def checkMatches(fname,posfeats,negfeats,frects,orb,pmatchdist=20,nmatchdist=20,draw=False):
'''Check whether image in fname is more similar to features in posfeats or features in negfeats'''
posmatches=getMatches(fname,posfeats,frects,orb,draw)
negmatches=getMatches(fname,negfeats,frects,orb,draw)
testmatches=testMatches(posmatches,negmatches,pmatchdist,nmatchdist)
if draw:
for i,gm,feat in zip(range(0,len(posfeats)),posmatches,posfeats):
plate=gm["plate"]
kp_plate=gm["kp_plate"]
feature=feat["plate"]
kp_feature=feat["kp"]
fmatches=testmatches["posMatches"][i]
print("Number of matches: "+str(len(fmatches)))
plt.figure(figsize=(20,20))
img3 = cv2.drawMatches(feature,kp_feature,plate,kp_plate,fmatches, flags=2,outImg=None)
plt.imshow(img3),plt.savefig(os.path.basename(fname)[0:-4]+"_PosMatches_{:03}.png".format(i),bbox_inches='tight', pad_inches=0)
for i,gm,feat in zip(range(0,len(negfeats)),negmatches,negfeats):
plate=gm["plate"]
kp_plate=gm["kp_plate"]
feature=feat["plate"]
kp_feature=feat["kp"]
fmatches=testmatches["negMatches"][i]
print("Number of matches: "+str(len(fmatches)))
plt.figure(figsize=(20,20))
img3 = cv2.drawMatches(feature,kp_feature,plate,kp_plate,fmatches, flags=2,outImg=None)
plt.imshow(img3),plt.savefig(os.path.basename(fname)[0:-4]+"_NegMatches_{:03}.png".format(i),bbox_inches='tight', pad_inches=0)
return(testmatches["hit"])
def img_match(f1, f2):
from matplotlib import pyplot as plt
img1 = cv2.imread(f1, 0) # Query Image
img2 = cv2.imread(f2, 0) # Training Image
# Initiate SIFT detector
orb = cv2.ORB()
# find the keypoints and descriptors with SIFT
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)
# create BFMatcher object
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
# Match descriptors.
matches = bf.match(des1, des2)
# Sort them in the order of their distance.
matches = sorted(matches, key=lambda x: x.distance)
# Draw first 10 matches.
img3 = cv2.drawMatches(img1, kp1, img2, kp2, matches[:10], flags=2)
plt.imshow(img3), plt.show()
def find_homography_matrix():
kp1, des1, img1, kp2, des2, img2 = find_features()
flann = cv2.FlannBasedMatcher(index_params,search_params)
matches = flann.knnMatch(des1,des2,k=2)
# Store all good matches as per Lowe's ratio test
good = []
for match in matches:
if len(match) == 2:
m,n = match
if m.distance < 0.7 * n.distance:
good.append(m)
if len(good) > MIN_MATCH_COUNT:
src_pts = np.float32([kp1[m.queryIdx].pt for m in good]).reshape(-1,1,2)
dst_pts = np.float32([kp2[m.trainIdx].pt for m in good]).reshape(-1,1,2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
matchesMask = mask.ravel().tolist()
print M
h,w = img1.shape
pts = np.float32([[0,0],[0,h-1],[w-1,h-1],[w-1,0]]).reshape(-1,1,2)
dst = cv2.perspectiveTransform(pts, M)
img2 = cv2.polylines(img2,[np.int32(dst)],True,(0,0,255),3,cv2.LINE_AA)
else:
print ("Not enough matches are found - %d/%d") %(len(good),MIN_MATCH_COUNT)
matchesMask = None
draw_params = dict(matchColor = (0,0,255),
singlePointColor = None,
matchesMask = matchesMask,
flags = 2)
img3 = cv2.drawMatches(img1, kp1, img2, kp2,good,None,**draw_params)
return img3, M
def bf_matcher(source1, source2):
"""
Question 1.3
Trace les mises en correspondance des caractéristiques de deux images selon la méthode de "Brute Force"
Parameters
----------
source1: la source de l'image de gauche
source2: la source de l'image de droite
"""
# Initiate SIFT detector
orb = cv2.ORB_create()
cv2.ocl.setUseOpenCL(False)
# find the keypoints and descriptors with SIFT
kp1, des1 = orb.detectAndCompute(source1, None)
kp2, des2 = orb.detectAndCompute(source2, None)
# create BFMatcher object
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
# Match descriptors.
matches = bf.match(des1, des2)
# Sort them in the order of their distance.
matches = sorted(matches, key=lambda x: x.distance)
# Draw first 10 matches.
img3 = cv2.drawMatches(source1, kp1, source2, kp2, matches[:10], None, flags=2)
plt.imshow(img3), plt.show()
def ORB_BF_Matching():
img1 = cv2.imread('box.png',0) # queryImage
img2 = cv2.imread('box_in_scene.png',0) # trainImage
# Initiate ORB detector
orb = cv2.ORB_create()
# find the keypoints and descriptors with SIFT
# cv2.ORB.detectAndCompute(image, mask[, descriptors[, useProvidedKeypoints]]) → keypoints, descriptors
kp1, des1 = orb.detectAndCompute(img1,None)
kp2, des2 = orb.detectAndCompute(img2,None)
# create BFMatcher object
# 第一个参数normType:指定要使用的距离测试类型，默认为cv2.Norm_L2.对于ORB,BRIEF,BRISK算法,要使用cv2.NORM_HAMMING
# crossCheck默认为False,若设置为True则需双向匹配，否则只需单项匹配即可
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
# Match descriptors.
matches = bf.match(des1,des2)
# Sort them in the order of their distance.
# The result of matches = bf.match(des1,des2) line is a list of DMatch objects. This DMatch object has following attributes:
# DMatch.distance - Distance between descriptors. The lower, the better it is
# DMatch.trainIdx - Index of the descriptor in train descriptors
# DMatch.queryIdx - Index of the descriptor in query descriptors
# DMatch.imgIdx - Index of the train image
matches = sorted(matches, key = lambda x:x.distance)
# Draw first 10 matches.
# cv2.drawMatches(img1, keypoints1, img2, keypoints2, matches1to2, outImg[, matchColor[, singlePointColor[, matchesMask[, flags]]]]]) → outImg
img3 = cv2.drawMatches(img1,kp1,img2,kp2,matches[:10], None,flags=2)
plt.imshow(img3),plt.show()
def createVisualMatch(im1, im2, detector, matcher): (kp1, des1) = detector.detectAndCompute(im1, None) (kp2, des2) = detector.detectAndCompute(im2, None) matches = matcher.match(des1,des2) im3 = cv.drawMatches(im1, kp1, im2, kp2, matches[:10], None, flags=2) return im3
def SIFTMatch(self, imagePath, display_results=False):
'''
Performs a match using Scale-Invariant Feature Transform algorithm.
Matching is done with Fast Library for Approximate Nearest Neighbors.
Lowe's ratio test is applied.
'''
sift = cv2.xfeatures2d.SIFT_create()
kp1, des1 = sift.detectAndCompute(self.image, None)
kp2, des2 = self.index[imagePath]
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5)
search_params = dict(checks=50)
flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.knnMatch(des1, des2, k=2)
filtered = list(filter(lambda x:x[0].distance < 0.7*x[1].distance, matches))
good = list(map(lambda x: x[0], filtered))
if display_results:
draw_params = dict(matchColor=(0,255,0),
singlePointColor=None,
flags=2)
result = cv2.drawMatches(self.image, kp1, training, kp2, good, None, **draw_params)
plt.imshow(result), plt.show()
return len(good)
def SURF(img2, debug):
# Initiate SURF detector
surf = cv2.xfeatures2d.SURF_create()
# find the keypoints and descriptors with SURF
kp1, des1 = surf.detectAndCompute(img1,None)
kp2, des2 = surf.detectAndCompute(img2,None)
#draw the keypoints
cv2.drawKeypoints(img1,kp1,None,(255,0,0),4)
# BFMatcher (Brute Force Matcher) Iniitialize with default params
bf = cv2.BFMatcher()
#do brute force matching with k nearest neighbors
matches = bf.knnMatch(des1,des2, k=2)
# Apply distance ratio test
good = []
for m,n in matches:
if m.distance < 0.7*n.distance:
good.append(m)
if len(good)>10:
src_pts = np.float32([ kp1[m.queryIdx].pt for m in good ]).reshape(-1,1,2)
dst_pts = np.float32([ kp2[m.trainIdx].pt for m in good ]).reshape(-1,1,2)
img3 = cv2.drawMatches(img1,kp1,img2,kp2,good,None,flags=2)
if img3!= None and debug:
plt.imshow(img3, 'gray'),plt.show()
print "Number of Features: ", len(good)
return good
def drawEpilinesAndMatches(imgL, imgR, pts1, pts2, kp1, kp2, good):
# Find epilines corresponding to points in right image (second image) and
# drawing its lines on left image
lines1 = cv2.computeCorrespondEpilines(pts2.reshape(-1,1,2), 2,F)
lines1 = lines1.reshape(-1,3)
img3_L = imgL
img3_R = imgR
img3_L, img3_R = drawlines(img3_L,img3_R,lines1,pts1,pts2)
fig = plt.figure()
plt.subplot(121)
plt.imshow(imgL), plt.title('Input L (no lines should be written on these variables?)')
plt.subplot(122)
plt.imshow(imgR), plt.title('Input R (no lines should be written on these variables?)')
plt.show()
# Find epilines corresponding to points in left image (first image) and
# drawing its lines on right image
lines2 = cv2.computeCorrespondEpilines(pts1.reshape(-1,1,2), 1,F)
lines2 = lines2.reshape(-1,3)
img4_L = imgL
img4_R = imgR
img4_L,img4_R = drawlines(img4_L,img4_R,lines2,pts2,pts1)
# cv2.drawMatchesKnn expects list of lists as matches.
# http://opencv-python-tutroals.readthedocs.org/en/latest/py_tutorials/py_feature2d/py_matcher/py_matcher.html
imgDummy = np.zeros((1,1))
img5 = cv2.drawMatches(imgL,kp1,imgR,kp2,good[:10],imgDummy)
return img3_L, img3_R, img4_L, img4_R, img5, lines1, lines2
def draw_matches(self, other, num=10):
kp1, des1 = self.orb_keypoints()
kp2, des2 = other.orb_keypoints()
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck = True)
matches = bf.match(des1, des2)
matches = sorted(matches, key = lambda x: x.distance)
return cv2.drawMatches(self._ocvimg, kp1, imgunwrap(other), kp2, matches[:num], flags=2)
def tryToMatchFeatures(orb, img1, pointInfo, img2):
kp2, des2 = pointInfo
# find the keypoints and descriptors with ORB
kp1, des1 = orb.detectAndCompute(img1,None)
if des1 is None or des2 is None:
return [], None, None
elif len(des1) == 0 or len(des2) == 0:
print("it thinks the key descriptions are empty")
return [], None, None
# BFMatcher with default params
bf = cv2.BFMatcher()
matches = bf.match(des1,des2)
sortedMatches = sorted(matches, key = lambda x: x.distance)
#for mat in matches:
# print mat.distance
goodMatches = [mat for mat in matches if mat.distance < 300]
print "Good match number:", len(goodMatches)
matchImage = cv2.drawMatches(img1, kp1, img2, kp2, goodMatches,
None, matchColor = (255, 255, 0), singlePointColor=(0, 0, 255))
cv2.imshow("Match Image", matchImage)
cv2.waitKey(0)
return goodMatches, kp1, kp2
def ORBMatch(self, imagePath, display_results=False):
'''
Matches query against specified image using the Oriented FAST and Rotated BRIEF algorithm.
Matching is done through Brute-Force.
