├── README.md
└── orb_ransac.py


/README.md:
--------------------------------------------------------------------------------
 1 | # Image-Matching-by-ORB-Feature-Using-RANSAC
 2 | 
 3 | Overview
 4 | -
 5 | In this program, when performing matching between images, RANSAC is applied to increase the reliability of matching using extracted ORB features.
 6 | 
 7 | Environment
 8 | -
 9 | Python3<br>
10 | OpenCV:3.4.3
11 | 
12 | Procedure
13 | -
14 | 1. Load image file
15 | 1. Extraction of feature points & feature quantities with ORB
16 | 1. Application of RANSAC
17 | 1. Write feature points on the image Output of result
18 | 
19 | Example of Result
20 | -
21 | ![rena-ransac](https://user-images.githubusercontent.com/43288669/61606127-c2c69c00-ac83-11e9-96b2-8be0ff3649b7.png)
22 | 


--------------------------------------------------------------------------------
/orb_ransac.py:
--------------------------------------------------------------------------------
 1 | import numpy as np
 2 | import cv2
 3 | from matplotlib import pyplot as plt
 4 | 
 5 | MIN_MATCH_COUNT = 10 #Number of minimum required feature points
 6 | good_match_rate = 0.15
 7 | 
 8 | img1 = cv2.imread('./image1.png',0) # queryImage
 9 | img2 = cv2.imread('./image2.png',0) # trainImage
10 |  
11 | # Initiate ORB detector
12 | orb = cv2.ORB_create()
13 | 
14 | # find the keypoints and descriptors with ORB
15 | kp1, des1 = orb.detectAndCompute(img1,None)
16 | kp2, des2 = orb.detectAndCompute(img2,None)
17 | 
18 | # create BFmatcher object
19 | bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
20 | #bf = cv2.BFMatcher(cv2.NORM_L1, crossCheck=False)
21 | # Match descriptors.
22 | matches = bf.match(des1,des2)
23 | 
24 | 
25 | matches = sorted(matches, key = lambda x:x.distance)
26 | good = matches[:int(len(matches) * good_match_rate)]
27 | 
28 | if len(good)>MIN_MATCH_COUNT:
29 |     src_pts = np.float32([ kp1[m.queryIdx].pt for m in good ]).reshape(-1,1,2)
30 |     dst_pts = np.float32([ kp2[m.trainIdx].pt for m in good ]).reshape(-1,1,2)
31 | 
32 |     M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC,5.0)
33 |     matchesMask = mask.ravel().tolist()
34 | 
35 |     h,w = img1.shape
36 |     pts = np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2)
37 |     dst = cv2.perspectiveTransform(pts,M)
38 | 
39 |     img2 = cv2.polylines(img2,[np.int32(dst)],True,255,3, cv2.LINE_AA)
40 | 
41 | else:
42 |     print ("Not enough matches are found - %d/%d") % (len(good),MIN_MATCH_COUNT)
43 |     matchesMask = None
44 | 
45 | draw_params = dict(matchColor = (0,255,0), # draw matches in green color
46 |                    singlePointColor = None,
47 |                    matchesMask = matchesMask, # draw only inliers
48 |                    flags = 2)
49 | 
50 | img3 = cv2.drawMatches(img1,kp1,img2,kp2,good,None,**draw_params)
51 | 
52 | plt.imshow(img3, 'gray'),plt.show()
53 | 


--------------------------------------------------------------------------------