_format="jpg",
max_aspect_ratio_delta=0.1, match_aspect_ratio=False,
minimum_matches=10,
json_output_filename=None, max_master_edge=4096, max_output_edge=2048):
thumbnail_basename, thumbnail_image = open_image(thumbnail_filename)
source_basename, source_image = open_image(source_filename)
if (((source_image.shape[0] <= thumbnail_image.shape[0])
or (source_image.shape[1] <= thumbnail_image.shape[1]))):
raise RuntimeError("Master file wasn't larger than the thumbnail: %r vs %r" % (source_image.shape,
thumbnail_image.shape))
logging.info("Attempting to locate %s within %s", thumbnail_filename, source_filename)
full_source_image = source_image
if max_master_edge and any(i for i in source_image.shape if i > max_master_edge):
logging.info("Resizing master to fit within %d pixels", max_master_edge)
source_image = fit_image_within(source_image, max_master_edge, max_master_edge)
logging.info('Finding common features')
kp_pairs = match_images(thumbnail_image, source_image)
if len(kp_pairs) >= minimum_matches:
title = "Found %d matches" % len(kp_pairs)
logging.info(title)
H, mask = find_homography(kp_pairs)
corners = get_scaled_corners(thumbnail_image, source_image, full_source_image, kp_pairs, H)
new_thumbnail, corners, rotation = reconstruct_thumbnail(thumbnail_image, full_source_image, corners,
match_aspect_ratio=match_aspect_ratio,
max_aspect_ratio_delta=max_aspect_ratio_delta)
if json_output_filename:
with open(json_output_filename, mode='wb') as json_file:
json.dump({
"master": {
"source": source_filename,
"dimensions": {
"height": full_source_image.shape[0],
"width": full_source_image.shape[1],
}
},
"thumbnail": {
"source": thumbnail_filename,
"dimensions": {
"height": thumbnail_image.shape[0],
"width": thumbnail_image.shape[1],
}
},
"bounding_box": {
"height": corners[0][1] - corners[0][0],
"width": corners[1][1] - corners[1][0],
"x": corners[1][0],
"y": corners[0][0],
},
"rotation_degrees": rotation
}, json_file, indent=4)
if save_reconstruction:
new_filename = "%s.reconstructed.%s" % (thumbnail_basename, reconstruction_format)
new_thumb_img = fit_image_within(new_thumbnail, max_output_edge, max_output_edge)
cv2.imwrite(new_filename, new_thumb_img)
logging.info("Saved reconstructed %s thumbnail %s", new_thumb_img.shape[:2], new_filename)
else:
logging.warning("Found only %d matches; skipping reconstruction", len(kp_pairs))
title = "MATCH FAILED: %d pairs" % len(kp_pairs)
new_thumbnail = corners = H = mask = None
if display or save_visualization:
vis_image = visualize_matches(source_image, thumbnail_image, new_thumbnail, corners, kp_pairs, mask)
if save_visualization:
vis_filename = "%s.visualized%s" % os.path.splitext(thumbnail_filename)
cv2.imwrite(vis_filename, vis_image)
logging.info("Saved match visualization %s", vis_filename)
if display:
# This may or may not exist depending on whether OpenCV was compiled using the QT backend:
window_flags = getattr(cv, 'CV_WINDOW_NORMAL', cv.CV_WINDOW_AUTOSIZE)
window_title = '%s - %s' % (thumbnail_basename, title)
cv2.namedWindow(window_title, flags=window_flags)
cv2.imshow(window_title, vis_image)
cv2.waitKey()
cv2.destroyAllWindows()
def locate_thumbnail(thumbnail_filename,
source_filename,
display=False,
save_visualization=False,
save_reconstruction=False,
reconstruction_format="jpg",
max_aspect_ratio_delta=0.1,
match_aspect_ratio=False,
minimum_matches=10,
json_output_filename=None,
max_master_edge=4096,
max_output_edge=2048):
thumbnail_basename, thumbnail_image = open_image(thumbnail_filename)
source_basename, source_image = open_image(source_filename)
if (((source_image.shape[0] <= thumbnail_image.shape[0])
or (source_image.shape[1] <= thumbnail_image.shape[1]))):
raise RuntimeError(
"Master file wasn't larger than the thumbnail: %r vs %r" %
(source_image.shape, thumbnail_image.shape))
logging.info("Attempting to locate %s within %s", thumbnail_filename,
source_filename)
full_source_image = source_image
if max_master_edge and any(
i for i in source_image.shape if i > max_master_edge):
logging.info("Resizing master to fit within %d pixels",
max_master_edge)
source_image = fit_image_within(source_image, max_master_edge,
max_master_edge)
logging.info('Finding common features')
kp_pairs = match_images(thumbnail_image, source_image)
if len(kp_pairs) >= minimum_matches:
title = "Found %d matches" % len(kp_pairs)
logging.info(title)
H, mask = find_homography(kp_pairs)
corners = get_scaled_corners(thumbnail_image, source_image,
full_source_image, kp_pairs, H)
new_thumbnail, corners, rotation = reconstruct_thumbnail(
thumbnail_image,
full_source_image,
corners,
match_aspect_ratio=match_aspect_ratio,
max_aspect_ratio_delta=max_aspect_ratio_delta)
if json_output_filename:
with open(json_output_filename, mode='wb') as json_file:
json.dump(
{
"master": {
"source": source_filename,
"dimensions": {
"height": full_source_image.shape[0],
"width": full_source_image.shape[1],
}
},
"thumbnail": {
"source": thumbnail_filename,
"dimensions": {
"height": thumbnail_image.shape[0],
"width": thumbnail_image.shape[1],
}
},
"bounding_box": {
"height": corners[0][1] - corners[0][0],
"width": corners[1][1] - corners[1][0],
"x": corners[1][0],
"y": corners[0][0],
},
"rotation_degrees": rotation
},
json_file,
indent=4)
if save_reconstruction:
new_filename = "%s.reconstructed.%s" % (thumbnail_basename,
reconstruction_format)
new_thumb_img = fit_image_within(new_thumbnail, max_output_edge,
max_output_edge)
cv2.imwrite(new_filename, new_thumb_img)
logging.info("Saved reconstructed %s thumbnail %s",
new_thumb_img.shape[:2], new_filename)
else:
logging.warning("Found only %d matches; skipping reconstruction",
len(kp_pairs))
title = "MATCH FAILED: %d pairs" % len(kp_pairs)
new_thumbnail = corners = H = mask = None
if display or save_visualization:
vis_image = visualize_matches(source_image, thumbnail_image,
new_thumbnail, corners, kp_pairs, mask)
if save_visualization:
vis_filename = "%s.visualized%s" % os.path.splitext(thumbnail_filename)
cv2.imwrite(vis_filename, vis_image)
logging.info("Saved match visualization %s", vis_filename)
if display:
# This may or may not exist depending on whether OpenCV was compiled using the QT backend:
window_flags = getattr(cv, 'CV_WINDOW_NORMAL', cv.CV_WINDOW_AUTOSIZE)
window_title = '%s - %s' % (thumbnail_basename, title)
cv2.namedWindow(window_title, flags=window_flags)
cv2.imshow(window_title, vis_image)
cv2.waitKey()
cv2.destroyAllWindows()