'''
orb = cv2.ORB_create()
kp1, des1 = orb.detectAndCompute(self.image, None)
kp2, des2 = self.index[imagePath]
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(des1, des2)
matches = sorted(matches, key=lambda x: x.distance)
if display_results:
draw_params = dict(matchColor=(0,255,0),
singlePointColor=None,
flags=2)
image = cv2.drawMatches(self.image, kp1, training, kp2, matches, None, **draw_params)
plt.imshow(image), plt.show()
return len(good)
def drawMatches(image1, image2, matches, filename):
'''
Takes two images and matches and saves an image of those matches
'''
matchImage = cv2.drawMatches(image1.img, image1.kps, image2.img, image2.kps, matches, image1.img, flags=2)
img = PILImage.fromarray(matchImage, 'RGB')
img.save(filename)
def drawmatches(matchesMask,img1,img2,kp1,kp2,good):
draw_params = dict(matchColor = (0,255,0), # draw matches in green color
singlePointColor = (255,0,0),
matchesMask = matchesMask, # draw only inliers
flags = 2)
img3 = cv2.drawMatches(img1,kp1,img2,kp2,good,None,**draw_params)
cv2.imwrite('diffrrrsift1817.png',img3)
def calculate_translation_values(self, image, warp_matrix):
source_file = os.path.join(self.INPUT_DIR, image)
if self.RESET_MATRIX_EVERY_LOOP:
warp_matrix = self._create_warp_matrix() # reset
im2 = self._read_image_and_crop(source_file, read_as_8bit=True)
# proceed with downsized version
if self.DOWNSIZE:
im2_downsized = cv2.resize(im2, (0,0), fx=1.0/self.DOWNSIZE_FACTOR, fy=1.0/self.DOWNSIZE_FACTOR)
else:
im2_downsized = im2
im2_gray = cv2.cvtColor(im2_downsized, cv2.COLOR_BGR2GRAY)
if self.USE_SOBEL:
im2_gray = self._get_gradient(im2_gray)
if self.ALGORITHM == "ECC":
try:
(cc, warp_matrix) = cv2.findTransformECC(self.reference_image_gray, im2_gray, warp_matrix, self.WARP_MODE, self.CRITERIA)
except Exception as e:
raise e
if self.ALGORITHM == "ORB":
orb = cv2.ORB_create(self.MAX_FEATURES)
im2_gray = np.uint8(im2_gray)
keypoints2, descriptors2 = orb.detectAndCompute(im2_gray, None)
# Match features.
matcher = cv2.DescriptorMatcher_create(cv2.DESCRIPTOR_MATCHER_BRUTEFORCE_HAMMING)
matches = matcher.match(self.orb_descriptors1, descriptors2, None)
# Sort matches by score
matches.sort(key=lambda x: x.distance, reverse=False)
# Remove not so good matches
numGoodMatches = int(len(matches) * self.GOOD_MATCH_PERCENT)
matches = matches[:numGoodMatches]
# Draw top matches
imMatches = cv2.drawMatches(self.reference_image_gray, self.orb_keypoints1, im2_gray, keypoints2, matches, None)
cv2.imwrite(image + "_match.jpg", imMatches)
# Extract location of good matches
points1 = np.zeros((len(matches), 2), dtype=np.float32)
points2 = np.zeros((len(matches), 2), dtype=np.float32)
for i, match in enumerate(matches):
points1[i, :] = self.orb_keypoints1[match.queryIdx].pt
points2[i, :] = keypoints2[match.trainIdx].pt
# Find homography
h, mask = cv2.findHomography(points1, points2, cv2.RANSAC)
warp_matrix = h
return (im2, warp_matrix)
def SURFmatch(self, imageName):
'''
Performs a SURF image match of a query image against the specified image from
the dataset.
'''
print("SURF matching: " + imageName + "...")
MIN_MATCH_COUNT = 200
FLANN_INDEX_KDTREE = 0
query = cv2.imread(self.image)
training = cv2.imread(self.dataset + "/" + imageName)
surf = cv2.xfeatures2d.SURF_create()
kp1, des1 = surf.detectAndCompute(query, None)
kp2, des2 = surf.detectAndCompute(training, None)
index_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5)
search_params = dict(checks=50)
flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.knnMatch(des1, des2, k=2)
filtered = list(filter(lambda x:x[0].distance < 0.7*x[1].distance, matches))
good = list(map(lambda x: x[0], filtered))
if len(good) > MIN_MATCH_COUNT:
print("\tFound %s matches" % (len(good)))
src_pts = np.float32([ kp1[m.queryIdx].pt for m in good ]).reshape(-1,1,2)
dst_pts = np.float32([ kp2[m.trainIdx].pt for m in good ]).reshape(-1,1,2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
matchesMask = mask.ravel().tolist()
h,w = query.shape[:2]
pts = np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2)
dst = cv2.perspectiveTransform(pts,M)
width = cv2.imread(self.image).shape[1]
offset = 0.5 * (0.5 * (dst[0][0] + dst[1][0]) + 0.5 * (dst[2][0] - width + dst[3][0] - width))
self.estimates.append((offset[0] * 60/width + int(imageName[5:8]), len(good)))
training = cv2.polylines(training,[np.int32(dst)],True,255,3,cv2.LINE_AA)
else:
print("\tNot enough matches found - %d/%d" % (len(good), MIN_MATCH_COUNT))
if self.seeSURFMatchesFlag:
draw_params = dict(matchColor=(0,255,0),
singlePointColor=None,
matchesMask=matchesMask,
flags=2)
result = cv2.drawMatches(query,kp1,training,kp2,good,None,**draw_params)
plt.imshow(result, 'gray'), plt.show()
return len(good)
def siftFeature(prev, curr, mask):
sift = cv2.SIFT(nOctaveLayers=3, contrastThreshold=0.05, edgeThreshold=10)
#surf = cv2.SURF(1000)
kp1, des1 = sift.detectAndCompute(prev, mask)
kp2, des2 = sift.detectAndCompute(curr, None)
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5)
search_params = dict(checks=50)
flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.knnMatch(des1, des2, k=2)
# store all the good matches as per Lowe's ratio test.
good = []
for m, n in matches:
if m.distance < 0.6 * n.distance:
good.append(m)
print len(kp1), len(kp2), len(matches), len(good)
if len(good) > 10:
src_pts = np.float32(
[kp1[m.queryIdx].pt for m in good]).reshape(-1, 1, 2)
dst_pts = np.float32(
[kp2[m.trainIdx].pt for m in good]).reshape(-1, 1, 2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 3.0)
if M is None:
return None, None
matchesMask = mask.ravel().tolist()
img = curr.copy()
for idx, m in enumerate(matchesMask):
x1, y1 = map(int, src_pts[idx][0])
x2, y2 = map(int, dst_pts[idx][0])
if m == 1:
cv2.line(img, (x1, y1), (x2, y2), (0, 255, 0), 1)
cv2.circle(img, (x1, y1), 2, (0, 0, 255), -1)
cv2.circle(img, (x2, y2), 1, (0, 0, 255), -1)
draw_params = dict(matchColor=(0, 255, 0),
singlePointColor=None,
matchesMask=matchesMask,
flags=2)
img3 = cv2.drawMatches(prev,
kp1, curr, kp2, good, None, **draw_params)
cv2.imshow('sift', img3)
else:
print "Not enough matches are found - %d/%d" % (len(good), 50)
matchesMask = None
M = None
img = curr.copy()
return img, M
def main():
# This scene contains a Sword Monster.
scene = cv2.imread('hex_images/hex_-1_-1_2.png', cv2.IMREAD_COLOR)
mon_sword = cv2.imread('hoplite_assets/mon_sword_1.png', cv2.IMREAD_COLOR)
mon_bow = cv2.imread('hoplite_assets/mon_bow_1.png', cv2.IMREAD_COLOR)
mon_wizard = cv2.imread('hoplite_assets/mon_wizard_1.png',
cv2.IMREAD_COLOR)
sift = cv2.xfeatures2d.SURF_create()
# find the keypoints and descriptors with SIFT
keypoints_scene, descriptors_scene = sift.detectAndCompute(scene, None)
for mon in [mon_sword, mon_bow, mon_wizard]:
keypoints_mon, descriptors_mon = sift.detectAndCompute(mon, None)
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
search_params = dict(checks = 50)
flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.knnMatch(descriptors_scene, descriptors_mon, k=2)
