Created
November 1, 2013 20:17
-
-
Save kphilipp/7271334 to your computer and use it in GitHub Desktop.
demo for orb descriptor matching with opencv
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# further information: | |
# * http://stackoverflow.com/questions/11114349/how-to-visualize-descriptor-matching-using-opencv-module-in-python | |
# * http://docs.opencv.org/doc/tutorials/features2d/feature_homography/feature_homography.html#feature-homography | |
# * http://stackoverflow.com/questions/9539473/opencv-orb-not-finding-matches-once-rotation-scale-invariances-are-introduced | |
# * OpenCV 2 Computer Vision Application Programming Cookbook, Chapter 9 | |
import cv2 | |
import scipy as sp | |
import numpy as np | |
ratio = 0.65 | |
""" Clear matches for which NN ratio is > than threshold """ | |
def filter_distance(matches): | |
dist = [m.distance for m in matches] | |
thres_dist = (sum(dist) / len(dist)) * ratio | |
# keep only the reasonable matches | |
sel_matches = [m for m in matches if m.distance < thres_dist] | |
print '#selected matches:%d (out of %d)' % (len(sel_matches), len(matches)) | |
return sel_matches | |
""" keep only symmetric matches """ | |
def filter_asymmetric(matches, matches2): | |
sel_matches = [] | |
for match1 in matches: | |
for match2 in matches2: | |
if k_ftr[match1.queryIdx] == k_ftr[match2.trainIdx] and k_scene[match1.trainIdx] == k_scene[match2.queryIdx]: | |
sel_matches.append(match1) | |
break | |
return sel_matches | |
# Todo: filter_ransac | |
def filter_matches(matches, matches2): | |
matches = filter_distance(matches) | |
matches2 = filter_distance(matches2) | |
return filter_asymmetric(matches, matches2) | |
img1_path = "test6j.jpg" | |
img2_path = "features/castle.jpg" | |
img_scene = cv2.imread(img1_path, cv2.CV_LOAD_IMAGE_GRAYSCALE) | |
img_ftr = cv2.imread(img2_path, cv2.CV_LOAD_IMAGE_GRAYSCALE) | |
detector = cv2.FeatureDetector_create("ORB") #SURF | |
descriptor = cv2.DescriptorExtractor_create("ORB") #BRIEF | |
matcher = cv2.DescriptorMatcher_create("BruteForce-Hamming") #FlannBased #BruteForce-Hamming | |
# detect keypoints | |
kp_scene = detector.detect(img_scene) | |
kp_ftr = detector.detect(img_ftr) | |
print '#keypoints in image1: %d, image2: %d' % (len(kp_scene), len(kp_ftr)) | |
# descriptors | |
k_scene, d_scene = descriptor.compute(img_scene, kp_scene) | |
k_ftr, d_ftr = descriptor.compute(img_ftr, kp_ftr) | |
print '#keypoints in image1: %d, image2: %d' % (len(d_scene), len(d_ftr)) | |
# match the keypoints | |
matches = matcher.match(d_scene, d_ftr) | |
matches2 = matcher.match(d_ftr, d_scene) | |
# visualize the matches | |
print '#matches:', len(matches) | |
dist = [m.distance for m in matches] | |
print 'distance: min: %.3f' % min(dist) | |
print 'distance: mean: %.3f' % (sum(dist) / len(dist)) | |
print 'distance: max: %.3f' % max(dist) | |
""" filter matches """ | |
sel_matches = filter_matches(matches,matches2) | |
""" localize object """ | |
h_scene, w_scene = img_scene.shape[:2] | |
h_ftr, w_ftr = img_ftr.shape[:2] | |
ftr =[] | |
scene = [] | |
for m in sel_matches: | |
scene.append(k_scene[m.queryIdx].pt) | |
ftr.append(k_ftr[m.trainIdx].pt) | |
ftr = np.float32(ftr) | |
scene = np.float32(scene) | |
homography, mask = cv2.findHomography(ftr, scene, cv2.RANSAC) | |
ftr_corners = np.float32([[0, 0], [w_ftr, 0], [w_ftr, h_ftr], [0, h_ftr]]).reshape(1, -1, 2) | |
corners = np.int32( cv2.perspectiveTransform(ftr_corners, homography).reshape(-1, 2) ) | |
""" visualization """ | |
view = sp.zeros((max(h_scene, h_ftr), w_scene + w_ftr, 3), np.uint8) | |
view[:h_scene, :w_scene, 0] = img_scene | |
view[:h_ftr, w_scene:, 0] = img_ftr | |
view[:, :, 1] = view[:, :, 0] | |
view[:, :, 2] = view[:, :, 0] | |
for m in sel_matches: | |
# draw the keypoints | |
color = tuple([sp.random.randint(0, 255) for _ in xrange(3)]) | |
cv2.line(view, (int(k_scene[m.queryIdx].pt[0]), int(k_scene[m.queryIdx].pt[1])), | |
(int(k_ftr[m.trainIdx].pt[0] + w_scene), int(k_ftr[m.trainIdx].pt[1])), color, 2) | |
cv2.polylines(view, [np.int32([c+[w_scene,0] for c in ftr_corners])], True, (0, 255, 0), 2) | |
cv2.polylines(view, [corners], True, (0, 255, 0), 2) | |
#cv2.imshow("view", view) | |
cv2.imwrite("output.jpg", view) | |
#cv2.waitKey() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Are you sure that
if k_ftr[match1.queryIdx] == k_ftr[match2.trainIdx] and k_scene[match1.trainIdx] == k_scene[match2.queryIdx]:
needs k_ftr and k_scene?