# store all the good matches as per Lowe's ratio test.
good = []
for m, n in matches:
if m.distance < 0.7 * n.distance:
good.append(m)
if len(good) > MIN_MATCH_COUNT:
src_pts = np.float32([keypoints_scene[m.queryIdx].pt for m in good]).reshape(-1,1,2)
dst_pts = np.float32([keypoints_mon[m.trainIdx].pt for m in good]).reshape(-1,1,2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC,5.0)
matchesMask = mask.ravel().tolist()
h,w = scene.shape
pts = np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2)
dst = cv2.perspectiveTransform(pts,M)
mon = cv2.polylines(mon,[np.int32(dst)], True, 255, 3, cv2.LINE_AA)
else:
print "Not enough matches are found - %d/%d" % (len(good),MIN_MATCH_COUNT)
matchesMask = None
draw_params = dict(matchColor = (0,255,0), # draw matches in green color
singlePointColor = None,
matchesMask = matchesMask, # draw only inliers
flags = 2)
result = cv2.drawMatches(scene, keypoints_scene, mon, keypoints_mon,
good,None,**draw_params)
cv2.namedWindow('result', cv2.WINDOW_AUTOSIZE)
cv2.imshow('result', result)
cv2.waitKey(0)
cv2.destroyAllWindows()
def do_it(im1, im2):
orb = cv2.ORB_create()
m = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
kp1, desc1 = orb.detectAndCompute(im1, None)
kp2, desc2 = orb.detectAndCompute(im2, None)
matches = m.match(desc1, desc2)
matches = sorted(matches, key=lambda x: x.distance)
res = cv2.drawMatches(im1, kp1, im2, kp2, matches[:20], None, flags=2)
cv2.imshow("res", res)
cv2.waitKey(0)
cv2.destroyAllWindows()
def function6():
img1 = cv2.imread('face1.jpg',0)
img2 = cv2.imread('face2.jpg',0)
orb = cv2.ORB_create()
kp1, des1 = orb.detectAndCompute(img1,None)
kp2, des2 = orb.detectAndCompute(img2,None)
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(des1,des2)
matches = sorted(matches, key = lambda x:x.distance)
img3 = cv2.drawMatches(img1,kp1,img2,kp2,matches[:10], flags=2)
plt.imshow(img3)
plt.show()
def matchImage(self):
img1 = cv2.imread("/home/edgardo/Pictures/CamView/4-processed/canny_1_(7,7):1-90:200.png")
img2 = cv2.imread("/home/edgardo/Pictures/CamView/4-processed/canny_2_(7,7):1-90:200.png")
orb = cv2.ORB_create()
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(des1, des2)
matches = sorted(matches, key = lambda x:x.distance)
img3 = cv2.drawMatches(img1, kp1, img2, kp2, matches[:40], img2, flags = 2)
plt.imshow(img3),plt.show()
def Homography():
MIN_MATCH_COUNT = 10
img1 = cv2.imread('benzene4.jpg',0) # queryImage
img2 = cv2.imread('chemistry1.jpg',0) # trainImage
# Initiate ORB detector
orb = cv2.ORB_create(nfeatures=500, edgeThreshold=5)
# find the keypoints and descriptors with ORB
kp1, des1 = orb.detectAndCompute(img1,None)
kp2, des2 = orb.detectAndCompute(img2,None)
print 'kp1_length:',len(kp1)
print 'kp2_length:',len(kp2)
# FLANN parameters
FLANN_INDEX_LSH = 0
index_params= dict(algorithm = FLANN_INDEX_LSH, table_number = 6, key_size = 12,multi_probe_level = 1)
search_params = dict(checks=50) # or pass empty dictionary
flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.knnMatch(np.asarray(des1,np.float32),np.asarray(des2,np.float32), k=2)
print '匹配数:',len(matches)
good = []
for m,n in matches:
if m.distance < 0.8*n.distance:
good.append(m)
if len(good)>MIN_MATCH_COUNT:
src_pts = np.float32([kp1[m.queryIdx].pt for m in good]).reshape(-1,1,2)
dst_pts = np.float32([kp2[m.trainIdx].pt for m in good]).reshape(-1,1,2)
# cv2.findHomography(srcPoints, dstPoints[, method[, ransacReprojThreshold[, mask]]]) → retval, mask
# ransacReprojThreshold –Maximum allowed reprojection error to treat a point pair as an inlier (used in the RANSAC method only)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
matchesMask = mask.ravel().tolist()
h,w = img1.shape
pts = np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2)
dst = cv2.perspectiveTransform(pts,M)
# cv2.polylines(img, pts, isClosed, color[, thickness[, lineType[, shift]]]) → img
img2 = cv2.polylines(img2,[np.int32(dst)],True, 127, 3, cv2.LINE_AA)
else:
print "Not enough matches are found - %d/%d" % (len(good),MIN_MATCH_COUNT)
matchesMask = None
draw_params = dict( matchColor = (0,255,0), # draw matches in green color
singlePointColor = None,
matchesMask = matchesMask, # draw only inliers
flags = 2)
img3 = cv2.drawMatches(img1,kp1,img2,kp2,good,None,**draw_params)
plt.imshow(img3, 'gray'),plt.show()
def templateSURF(self):
if 'surf' not in globals():
global surf
print("creating surf component")
surf = cv2.xfeatures2d.SURF_create(400)
# find the keypoints and descriptors with SURF
kpT, desT = surf.detectAndCompute(templateImage, None)
kpA, desA = surf.detectAndCompute(activeImage, None)
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5)
search_params = dict(checks=50)
flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.knnMatch(desT, desA, k=2)
# store all the good matches as per Lowe's ratio test
good = []
for m, n in matches:
if m.distance < 0.7*n.distance:
good.append(m)
MIN_MATCH_COUNT = self.ui.sbFeatureMinCnt.value()
print(len(good))
if len(good)>MIN_MATCH_COUNT:
src_pts = np.float32([ kpT[m.queryIdx].pt for m in good ]).reshape(-1,1,2)
dst_pts = np.float32([ kpA[m.trainIdx].pt for m in good ]).reshape(-1,1,2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC,5.0)
matchesMask = mask.ravel().tolist()
h, w = activeImage.shape
pts = np.float32([ [0, 0], [0,h-1], [w-1, h-1], [w-1, 0] ]).reshape(-1,1,2)
dst = cv2.perspectiveTransform(pts, M)
global activeImage
activeImage = activeImage.copy()
activeImage = cv2.polylines(activeImage, [np.int32(dst)], True, 255, 3, cv2.LINE_AA)
else:
print "Not enough matches are found -%d/%d" % (len(good), MIN_MATCH_COUNT)
matchesMask = None
draw_params = dict(matchColor = (0,255, 0), # draw matches in green color
singlePointColor = None,
matchesMask = matchesMask, # draw only inliers
flags = 2)
matchedImage = cv2.drawMatches(templateImage, kpT, activeImage, kpA, good, None, **draw_params)
cv2.imshow("matched image", matchedImage)
def match(img1, img2, MIN_MATCH_COUNT = 10):
surf = cv2.xfeatures2d.SIFT_create()
# find the keypoints and descriptors with SIFT
kp1, des1 = surf.detectAndCompute(img1, None)
kp2, des2 = surf.detectAndCompute(img2, None)
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
search_params = dict(checks = 50)
flann = cv2.FlannBasedMatcher(index_params, search_params)
# bf = cv2.BFMatcher()
matches = flann.knnMatch(des1, des2, k=2)
# matches = bf.knnMatch(des1, des2, k=2)
# store all the good matches as per Lowe's ratio test.
good = []
for m, n in matches:
if m.distance < 0.7 * n.distance:
good.append(m)
print "Good Matched Point:", len(good)
if len(good) > MIN_MATCH_COUNT:
src_pts = np.float32([ kp1[m.queryIdx].pt for m in good ]).reshape(-1, 1, 2)
dst_pts = np.float32([ kp2[m.trainIdx].pt for m in good ]).reshape(-1, 1, 2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
matchesMask = mask.ravel().tolist()
h, w, _ = img1.shape
pts = np.float32([ [0, 0], [0, h-1], [w-1, h-1], [w-1, 0] ]).reshape(-1, 1, 2)
dst = cv2.perspectiveTransform(pts, M)
img2 = cv2.polylines(img2, [np.int32(dst)], True, 255, 3, cv2.LINE_AA)
else:
print "Not enough matches are found - %d/%d" % (len(good), MIN_MATCH_COUNT)
matchesMask = None
return False, ''
draw_params = dict(matchColor = (0, 255, 0), # draw matches in green color
singlePointColor = None,
matchesMask = matchesMask, # draw only inliers
flags = 2)
img3 = cv2.drawMatches(img1, kp1, img2, kp2, good, None, **draw_params)
plt.imshow(img3, 'gray'), plt.show()
dst_img = cv2.warpPerspective(img1, M, (h * 2, w))
# dst_img = cv2.resize(dst_img, None, fx=0.25, fy=0.25, interpolation = cv2.INTER_CUBIC)
cv2.imwrite('/Users/snake/Documents/images/result.jpg',dst_img)
return True, dst_img
def draw_matches(reference_features, unknown_features, mask, good_pts):
fig = plt.figure()
draw_params = dict(matchColor = (0,255,0), # draw matches in green color
singlePointColor = (255,0,0),
matchesMask = mask,
flags = 2)
img3 = cv2.drawMatches(reference_features["image"],
reference_features["kps"],
unknown_features["image"],
unknown_features["kps"],
good_pts,None,**draw_params)
plt.imshow(img3)
return fig
def harris_match(g_first,g_second):
global curr_img
global mpc_pos
h_first = harris_only(g_first)
h_second = harris_only(g_second)
#Match the identical corners in the two given input images
ofrb_var = cv2.ORB_create()
(f_key,f_dscrpt) = ofrb_var.detectAndCompute(h_first, None)
(s_key,s_dscrpt) = ofrb_var.detectAndCompute(h_second, None)
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
mtchs = bf.match(f_dscrpt,s_dscrpt)
mtchs = sorted(mtchs, key = lambda x:x.distance)
curr_img = cv2.drawMatches(h_first,f_key,h_second,s_key,mtchs[:mpc_pos],None,flags=2)
def draw_kps_match(img1, img2):
# Initiate SIFT detector
sift = cv2.xfeatures2d.SURF_create()
# find the key points and descriptors with SIFT
kp1, des1 = sift.detectAndCompute(img1, None)
kp2, des2 = sift.detectAndCompute(img2, None)
bf = cv2.BFMatcher(cv2.NORM_L1, crossCheck=True)
matches = bf.match(des1, des2)
matches_sorted = sorted(matches, key=lambda x: x.distance)
img3 = cv2.drawMatches(img1, kp1, img2, kp2, matches1to2=matches_sorted[:200], outImg=np.ndarray(img1.shape))
return img3
def match(self, img1, img2, draw):
kp1, des1 = self.corner(img1)
kp2, des2 = self.corner(img2)
# if ver2 == 'FLANN':
# # FLANN parameters
# FLANN_INDEX_KDTREE = 0
# index_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5)
# search_params = dict(checks=50) # or pass empty dictionary
#
# flann = cv2.FlannBasedMatcher(index_params, search_params)
#
# matches = flann.knnMatch(des1, des2, k=2)
#
# # Need to draw only good matches, so create a mask
# matchesMask = [[0, 0] for i in xrange(len(matches))]
#
# # ratio test as per Lowe's paper
# for i, (m, n) in enumerate(matches):
# if m.distance < 0.7 * n.distance:
# matchesMask[i] = [1, 0]
#
# draw_params = dict(matchColor=(0, 255, 0),
# singlePointColor=(255, 0, 0),
# matchesMask=matchesMask,
# flags=0)
#
# return cv2.drawMatchesKnn(img1, kp1, img2, kp2, matches[:self.matchPoints], None, **draw_params)
matches = self.bf.knnMatch(des1, des2, k=2)
# matches = sorted(matches, key = lambda x:x.distance)
good = []
for m, n in matches:
if m.distance < 0.6 * n.distance:
good.append(m)
matches = good
list_kp1 = []
list_kp2 = []
for m in good:
img1_idx = m.queryIdx
img2_idx = m.trainIdx
(x1, y1) = kp1[img1_idx].pt
(x2, y2) = kp2[img2_idx].pt
list_kp1.append((x1, y1))
list_kp2.append((x2, y2))
if draw:
return cv2.drawMatches(img1, kp1, img2, kp2, good[:self.matchPoints], None, flags=2)
else:
return list_kp1, list_kp2
def findPatternORB(frame, orb, kp, des, template, preview):
#find keypoints for new frame
kp2, des2 = orb.detectAndCompute(frame, None)
found = False
x, y = 0, 0
#use brute force matching
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(des, des2)
matches = sorted(matches, key = lambda x:x.distance)
#There is a match if there is at least 10 matches and the distance of the
# furthest one is not too far
if len(matches) > 6 and matches[6].distance < 50:
found = True
frame = cv2.drawMatches(template, kp, frame, kp2, matches[:6], outImg=None, flags=2)
return found, (x, y), frame
def sift_compute(file1,file2,result):
global cnt
global selfile
global selcategory
img1 = cv2.imread(file1)
img2 = cv2.imread(file2)
#show_rgb_img(img1);
#plt.show()
img1_gray = to_gray(img1)
img2_gray = to_gray(img2)
#plt.imshow(img1_gray, cmap='gray'),plt.show();
# generate SIFT keypoints and descriptors
img1_kp, img1_desc = gen_sift_features(img1_gray)
img2_kp, img2_desc = gen_sift_features(img2_gray)
#print ('Here are what our SIFT features look like for the image:')
#show_sift_features(img1_gray, img1_front, img1_kp);
#plt.show()
# create a BFMatcher object which will match up the SIFT features
bf = cv2.BFMatcher(cv2.NORM_L2, crossCheck=True)
matches = bf.match(img1_desc, img2_desc)
# Sort the matches in the order of their distance.
matches = sorted(matches, key = lambda x:x.distance)
file = os.path.basename(file2)
parent=os.path.basename(os.path.dirname(file2))
if(len(matches)>= cnt):
cnt = len(matches)
selfile = file
selcategory = parent
print("{}<><><>{}<><><>{}".format(file2,len(matches),parent))
# draw the top N matches
N_MATCHES = 100
match_img = cv2.drawMatches( img1, img1_kp, img2, img2_kp, matches[:N_MATCHES], img2.copy(), flags=0)
#plt.figure(figsize=(12,6))
#plt.imshow(match_img);
#plt.show()
cv2.imwrite(result,match_img)
def Matches(image, Match_image, angle, box, point, CenterX, CenterY):
x = 0
y = 0
while y < len(Match_image):
# cv.imshow("box", Match_image[y])
# cv.imshow("image", image)
# 创建sift特征检测器
sift = cv.xfeatures2d.SIFT_create()
kp1, des1 = sift.detectAndCompute(Match_image[y], None)
kp2, des2 = sift.detectAndCompute(image, None)
# 匹配
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5)
search_params = dict(checks=50)
flann = cv.FlannBasedMatcher(index_params, search_params)
matches = flann.knnMatch(des1, des2, k=2)
goodMatches = []
# 筛选出好的描述子
for m, n in matches:
if m.distance < 0.7 * n.distance:
goodMatches.append(m)
# print('good',len(goodMatches)
# 单独保存 obj 和 scene 好的点位置
obj_pts = []
scene_pts = []
if len(goodMatches) > MIN_MATCH_COUNT:
# 获取关键点的坐标
obj_pts = np.float32([kp1[m.queryIdx].pt
for m in goodMatches]).reshape(-1, 1, 2)
scene_pts = np.float32([kp2[m.trainIdx].pt
for m in goodMatches]).reshape(-1, 1, 2)
#计算变换矩阵和MASK
M, mask = cv.findHomography(obj_pts, scene_pts, cv.RANSAC, 5.0)
matchesMask = mask.ravel().tolist()
h, w = Match_image[y].shape
# pts 為box照片的四個角落座標
pts = np.float32([[0, 0], [0, h], [w, h], [w,
0]]).reshape(-1, 1, 2)
# cv.circle(box, (0,0), 10, (1, 227, 254), -1) #左上
# cv.circle(box, (0,440), 10, (1, 227, 254), -1) #左下
# cv.circle(box, (236,440), 10, (1, 227, 254), -1) #右下
# cv.circle(box, (236,0), 10, (1, 227, 254), -1) #右上
# print("PTS : ")
# print(pts)
# print(pts[0][0][1]) #左上
# print(pts[1][0][1]) #左下
# print(pts[2][0][1]) #右下
# print(pts[3][0][1]) #右上
# 使用得对到的变换矩阵原图像的四个角进行变换，获得在目标图像上对应的坐标
dst = cv.perspectiveTransform(pts, M).reshape(-1, 2)
# print("DST : ")
# print(dst)
# dst[0] 左上
# dst[1] 左下
# dst[2] 右下
# dst[3] 右上
# 求左上->右下 中心點
X_abs = int((dst[0][0] + dst[2][0]) / 2)
Y_abs = int((dst[0][1] + dst[2][1]) / 2)
print("X_abs : ", X_abs)
print("Y_abs : ", Y_abs)
cv.circle(image, (X_abs, Y_abs), 10, (1, 227, 254), -1)
# 求右上->左下 中心點
X_abs_2 = int((dst[1][0] + dst[3][0]) / 2)
Y_abs_2 = int((dst[1][1] + dst[3][1]) / 2)
print("X_abs_2 : ", X_abs_2)
print("Y_abs_2 : ", Y_abs_2)
# 求左上->右下 斜邊
X_side = int((dst[0][0] + dst[2][0]))
Y_side = int((dst[0][1] + dst[2][1]))
print("X_side : ", X_side)
print("Y_side : ", Y_side)
# 求右上->左下 斜邊
X_side_2 = int((dst[1][0] + dst[3][0]))
Y_side_2 = int((dst[1][1] + dst[3][1]))
print("X_side_2 : ", X_side_2)
print("Y_side_2 : ", Y_side_2)
X_side_Less = abs(X_side - X_side_2)
Y_side_Less = abs(Y_side - Y_side_2)
X_center_point_Less = abs(X_abs - X_abs_2)
Y_center_point_Less = abs(Y_abs - Y_abs_2)
print("X_sidet_Less : ", X_side_Less)
print("Y_side_Less : ", Y_side_Less)
print("X_center_point_Less : ", X_center_point_Less)
print("Y_center_point_Less : ", Y_center_point_Less)
# 求角度
X1 = pts[2][0][0] - pts[0][0][0]
Y1 = pts[2][0][1] - pts[0][0][1]
# print(X1,Y1)
X2 = dst[2][0] - dst[0][0]
Y2 = dst[2][1] - dst[0][1]
# print(X2,Y2)
angle1 = np.rad2deg(np.arctan2(Y1, X1))
angle2 = np.rad2deg(np.arctan2(Y2, X2))
# print(angle1,angle2)
angle_diff = angle2 - angle1
print('angle_1', angle1)
print('angle_2', angle2)
print('angle', angle_diff)
print("scene_pts : ", len(scene_pts))
# 加上偏移量
for i in range(4):
dst[i][0] += w
draw_params = dict(singlePointColor=None,
matchesMask=matchesMask,
flags=2)
# result = cv.drawMatches(box, kp1, image, kp2, goodMatches, None)
result = cv.drawMatches(Match_image[y], kp1, image, kp2,
goodMatches, None, **draw_params)
cv.polylines(result, [np.int32(dst)], True, (0, 0, 255), 3,
cv.LINE_AA)
# cv.namedWindow('flann-match', cv.WINDOW_NORMAL) #設置為WINDOW_NORMAL可以任意縮放
# cv.imshow('flann-match',result)
if (X_side >= 50 and X_side <= 2000) and (Y_side >= 50
and Y_side <= 2000):
if ((X_side_Less == 0 and Y_side_Less == 0)
or (X_side_Less <= 6 and Y_side_Less <= 6)):
if ((X_center_point_Less == 0 and Y_center_point_Less == 0)
or (X_center_point_Less <= 5
and Y_center_point_Less <= 5)):
angle.append(angle_diff)
point.append(len(scene_pts))
CenterX.append(X_abs)
CenterY.append(Y_abs)
if (y == 0):
box.append('W')
if (y == 1):
box.append('R')
if (y == 2):
box.append('Y')
if (y == 3):
box.append('G')
if (y == 4):
box.append('WP')
if (y == 5):
box.append('GP')
y = y + 1
# cv.waitKey(0)
# cv.destroyAllWindows()
else:
print("Not enough matches are found - %d/%d" %
(len(goodMatches), MIN_MATCH_COUNT))
# matchesMask = None
y = y + 1
x2 = max(Xs)
y1 = min(Ys)
y2 = max(Ys)
hight = y2 - y1
width = x2 - x1
crop_img = img1[y1:y1 + hight, x1:x1 + width]
cv2.imshow('draw_img', draw_img)
cv2.imshow('crop_img', crop_img)
# 绘制边框
# cv2.polylines(canvas,[np.int32(dst)],True,(0,255,0),3, cv2.LINE_AA)
else:
## (9) Crop the matched region from scene
print("Not enough matches are found - ".format(len(good), MIN_MATCH_COUNT))
## (8) drawMatches
matched = cv2.drawMatches(img1, kpts1, canvas, kpts2, good, None, (0, 255, 0))
#参数2：左上角坐标，参数3：右下角坐标
# cv2.rectangle(matched, (384, 0), (510, 128), (0, 0, 255), 3)
# cv2.rectangle(matched, (590, 197), (885, 490), (0, 0, 255), 3)
# cv2.rectangle(matched, (384, 0), (510, 128), (0, 0, 255), 3)
# 画一个填充红色的圆，参数2：圆心坐标，参数3：半径
# cv2.circle(matched, (678, 890), 63, (255, 0, 0), -1)
#在图中画一个小矩形 最后一个是粗细
#matched = cv2.line(canvas, kpts1, kpts2, (255, 0, 0))
h, w = img1.shape[:2]
#pts = np.float32([[0,0],[0,h-1],[w-1,h-1],[w-1,0]]).reshape(-1,1,2)
#dst = cv2.perspectiveTransform(pts, M)
#perspectiveM = cv2.getPerspectiveTransform(np.float32(dst),pts)
#found = cv2.warpPerspective(img2,perspectiveM,(w,h))
## (10) save and display
def match1():
parser = argparse.ArgumentParser(
description='Code for Feature Matching with FLANN tutorial.')
parser.add_argument('--input1',
help='Path to input image 1.',
default='rec_small1.jpg')
parser.add_argument('--input2',
help='Path to input image 2.',
default='facing1.jpg')
args = parser.parse_args()
img_object = cv.imread(args.input1, cv.IMREAD_GRAYSCALE)
img_scene = cv.imread(args.input2, cv.IMREAD_GRAYSCALE)
if img_object is None or img_scene is None:
print('Could not open or find the images!')
exit(0)
#-- Step 1: Detect the keypoints using SURF Detector, compute the descriptors
minHessian = 400
detector = cv.xfeatures2d_SURF.create(hessianThreshold=minHessian)
keypoints_obj, descriptors_obj = detector.detectAndCompute(
img_object, None)
keypoints_scene, descriptors_scene = detector.detectAndCompute(
img_scene, None)
#-- Step 2: Matching descriptor vectors with a FLANN based matcher
# Since SURF is a floating-point descriptor NORM_L2 is used
matcher = cv.DescriptorMatcher_create(cv.DescriptorMatcher_FLANNBASED)
knn_matches = matcher.knnMatch(descriptors_obj, descriptors_scene, 2)
#-- Filter matches using the Lowe's ratio test
ratio_thresh = 0.75
good_matches = []
for m, n in knn_matches:
if m.distance < ratio_thresh * n.distance:
good_matches.append(m)
#-- Draw matches
img_matches = np.empty(
(max(img_object.shape[0],
img_scene.shape[0]), img_object.shape[1] + img_scene.shape[1], 3),
dtype=np.uint8)
cv.drawMatches(img_object,
keypoints_obj,
img_scene,
keypoints_scene,
good_matches,
img_matches,
flags=cv.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)
#-- Localize the object
obj = np.empty((len(good_matches), 2), dtype=np.float32)
scene = np.empty((len(good_matches), 2), dtype=np.float32)
for i in range(len(good_matches)):
#-- Get the keypoints from the good matches
obj[i, 0] = keypoints_obj[good_matches[i].queryIdx].pt[0]
obj[i, 1] = keypoints_obj[good_matches[i].queryIdx].pt[1]
scene[i, 0] = keypoints_scene[good_matches[i].trainIdx].pt[0]
scene[i, 1] = keypoints_scene[good_matches[i].trainIdx].pt[1]
H, _ = cv.findHomography(obj, scene, cv.RANSAC)
#-- Get the corners from the image_1 ( the object to be "detected" )
obj_corners = np.empty((4, 1, 2), dtype=np.float32)
obj_corners[0, 0, 0] = 0
obj_corners[0, 0, 1] = 0
obj_corners[1, 0, 0] = img_object.shape[1]
obj_corners[1, 0, 1] = 0
obj_corners[2, 0, 0] = img_object.shape[1]
obj_corners[2, 0, 1] = img_object.shape[0]
obj_corners[3, 0, 0] = 0
obj_corners[3, 0, 1] = img_object.shape[0]
scene_corners = cv.perspectiveTransform(obj_corners, H)
#-- Draw lines between the corners (the mapped object in the scene - image_2 )
cv.line(img_matches, (int(scene_corners[0,0,0] + img_object.shape[1]), int(scene_corners[0,0,1])),\
(int(scene_corners[1,0,0] + img_object.shape[1]), int(scene_corners[1,0,1])), (0,255,0), 4)
cv.line(img_matches, (int(scene_corners[1,0,0] + img_object.shape[1]), int(scene_corners[1,0,1])),\
(int(scene_corners[2,0,0] + img_object.shape[1]), int(scene_corners[2,0,1])), (0,255,0), 4)
cv.line(img_matches, (int(scene_corners[2,0,0] + img_object.shape[1]), int(scene_corners[2,0,1])),\
(int(scene_corners[3,0,0] + img_object.shape[1]), int(scene_corners[3,0,1])), (0,255,0), 4)
cv.line(img_matches, (int(scene_corners[3,0,0] + img_object.shape[1]), int(scene_corners[3,0,1])),\
(int(scene_corners[0,0,0] + img_object.shape[1]), int(scene_corners[0,0,1])), (0,255,0), 4)
#-- Show detected matches
cv.imshow('Good Matches & Object detection', img_matches)
# cv.imwrite('Good Matches&Object detection1.jpg',img_matches)
# cv.imshow('Good Matches&Object detection1.jpg',img_matches)
# print(int(scene_corners[0,0,0]),img_object.shape[1],int(scene_corners[0,0,1]))
# print('Good Matches&Object detection1'+str(c) + '.jpg')
center_point0 = (int(scene_corners[0, 0, 0] + img_object.shape[1]) +
int(scene_corners[2, 0, 0] + img_object.shape[1])) / 2
center_point1 = (int(scene_corners[0, 0, 1]) +
int(scene_corners[2, 0, 1])) / 2
center_point = [center_point0, center_point1]
# return(center_point)
print(center_point)
while (True):
# Press Q on keyboard to exit
if cv.waitKey(25) & 0xFF == ord('q'):
break
def plot_matches(self,
image1,
keypoints1,
image2,
keypoints2,
matches,
color_matches,
color_keypoints,
mask=None,
color_outliers=None):
"""!
Draws the given keypoints and matches onto the given images. The resulting image is of width image1.width + image2.width
An optional mask determines what keypoints to draw and if color_outliers is set additionaly the mask will be treated as inliers and outliers and will be drawn in different colors.
@param image1 The first image
@param keypoints1 The keypoints of the first image
@param image2 The second image
@param keypoints2 The keypoints of the second image.
@param matches The matches to plot
@param color_matches: Color Triple (blue, green, red) with values from 0 to 255. Matches will be drawn in this color.
@param color_keypoints Color Triple (blue, green, red) with values from 0 to 255. Keypoints will be drawn in this color.
@param mask optional paremeter. Mask determining which matches are drawn. Has to be of same size as matches.
@param color_outliers optional parameter. Color Triple (blue, green, red) with values from 0 to 255. Requires mask to be set as well. The masks values will be handled as inliers and will be drawn in color_matches and the outliers will be drawn in color_outliers.
@return image_with_matches: returns image of size image1.width + image2.width with keypoints and matches drawn according to parameters
"""
if image1 is None:
raise ValueError('Image1 is not a valid image.')
if image2 is None:
raise ValueError('Image2 is not a valid image.')
if (not isinstance(color_matches, tuple)
) or (len(color_matches) != 3) or (not all(
isinstance(x, int)
for x in color_matches)) or (not all(x >= 0 and x <= 255
for x in color_matches)):
raise ValueError(
'Color_matches needs to be of format (b, g, r) with integer values from 0 to 255.'
)
if (not isinstance(color_keypoints,
tuple)) or (len(color_keypoints) != 3) or (not all(
isinstance(x, int)
for x in color_keypoints)) or (not all(
x >= 0 and x <= 255
for x in color_keypoints)):
raise ValueError(
'Color_keypoints needs to be of format (b, g, r) with integer values from 0 to 255.'
)
if mask is None and color_outliers is not None:
raise ValueError(
'If color_outliers is set, mask needs to be set as well.')
# Plot all matches
if mask is None and color_outliers is None:
return cv2.drawMatches(image1, keypoints1, image2, keypoints2,
matches, None, color_matches,
color_keypoints)
if len(mask) != len(matches):
raise ValueError('Mask and matches need to be of same length.')
if mask is not None and color_outliers is None:
return cv2.drawMatches(image1, keypoints1, image2, keypoints2,
matches, None, color_matches,
color_keypoints, mask)
if (not isinstance(color_outliers,
tuple)) or (len(color_outliers) != 3) or (not all(
isinstance(x, int)
for x in color_outliers)) or (not all(
x >= 0 and x <= 255
for x in color_outliers)):
raise ValueError(
'Color_outliers needs to be of format (b, g, r) with integer values from 0 to 255.'
)
# Plot inliers and outliers in different colors
inliers = mask.ravel().tolist()
outliers = inliers.copy()
for i in range(len(outliers)):
if outliers[i] == 0:
outliers[i] = 1
else:
outliers[i] = 0
out = cv2.drawMatches(image1, keypoints1, image2, keypoints2, matches,
None, color_matches, color_keypoints, inliers)
return cv2.drawMatches(image1, keypoints1, image2, keypoints2, matches,
out, color_outliers, color_keypoints, outliers,
1)
# Match descriptors.
matches20 = bf.match(des20, des)
matches50 = bf.match(des50, des)
matches100 = bf.match(des100, des)
matches_puste = bf.match(des_puste, des)
# Sort them in the order of their distance.
matches20 = sorted(matches20, key=lambda x: x.distance)
matches50 = sorted(matches50, key=lambda x: x.distance)
matches100 = sorted(matches100, key=lambda x: x.distance)
matches_puste = sorted(matches_puste, key=lambda x: x.distance)
# Draw matches.
if len(matches20) > len(matches50) and len(matches20) > len(matches100):
img3 = cv2.drawMatches(img20, kp20, img, kp, matches20, None, flags=2)
print("*** WYKRYTO 20ZŁ ***")
elif len(matches50) > len(matches20) and len(matches50) > len(matches100):
img3 = cv2.drawMatches(img50, kp50, img, kp, matches50, None, flags=2)
print("*** WYKRYTO 50ZŁ ***")
elif len(matches100) > len(matches20) and len(matches100) > len(matches50):
img3 = cv2.drawMatches(img100, kp100, img, kp, matches100, None, flags=2)
print("*** WYKRYTO 100ZŁ ***")
else:
img3 = cv2.drawMatches(img_puste, kp_puste, img, kp, matches_puste, None, flags=2)
print("*** NIE WYKRYTO NIC ***")
while True:
cv2.imshow('frame', img3)
if cv2.waitKey(1) & 0xFF == ord('q'):
def draw_key_sift(path1, path2):
# bf = cv2.BFMatcher()
# match = bf.knnMatch(des1, des2, k=2)
# match2 = bf.knnMatch(des1, des3, k=2)
# good = []
# good2 = []
# for m, n in match:
# if m.distance < 0.75 * n.distance:
# good.append([m])
#
# for m, n in match2:
# if m.distance < 0.75 * n.distance:
# good2.append([m])
# print(len(good))
# print(len(good2))
# # print("n: ", match.n, "\n m", match.m)
# img4 = cv2.drawMatchesKnn(img1, key1, img2, key2, good, None, flags=2)
# img5 = cv2.drawMatchesKnn(img1, key1, img3, key3, good, None, flags=2)
# FLANN parameters
img1 = cv2.imread(path1)
img2 = cv2.imread(path2)
sift = cv2.xfeatures2d.SIFT_create()
key1, des1 = sift.detectAndCompute(img1, None)
key2, des2 = sift.detectAndCompute(img2, None)
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5)
search_params = dict(checks=50) # or pass empty dictionary
flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.knnMatch(des1, des2, k=2)
# Need to draw only good matches, so create a mask
# matchesMask1 = [[0, 0] for i in range(len(matches1))]
# matchesMask2 = [[0, 0] for i in range(len(matches2))]
# ratio test as per Lowe's paper
# for i, (m, n) in enumerate(matches1):
# if m.distance < 0.7 * n.distance:
# matchesMask1[i] = [1, 0]
#
# for i, (m, n) in enumerate(matches2):
# if m.distance < 0.7 * n.distance:
# matchesMask2[i] = [1, 0]
#
# draw_params1 = dict(matchColor=(0, 255, 0),
# singlePointColor=(255, 0, 0),
# matchesMask=matchesMask1,
# flags=0)
#
# draw_params2 = dict(matchColor=(0, 255, 0),
# singlePointColor=(255, 0, 0),
# matchesMask=matchesMask2,
# flags=0)
# if len(matches1) >= len(matches2):
# img6 = cv2.drawMatchesKnn(img1, key1, img2, key2, matches1, None, **draw_params1)
# cv2.imshow("Image", img6)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
# else:
good = []
for m, n in matches:
if m.distance < 0.7 * n.distance:
good.append(m)
MIN_MATCH_COUNT = 10
if len(good) > MIN_MATCH_COUNT:
src_pts = np.float32([key1[m.queryIdx].pt
for m in good]).reshape(-1, 1, 2)
dst_pts = np.float32([key2[m.trainIdx].pt
for m in good]).reshape(-1, 1, 2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
matchesMask = mask.ravel().tolist()
h, w, d = img1.shape
pts = np.float32([[0, 0], [0, h - 1], [w - 1, h - 1],
[w - 1, 0]]).reshape(-1, 1, 2)
dst = cv2.perspectiveTransform(pts, M)
img2 = cv2.polylines(img2, [np.int32(dst)], True, 255, 3, cv2.LINE_AA)
else:
matchesMask = None
draw_params = dict(
matchColor=(0, 255, 0), # draw matches in green color
singlePointColor=None,
matchesMask=matchesMask, # draw only inliers
flags=2)
matching_result = cv2.drawMatches(img1, key1, img2, key2, good, None,
**draw_params)
img = Image.fromarray(cv2.cvtColor(matching_result, cv2.COLOR_BGR2RGB))
return img
matches = bf.match(des1, des2)
good = sorted(matches, key=lambda x: x.distance)
if len(good) > MIN_MATCH_COUNT:
src_pts = np.float32([kp1[m.queryIdx].pt for m in good]).reshape(-1, 1, 2)
dst_pts = np.float32([kp2[m.trainIdx].pt for m in good]).reshape(-1, 1, 2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
if np.shape(M) == ():
print("No transformation possible")
#return None, None
## derive rotation angle from homography
theta = -math.atan2(M[0, 1], M[0, 0]) * 180 / math.pi
total_move = src_pts[0][0] - dst_pts[0][0]
Horizontal = total_move[0]
Vertical = total_move[1]
print "Displacement in X:", Horizontal
print "Displacement in Y:", Vertical
print "Angle detected:", abs(theta)
img3 = cv2.drawMatches(img1, kp1, img2, kp2, matches[1:10], None, flags=2)
cv2.imwrite("output.jpg", img3)
plt.figure(5)
plt.imshow((cv2.drawKeypoints(train_img_gray, train_kp, train_img.copy())))
plt.title('Train Image Keypoints')
plt.figure(6)
plt.imshow((cv2.drawKeypoints(query_img_gray, query_kp, query_img.copy())))
plt.title('Query Image Keypoints')
# create a BFMatcher object which will match up the SIFT features
bf = cv2.BFMatcher(cv2.NORM_L2, crossCheck=True)
matches = bf.match(train_desc, query_desc)
# Sort the matches in the order of their distance.
matches = sorted(matches, key=lambda x: x.distance)
# draw the top N matches
N_MATCHES = 100
match_img = cv2.drawMatches(train_img,
train_kp,
query_img,
query_kp,
matches[:N_MATCHES],
query_img.copy(),
flags=0)
plt.figure(7)
plt.imshow(match_img)
plt.title('SIFT Detection')
plt.show()
[kp1[matches[i].queryIdx].pt for i in range(len(matches))])
right_point = np.asarray(
[kp2[matches[i].trainIdx].pt for i in range(len(matches))])
left_point[:, 0] += xg
left_point[:, 1] += yg
right_point[:, 0] += xd
right_point[:, 1] += yd
while len(left_point) < nbr:
left_point = np.append(left_point, [[1e17, 1e17]], axis=0)
right_point = np.append(right_point, [[1e17, 1e17]],
axis=0)
img = cv2.drawMatches(frameg[yg:yg + hg, xg:xg + wg],
kp1,
framed[yd:yd + hd, xd:xd + wd],
kp2,
matches,
None,
flags=2)
cv2.namedWindow("ok", cv2.WINDOW_NORMAL)
cv2.imshow("ok", img)
key = cv2.waitKey(1)
if key == 'q':
break
if count > 0:
trajectory_camera_coord_gauche = np.append(
trajectory_camera_coord_gauche, left_point, axis=0)
trajectory_camera_coord_droite = np.append(
trajectory_camera_coord_droite, right_point, axis=0)
if count == 0:
print('good matches:%d/%d' % (len(good_matches), len(matches)))
# 좋은 매칭점의 queryIdx로 원본 영상의 좌표 구하기
src_pts = np.float32([kp1[m.queryIdx].pt for m in good_matches])
# 좋은 매칭점의 trainIdx로 대상 영상의 좌표 구하기
dst_pts = np.float32([kp2[m.trainIdx].pt for m in good_matches])
# findHomography() 함수로 원근 변환 행렬 구하기
mtrx, mask = cv2.findHomography(src_pts, dst_pts)
# 원본 영상 크기로 변환 영역 좌표 생성
h, w, = img1.shape[:2]
pts = np.float32([[[0, 0]], [[0, h - 1]], [[w - 1, h - 1]], [[w - 1, 0]]])
# perspectiveTransform() 함수로 원본 영상 좌표를 원근 변환시킴
dst = cv2.perspectiveTransform(pts, mtrx)
# 변환 좌표 영역을 대상 영상에 그리기
img2 = cv2.polylines(img2, [np.int32(dst)], True, 255, 3, cv2.LINE_AA)
# 좋은 매칭 그려서 출력
res = cv2.drawMatches(img1,
kp1,
img2,
kp2,
good_matches,
None,
flags=cv2.DRAW_MATCHES_FLAGS_NOT_DRAW_SINGLE_POINTS)
cv2.imshow('Matching Homography', res)
cv2.waitKey()
cv2.destroyAllWindows()
feature = cv2.KAZE_create()
#feature = cv2.AKAZE_create()
#feature = cv2.ORB_create()
# 특징점 검출 및 기술자 계산
kp1, desc1 = feature.detectAndCompute(src1, None)
kp2, desc2 = feature.detectAndCompute(src2, None)
# 특징점 매칭
matcher = cv2.BFMatcher_create()
#matcher = cv2.BFMatcher_create(cv2.NORM_HAMMING)
matches = matcher.knnMatch(desc1, desc2, 2)
# 좋은 매칭 결과 선별
good_matches = []
for m in matches:
if m[0].distance / m[1].distance < 0.7:
good_matches.append(m[0])
print('# of kp1:', len(kp1))
print('# of kp2:', len(kp2))
print('# of matches:', len(matches))
print('# of good_matches:', len(good_matches))
# 특징점 매칭 결과 영상 생성
dst = cv2.drawMatches(src1, kp1, src2, kp2, good_matches, None)
cv2.imshow('dst', dst)
cv2.waitKey()
cv2.destroyAllWindows()
def detectRDT(self, img, cnt=5):
print('[INFO] start detectRDT')
startTime = time.time()
height, width = img.shape
p1 = (0, int(height * (1 - (VIEW_FINDER_SCALE_W) / CROP_RATIO) / 2))
p2 = (int(width - p1[0]), int(height - p1[1]))
#"""
# ==================
# TODO: this mask is still HARDCODEDDDDDD!!!!!!
# =================
#"""
#p1 = (int(width * (1 - VIEW_FINDER_SCALE_W)/2 * CROP_RATIO) , 0)
#p2 = (int(width - p1[0]), int(height - p1[1] - 65))
roi = img[p1[1]:p2[1], p1[0]:p2[0]]
# img = roi
mask = np.zeros((height, width), np.uint8)
mask[p1[1]:p2[1], p1[0]:p2[0]] = 255
# show_image(img)
# show_image(mask)
# keypoints, descriptors = self.siftDetector.detectAndCompute(img, None)
# show_image(mask)
keypoints, descriptors = self.siftDetector.detectAndCompute(img, mask)
print('[INFO] detect/compute time: ', time.time() - startTime)
print('[INFO] descriptors')
print(descriptors)
# TODO: Find condition for this
# if (descriptors == None or all(descriptors)):
# print('[WARNING] No Features on input')
# return None
# Matching
if descriptors is None:
return None
matches = self.matcher.knnMatch(self.refSiftDescriptors,
descriptors,
k=2)
print('[INFO] Finish matching')
print('matches', matches)
# Apply ratio test
good = []
if matches is None or len(matches) == 0 or len(matches[0]) < 2:
return None
for m, n in matches:
if m.distance < 0.80 * n.distance:
good.append(m)
matchingImage = cv.drawMatches(self.fluRefImg,
self.refSiftKeyPoints,
img,
keypoints,
good,
None,
flags=2)
# plt.imshow(matchingImage)
# plt.title('SIFT Brute Force matching')
# plt.show()
sum = 0
distance = 0
count = 0
# store all the good matches as per Lowe's ratio test.
img2 = None
dst = None
print('[INFO] matches')
if len(good) > MIN_MATCH_COUNT:
src_pts = np.float32([
self.refSiftKeyPoints[m.queryIdx].pt for m in good
]).reshape(-1, 1, 2)
dst_pts = np.float32([keypoints[m.trainIdx].pt
for m in good]).reshape(-1, 1, 2)
# print('src_pts', src_pts)
# print('dst_pst', dst_pts)
M, mask = cv.findHomography(src_pts, dst_pts, cv.RANSAC, cnt)
print('[INFO] Finish finding Homography')
# print('[INFO] M Matrix', M)
# print('[INFO] mask', mask)
matchesMask = mask.ravel().tolist()
h, w = self.fluRefImg.shape
pts = np.float32([[0, 0], [w - 1, 0], [w - 1, h - 1],
[0, h - 1]]).reshape(-1, 1, 2)
if M is None or M.size == 0:
return None
dst = cv.perspectiveTransform(pts, M)
print('[INFO] dst transformation pts', dst)
img2 = np.copy(img)
img2 = cv.polylines(img2, [np.int32(dst)], True, (255, 0, 0))
pts_box = cv.minAreaRect(dst)
box = cv.boxPoints(pts_box) # cv2.boxPoints(rect) for OpenCV 3.x
box = np.int0(box)
cv.drawContours(img2, [box], 0, (0, 0, 255), 2)
print('[INFO] finish perspective transform')
print(box)
# show_image(roi)
# show_image(img2)
# img2=None
new_dst = np.copy(dst)
#for i in list(range(0, 4)):
# min_dist = 99999999
# min_j = -1
# for j in list(range(0, 4)):
# print(box[i], dst[j])
# dist = pow(box[i][0]-dst[j][0][0], 2) + pow(box[i][1]-dst[j][0][1], 2)
# if dist < min_dist:
# print('---min', dist, j, box[i], new_dst[j])
# min_dist = dist
# min_j = j
# new_dst[min_j][0][0] = box[i][0]
# new_dst[min_j][0][1] = box[i][1]
for i in list(range(0, 4)):
if pts_box[2] < -45:
new_dst[(i + 2) % 4] = [box[i]]
else:
new_dst[(i + 3) % 4] = [box[i]]
dst = np.copy(new_dst)
print(dst)
else:
print("Not enough matches are found - {}/{}".format(
len(good), MIN_MATCH_COUNT))
matchesMask = None
return None
draw_params = dict(
matchColor=(255, 0, 0), # draw matches in green color
singlePointColor=None,
matchesMask=matchesMask, # draw only inliers
flags=2)
img3 = cv.drawMatches(self.fluRefImg, self.refSiftKeyPoints, img2,
keypoints, good, None, **draw_params)
#plt.imshow(img3, 'gray'),plt.show()
# show_image(img3)
h, w = self.fluRefImg.shape
#refBoundary = np.float32([ [0,0], [0,h-1],[w-1,h-1],[w-1,0]]).reshape(-1,1,2)
refBoundary = np.float32([[0, 0], [w - 1, 0], [w - 1, h - 1],
[0, h - 1]]).reshape(-1, 1, 2)
print('Refboundary', refBoundary)
print('Boundary', dst)
M = cv.getPerspectiveTransform(dst, refBoundary)
print('M matrixxxx', M)
transformedImage = cv.warpPerspective(
img, M, (self.fluRefImg.shape[1], self.fluRefImg.shape[0]))
# show_image(roi)
# show_image(transformedImage)
return dst
#gftt match nur pro Ecke (also in unmittelbarer Umgebung) nur einmal während Harris das nicht tut.
#2
gftt_img = cv2.goodFeaturesToTrack(blur_img,20,0.1,1)
show_result(img_left,img_right,gftt_img,True,path="out/goodFeaturesToTrack.png")
#3
orb = cv2.ORB_create(3000)
print(orb)
kp1, des1 = orb.detectAndCompute(img_left,None)
kp2, des2 = orb.detectAndCompute(img_right,None)
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(img_left,img_right)
matches = sorted(matches, key = lambda x:x.distance)
orb_img = np.hstack((img_left,img_right))
kp3, des3 = orb.detectAndCompute(orb_img,None)
orb_img = cv2.drawMatches(img_left,kp1,img_right,kp2,matches,orb_img, flags=2)
orb_img = cv2.drawKeypoints(orb_img,kp3,orb_img,color=(0,0,255))
cv2.namedWindow("img",cv2.WINDOW_NORMAL)
cv2.imwrite("out/orb.png",orb_img)
cv2.imshow("img",orb_img)
cv2.waitKey(0)
# 0, 255, 0), flags=0)
# kp2img = cv2.drawKeypoints(img2, kp2, None, color=(
# 0, 255, 0), flags=0)
# cv2.imshow('img1kp', kp1img)
# cv2.imshow('img2kp', kp2img)
# match 两张图片的 `keypoints`
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(des1, des2)
matches = sorted(matches, key=lambda x: x.distance)
# 画出20对匹配的 `keypoints`
matches_img = cv2.drawMatches(
img1,
kp1,
img2,
kp2,
matches[:20],
None,
flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)
# 执行 `RANSAC` 算法，计算出变换矩阵 `H`
A = np.array([kp1[m.queryIdx].pt for m in matches])
B = np.array([kp2[m.trainIdx].pt for m in matches])
H = ransacMatching(A, B)
# H = ransacMatching2(A, B)
# 画出原始图片的边框(红色)，以及该边框经过 `H` 变换得到的边框(蓝色)
rows, cols, _ = img1.shape
pts = np.array(
[[0, 0], [0, rows - 1], [cols - 1, rows - 1], [cols - 1, 0]],
dtype=np.float32).reshape(-1, 1, 2)
kp1, des1 = orb.detectAndCompute(img1_01, None)
kp2, des2 = orb.detectAndCompute(img1_02, None)
bf = cv2.BFMatcher.create()
matches = bf.knnMatch(des1, des2, k=2)
#matches = sorted(matches,key=lambda x:x.distance)。
## 调试寻找最优的匹配点。
goodPoints = []
for m, n in matches:
print(m.queryIdx, m.trainIdx)
if m.distance < 0.65 * n.distance:
goodPoints.append(m)
print("最优的匹配点的个数：", len(goodPoints))
draw_Params = dict(matchColor=(0, 255, 0), singlePointColor=None, flags=2)
img3 = cv2.drawMatches(img1_01, kp1, img1_02, kp2, goodPoints, None,
**draw_Params)
MIN_MATCH_COUNT = 10
if len(goodPoints) > MIN_MATCH_COUNT:
src_pts = np.float32([kp1[m.queryIdx].pt
for m in goodPoints]).reshape(-1, 1, 2)
dst_pts = np.float32([kp2[m.trainIdx].pt
for m in goodPoints]).reshape(-1, 1, 2)
M, mask = cv2.findHomography(dst_pts, src_pts, cv2.RANSAC, 5.0)
h, w = img1_gray.shape
pts = np.float32([[0, 0], [0, h - 1], [w - 1, h - 1],
[w - 1, 0]]).reshape(-1, 1, 2)
dst = cv2.perspectiveTransform(pts, M)
img2_gray = cv2.polylines(img1_gray, [np.int32(dst)], True, 255, 3,
def alignImages(im1, im2, meth='ORB'):
input_img1 = im1.copy()
input_img2 = im2.copy()
# input_img1 = cv2.cvtColor(input_img1, cv2.COLOR_BGR2GRAY)
# input_img2 = cv2.cvtColor(input_img2, cv2.COLOR_BGR2GRAY)
if TF == '/TRUE/':
cross = True
else:
cross = False
print(meth)
if meth == 'ORB':
# Detect ORB features and compute descriptors.
detector = cv2.ORB_create(MAX_FEATURES)
matcher = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=cross)
# Convert images to grayscale
elif meth == 'SURF':
detector = cv2.xfeatures2d.SURF_create(MAX_FEATURES)
matcher = cv2.BFMatcher(cv2.NORM_L2, crossCheck=cross)
elif meth == 'SIFT':
detector = cv2.xfeatures2d.SIFT_create(MAX_FEATURES)
matcher = cv2.BFMatcher(cv2.NORM_L2, crossCheck=cross)
keypoints1, descriptors1 = detector.detectAndCompute(input_img1, None)
keypoints2, descriptors2 = detector.detectAndCompute(input_img2, None)
# Match features.
# create BFMatcher object
# matcher = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
# matcher = cv2.DescriptorMatcher_create(cv2.DESCRIPTOR_MATCHER_BRUTEFORCE_HAMMING)
matches = matcher.match(descriptors1, descriptors2, None)
# Sort matches by score
matches.sort(key=lambda x: x.distance, reverse=False)
# Remove not so good matches
numGoodMatches = int(len(matches) * GOOD_MATCH_PERCENT)
matches = matches[:numGoodMatches]
# print(type(keypoints1))
# Extract location of good matches
points1 = np.zeros((len(matches), 2), dtype=np.float32)
points2 = np.zeros((len(matches), 2), dtype=np.float32)
new_matches = []
for i, match in enumerate(matches):
if keypoints1[match.queryIdx].pt[0] > 360 and keypoints1[
match.queryIdx].pt[0] < 720 and keypoints1[match.queryIdx].pt[
1] > 270 and keypoints1[match.queryIdx].pt[1] < 810:
if keypoints2[match.queryIdx].pt[0] > 360 and keypoints2[
match.queryIdx].pt[0] < 720 and keypoints2[
match.queryIdx].pt[1] > 270 and keypoints2[
match.queryIdx].pt[1] < 810:
new_matches.append(match)
points1[i, :] = keypoints1[match.queryIdx].pt
points2[i, :] = keypoints2[match.trainIdx].pt
# Draw top matches
imMatches = cv2.drawMatches(im1, keypoints1, im2, keypoints2, new_matches,
None)
# cv2.imwrite("matches.jpg", imMatches)
# Find homography
h, mask = cv2.findHomography(points1, points2, cv2.RANSAC)
# Use homography
height, width, channels = im2.shape
im1Reg = cv2.warpPerspective(im1, h, (width, height))
return im1Reg, h, imMatches, matches, keypoints1, keypoints2
img = cv2.imread(filename)
match = cv2.imread(matchfile)
img=cv2.resize(img, (0, 0), fx=0.1, fy=0.1)
match=cv2.resize(match,(0,0),fx=0.1,fy=0.1)
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
matchgray=cv2.cvtColor(match,cv2.COLOR_BGR2GRAY)
bfmatcher = cv2.BFMatcher_create(cv2.NORM_L2,crossCheck=True)
sift = cv2.xfeatures2d.SIFT_create()
corners = cv2.cornerHarris(gray,2,0,0.04)
kpsCorners = np.argwhere(corners>0.1*corners.max())
kpsCorners = [cv2.KeyPoint(pt[1],pt[0],3) for pt in kpsCorners]
grayWithCorners = cv2.drawKeypoints(gray,kpsCorners,None,flags=cv2.DRAW_MATCHES_FLAGS_DEFAULT)
kpsCorners,dscCorners = sift.compute(gray,kpsCorners)
kp = sift.detect(gray,None)
grayWithSift =cv2.drawKeypoints(gray,kp,None,flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
kp,dsc =sift.compute(gray,kp)
corners = cv2.cornerHarris(matchgray,2,0,0.04)
kpsCorners2 = np.argwhere(corners>0.1*corners.max())
kpsCorners2 = [cv2.KeyPoint(pt[1],pt[0],3) for pt in kpsCorners2]
#grayWithCorners = cv2.drawKeypoints(gray,kpsCorners,None,flags=cv2.DRAW_MATCHES_FLAGS_DEFAULT)
kpsCorners2,dscCorners2 = sift.compute(matchgray,kpsCorners2)
matchesCorners = bfmatcher.match(dscCorners,dscCorners2)
matchesCorners = sorted(matchesCorners,key=lambda x:x.distance)
img3 = cv2.drawMatches(img,kpsCorners,match,kpsCorners2,matchesCorners[:10],None,flags=2)
cv2.imshow("match",img3)
if cv2.waitKey(0) & 0xff == 27:
cv2.destroyAllWindows()
for m, n in matches:
if m.distance < 0.7 * n.distance:
good.append(m)
if len(good) > MIN_MATCH_COUNT:
src_pts = np.float32([kp1[m.queryIdx].pt for m in good]).reshape(-1, 1, 2)
dst_pts = np.float32([kp2[m.trainIdx].pt for m in good]).reshape(-1, 1, 2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
matchesMask = mask.ravel().tolist()
h, w = img1.shape
pts = np.float32([[0, 0], [0, h - 1], [w - 1, h - 1],
[w - 1, 0]]).reshape(-1, 1, 2)
dst = cv2.perspectiveTransform(pts, M)
img2 = cv2.polylines(img2, [np.int32(dst)], True, 255, 3, cv2.LINE_AA)
else:
print("Not enough matches are found - %d/%d" % len(good), MIN_MATCH_COUNT)
matchesMask = None
draw_params = dict(
matchColor=(0, 255, 0), # draw matches in green color
singlePointColor=None,
matchesMask=matchesMask, # draw only inliers
flags=2)
img3 = cv2.drawMatches(img1, kp1, img2, kp2, good, None, **draw_params)
plt.imshow(img3, 'gray'), plt.show()
matches = []
# loop over the raw matches
for m,n in rawMatches:
# We ensure the distance is within a certain ratio of each other (i.e. Lowe's ratio test)
if m.distance < n.distance * ratio:
matches.append(m)
return matches
print("Using: {} feature matcher".format(feature_matching))
fig = plt.figure(figsize=(20,8))
if feature_matching == 'bf':
matches = matchKeyPointsBF(featuresA, featuresB, method=feature_extractor)
img3 = cv2.drawMatches(img2,kpsA,img1,kpsB,matches[:100],
None,flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)
elif feature_matching == 'knn':
matches = matchKeyPointsKNN(featuresA, featuresB, ratio=0.75, method=feature_extractor)
img3 = cv2.drawMatches(img2,kpsA,img1,kpsB,np.random.choice(matches,100),
None,flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)
plt.imshow(img3)
plt.show()
def getHomography(kpsA, kpsB, featuresA, featuresB, matches, reprojThresh):
# convert the keypoints to numpy arrays
kpsA = np.float32([kp.pt for kp in kpsA])
kpsB = np.float32([kp.pt for kp in kpsB])
if len(matches) > 4:
def matchAndImageCut(self, sift, origin, ori_kp, ori_des, typeName, featureConfig, imageConfig, job_id):
# TODO: check file exists
img_template = cv2.imread(featureConfig['file'], cv2.IMREAD_GRAYSCALE)
img_detect = origin.copy()
min_match_count = featureConfig['option'].get('minMatchCount', 50)
distance_threshold = featureConfig['option'].get('matchDistance', 0.5)
tpl_kp, tpl_des = sift.detectAndCompute(img_template, None)
index_params = dict(algorithm=0, trees=5) # algorithm = FLANN_INDEX_KDTREE
search_params = dict(checks=50)
flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.knnMatch(tpl_des, ori_des, k=2)
# store all the good matches as per Lowe's ratio test.
good = []
for m, n in matches:
if m.distance < distance_threshold * n.distance:
good.append(m)
logging.info("Feature [%s] matches %s, min=%s, threshold=%.2f, good=%s" % (
typeName, len(matches), min_match_count, distance_threshold, len(good)))
if len(good) > min_match_count:
src_pts = np.float32([tpl_kp[m.queryIdx].pt for m in good]).reshape(-1, 1, 2)
dst_pts = np.float32([ori_kp[m.trainIdx].pt for m in good]).reshape(-1, 1, 2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC, 5.0)
# draw feature polyline in origin image
pts = np.float32([[0, 0], [0, img_template.shape[0] - 1],
[img_template.shape[1] - 1, img_template.shape[0] - 1],
[img_template.shape[1] - 1, 0]]).reshape(-1, 1, 2)
dst = cv2.perspectiveTransform(pts, M)
cv2.polylines(img_detect, [np.int32(dst)], True, 0, 1)
# draw detected image
matchesMask = mask.ravel().tolist()
draw_params = dict(matchColor=(0, 255, 0), # draw matches in green color
singlePointColor=None,
matchesMask=matchesMask, # draw only inliers
flags=2)
draw_img = cv2.drawMatches(img_template, tpl_kp, origin, ori_kp, good, None, **draw_params)
tools.writeImageJob(draw_img, job_id + '/step1', 'draw matching %s' % typeName)
# draw normalize image's polyline in origin image
normalized_pts = np.float32([
[-1 * featureConfig['x'], -1 * featureConfig['y']],
[-1 * featureConfig['x'], imageConfig['h'] - featureConfig['y'] - 1],
[imageConfig['w'] - featureConfig['x'] - 1, imageConfig['h'] - featureConfig['y'] - 1],
[imageConfig['w'] - featureConfig['x'] - 1, -1 * featureConfig['y']]]) \
.reshape(-1, 1, 2)
normalized_dst = cv2.perspectiveTransform(normalized_pts, M)
cv2.polylines(img_detect, [np.int32(normalized_dst)], True, 0, 2)
# add offset to src_pts so that it can create right matrix
for p in src_pts:
p[0][0] += featureConfig.get('x', 0)
p[0][1] += featureConfig.get('y', 0)
M2, mask2 = cv2.findHomography(dst_pts, src_pts, cv2.RANSAC, 5.0)
normalized_polygons = []
for d in np.int32(normalized_dst).tolist():
normalized_polygons.append({
'x': d[0][0],
'y': d[0][1]
})
return float(len(good)) / float(len(matches)), img_detect, M2, normalized_polygons
else:
return 0, None, None, None
default='box_in_scene.png')
args = parser.parse_args()
img1 = cv.imread(cv.samples.findFile(args.input1), cv.IMREAD_GRAYSCALE)
img2 = cv.imread(cv.samples.findFile(args.input2), cv.IMREAD_GRAYSCALE)
if img1 is None or img2 is None:
print('Could not open or find the images!')
exit(0)
#-- Step 1: Detect the keypoints using SURF Detector, compute the descriptors
minHessian = 400
detector = cv.xfeatures2d_SURF.create(hessianThreshold=minHessian)
keypoints1, descriptors1 = detector.detectAndCompute(img1, None)
keypoints2, descriptors2 = detector.detectAndCompute(img2, None)
#-- Step 2: Matching descriptor vectors with a brute force matcher
# Since SURF is a floating-point descriptor NORM_L2 is used
matcher = cv.DescriptorMatcher_create(cv.DescriptorMatcher_BRUTEFORCE)
matches = matcher.match(descriptors1, descriptors2)
#-- Draw matches
img_matches = np.empty(
(max(img1.shape[0], img2.shape[0]), img1.shape[1] + img2.shape[1], 3),
dtype=np.uint8)
cv.drawMatches(img1, keypoints1, img2, keypoints2, matches, img_matches)
#-- Show detected matches
cv.imshow('Matches', img_matches)
cv.waitKey()
kp2, des2 = sift.detectAndCompute(im2, None)
bf = cv2.BFMatcher(cv2.NORM_L2, crossCheck=True)
matches = bf.match(des1, des2)
matches = sorted(matches, key=lambda x: x.distance)
src_pts = np.array(np.float32([kp1[m.queryIdx].pt for m in matches]))
dst_pts = np.array(np.float32([kp2[m.trainIdx].pt for m in matches]))
model, inlier = ransac((src_pts, dst_pts),
AffineTransform,
min_samples=4,
residual_threshold=8,
max_trials=10000)
n_inliers = np.sum(inlier)
print('Number of inliers: %d.' % n_inliers)
inlier_keypoints_left = [
cv2.KeyPoint(point[0], point[1], 1) for point in src_pts[inlier]
]
inlier_keypoints_right = [
cv2.KeyPoint(point[0], point[1], 1) for point in dst_pts[inlier]
]
placeholder_matches = [cv2.DMatch(idx, idx, 1) for idx in range(n_inliers)]
image3 = cv2.drawMatches(im1, inlier_keypoints_left, im2,
inlier_keypoints_right, placeholder_matches, None)
cv2.imwrite(
'/scratch/udit/robotcar/overcast/ipm/sift/pair' + str(i) + '.png',
image3)
print(i)
for m in good]).reshape(-1, 1, 2)
ptsImage2 = np.array([keypointsImage2[m.trainIdx].pt
for m in good]).reshape(-1, 1, 2)
# Getting homography matrix after applying RANSAC on
# well matched keypoints on both images with projection error <= 1
H, mask = cv2.findHomography(ptsImage1, ptsImage2, cv2.RANSAC)
print('Homography Matrix:')
print(H)
# Get 10 inlier matches after applying RANSAC
matchesMask = getInliers(mask, 10)
inlierImage = cv2.drawMatches(image1,
keypointsImage1,
image2,
keypointsImage2,
good,
None,
matchesMask=matchesMask,
flags=2)
cv2.imwrite('Results/task1_matches.jpg', inlierImage)
# Getting corners of image 1 in the 2nd plane
h, w, d = image1.shape
image1Corners = np.float32([[0, 0], [0, h - 1], [w - 1, h - 1],
[w - 1, 0]]).reshape(-1, 1, 2)
image1CornersPlane2 = np.squeeze(cv2.perspectiveTransform(image1Corners, H))
# The following function gives the max dimensions that image1 can
# take in the second plane (for displaying all pixels)
xMin, yMin, xMax, yMax = getExtremePoints(image1CornersPlane2)
def KAZE(prev_frame, frame):
## [load]
gray0 = cv.cvtColor(prev_frame, cv.COLOR_BGR2GRAY)
gray1 = cv.cvtColor(frame, cv.COLOR_BGR2GRAY)
## [load]
## [KAZE]
kaze = cv.KAZE_create()
t1d = cv.getTickCount()
kpts0, desc0 = kaze.detectAndCompute(gray0, None)
kpts1, desc1 = kaze.detectAndCompute(gray1, None)
t2d = cv.getTickCount()
# time for detection
tDetectKaze = 1000 * (t2d - t1d) / cv.getTickFrequency()
## [KAZE]
## [Brute-Force matching]
matcher = cv.BFMatcher(cv.NORM_L2, crossCheck=False)
t1m = cv.getTickCount()
matches = matcher.knnMatch(desc0, desc1, 2)
t2m = cv.getTickCount()
# time for detection
tMatchKaze = 1000 * (t2m - t1m) / cv.getTickFrequency()
## [Brute-Force matching]
## [ratio test filtering]
matched = []
match_ratio = 0.75
for m, n in matches:
if m.distance < match_ratio * n.distance:
matched.append(m)
## [ratio test filtering]
## [draw final matches]
res = np.empty((max(gray0.shape[0],
gray1.shape[0]), gray0.shape[1] + gray1.shape[1], 3),
dtype=np.uint8)
cv.drawMatches(gray0,
kpts0,
gray1,
kpts1,
matched,
res,
flags=cv.DRAW_MATCHES_FLAGS_NOT_DRAW_SINGLE_POINTS)
## [draw final matches]
## [RESULTS]
print('\nKAZE Matching Results')
print('*******************************')
print('# Keypoints 1: \t',
len(kpts0))
print('# Keypoints 2: \t',
len(kpts1))
print('# Matches: \t',
len(matched))
print('# Detection and Description Time (ms): \t',
tDetectKaze)
print('# Matching Time (ms): \t',
tMatchKaze)
return res
kp1,des1 = orb.detectAndCompute(reeses,mask=None) kp2,des2 = orb.detectAndCompute(cereals,mask=None) bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True) matches = bf.match(des1,des2) # matches[i].distance -> level of similarity # Sort the matches by distance matches = sorted(matches, key=lambda x:x.distance) ## Draw Lines where the matches are reeses_matches = cv2.drawMatches(reeses,kp1,cereals,kp2,matches[:25], None) # -------------------- Sift sift = cv2.xfeatures2d.SIFT_create() kp1,des1 = sift.detectAndCompute(reeses,mask=None) kp2,des2 = sift.detectAndCompute(cereals,mask=None) bf = cv2.BFMatcher() # k number of best matches # Return one array of 2 matches [[match1,match2],...] matches = bf.knnMatch(des1,des2,k=2) goodMatches = []
# =============================================================================
import cv2
import numpy as np
img = cv2.imread('wd.jpg')
img2= cv2.imread('wd2.jpg')
gray1 = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)
sift = cv2.xfeatures2d.SIFT_create()
kp1, des1 = sift.detectAndCompute(gray1, None)
kp2, des2 = sift.detectAndCompute(gray2, None)
bf = cv2.BFMatcher()
matches = bf.match(des1, des2)
matches = sorted(matches, key = lambda x:x.distance)
result = np.zeros((img.shape[1] + img2.shape[1], img.shape[0]),np.uint8)
result = cv2.drawMatches(img, kp1, img2, kp2, matches[:30], result, (0,255,0),flags = 0)
cv2.imshow('result', result)
cv2.waitKey(0)
cv2.destroyAllWindows()
matchImg = np.zeros_like(img1)
grayImg1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
grayImg2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)
sift = cv2.xfeatures2d.SIFT_create()
kp1, des1 = sift.detectAndCompute(grayImg1, None)
kp2, des2 = sift.detectAndCompute(grayImg2, None)
bf = cv2.BFMatcher_create(normType=cv2.NORM_L2, crossCheck=True)
matches = bf.match(des1, des2)
matches = sorted(matches, key=lambda x: x.distance)
matchImg = cv2.drawMatches(img1,
kp1,
img2,
kp2,
matches,
matchImg,
flags=cv2.DRAW_MATCHES_FLAGS_NOT_DRAW_SINGLE_POINTS)
# bf = cv2.BFMatcher_create(normType=cv2.NORM_L2, crossCheck=False)
# matches = bf.knnMatch(des1, des2, k=2)
# 使用阈值筛选距离
# good = []
# for m, n in matches:
# if m.distance < 0.05*n.distance:
# good.append([m])
#
# matchImg = cv2.drawMatchesKnn(img1, kp1, img2, kp2, good, matchImg, flags=2)
cv2.imshow('Match Image', matchImg)
import cv2
import numpy as np
img1 = cv2.imread("the_book_thief.jpg", cv2.IMREAD_GRAYSCALE)
img2 = cv2.imread("me_holding_book.jpg", cv2.IMREAD_GRAYSCALE)
# ORB Detector
orb = cv2.ORB_create()
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)
# Brute Force Matching
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
matches = bf.match(des1, des2)
matches = sorted(matches, key=lambda x: x.distance)
matching_result = cv2.drawMatches(img1,
kp1,
img2,
kp2,
matches[:50],
None,
flags=2)
cv2.imshow("Img1", img1)
cv2.imshow("Img2", img2)
cv2.imshow("Matching result", matching_result)
cv2.waitKey(0)
cv2.destroyAllWindows()
def align_images(im1, im_ref, savematch=False):
"""
对齐图像
:param im1: 被对齐的图像，可为彩色或灰度
:param im_ref: 参考图像，类型必须和 im1 保持一致
:param savematch: 是否保存特征点匹配图，只能保存最后一次的匹配
:return:
"""
# 需要调整下面参数，保证全部试卷可以，又提高速度
MAX_FEATURES = 1000
GOOD_MATCH_PERCENT = 0.5
# Convert images to grayscale
if len(im1.shape) == 3:
im1Gray = cv2.cvtColor(im1, cv2.COLOR_BGR2GRAY)
im2Gray = cv2.cvtColor(im_ref, cv2.COLOR_BGR2GRAY)
else:
im1Gray = im1
im2Gray = im_ref
# Detect ORB features and compute descriptors.
orb = cv2.ORB_create(MAX_FEATURES)
keypoints1, descriptors1 = orb.detectAndCompute(im1Gray, None)
keypoints2, descriptors2 = orb.detectAndCompute(im2Gray, None)
# Match features.
matcher = cv2.DescriptorMatcher_create(
cv2.DESCRIPTOR_MATCHER_BRUTEFORCE_HAMMING)
matches = matcher.match(descriptors1, descriptors2, None)
# Sort matches by score
matches.sort(key=lambda x: x.distance, reverse=False)
# Remove not so good matches
numGoodMatches = int(len(matches) * GOOD_MATCH_PERCENT)
matches = matches[:numGoodMatches]
if savematch:
# Draw top matches
imMatches = cv2.drawMatches(im1, keypoints1, im_ref, keypoints2,
matches, None)
cv2.imwrite("../result/matches.jpg", imMatches)
# Extract location of good matches
points1 = np.zeros((len(matches), 2), dtype=np.float32)
points2 = np.zeros((len(matches), 2), dtype=np.float32)
for i, match in enumerate(matches):
points1[i, :] = keypoints1[match.queryIdx].pt
points2[i, :] = keypoints2[match.trainIdx].pt
# Find homography
# h, mask = cv2.findHomography(points1, points2, cv2.RANSAC)
# h, _ = cv2.estimateAffinePartial2D(points1, points2, method=cv2.RANSAC)
h, _ = cv2.estimateAffine2D(points1, points2, method=cv2.RANSAC)
# Use homography
height, width = im_ref.shape[:2]
# im1Reg = cv2.warpPerspective(im1, h, (width, height))
im1Reg = cv2.warpAffine(im1,
h, (width, height),
flags=cv2.INTER_LINEAR,
borderMode=cv2.BORDER_REPLICATE)
return im1Reg, h
good=np.asarray(good)
h,w = image_mountain1.shape[:-1]
pts = np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2)
dst = cv2.perspectiveTransform(pts,H)
result1 = panoTwoImages(image_mountain2_color, image_mountain1_color, H)
else:
print ("No good match - %d/%d" % (len(good),MIN_MATCH_COUNT))
matchesMask = None
draw_params = dict(matchColor = (0,200,0),
singlePointColor =(000, 0, 0),
matchesMask = matchesMask,
flags = 2)
img3 = cv2.drawMatches(image_mountain1_color,keypoints,image_mountain2_color,keypoints1,good,None,**draw_params)
cv2.imwrite("task1_matches.jpg",img3)
cv2.imwrite("task1_pano.jpg",result1)
#task2
tsucuba_left=cv2.imread("tsucuba_left.png", cv2.IMREAD_GRAYSCALE)
tsucuba_right=cv2.imread("tsucuba_right.png", cv2.IMREAD_GRAYSCALE)
tsucuba_left_color=cv2.imread("tsucuba_left.png")
tsucuba_right_color=cv2.imread("tsucuba_right.png")
sift=cv2.xfeatures2d.SIFT_create()
keypoints,descriptors=sift.detectAndCompute(tsucuba_left,None)
keypoints1,descriptors1=sift.detectAndCompute(tsucuba_right,None)
tsucuba_left=cv2.drawKeypoints(tsucuba_left_color,keypoints,None)
tsucuba_right=cv2.drawKeypoints(tsucuba_right_color,keypoints1,None)
cv2.imwrite("task2_sift1.jpg",tsucuba_left)
cv2.imwrite("task2_sift2.jpg",tsucuba_right)