1 import numpy as np
 2 import cv2 as cv
 3 filename = "picture/chessboard.png"
 4 img = cv.imread(filename)
 5 gray = cv.cvtColor(img,cv.COLOR_BGR2GRAY)
 6 gray = np.float32(gray)
 7 dst = cv.cornerHarris(gray,2,3,0.04)
 8 #result is dilated for marking the corners, not important
 9 dst = cv.dilate(dst,None)
10 # Threshold for an optimal value, it may vary depending on the image.
11 img[dst>0.01*dst.max()]=[0,0,255]
12 cv.imshow(\'dst\',img)
13 if cv.waitKey(0) & 0xff == 27:
14     cv.destroyAllWindows()

 1 import numpy as np
 2 import cv2 as cv
 3 import matplotlib.pyplot as plt
 4 img1 = cv.imread(\'picture/box.png\',0)          # queryImage
 5 img2 = cv.imread(\'picture/box_in_scene.png\',0) # trainImage
 6 # Initiate ORB detector
 7 orb = cv.ORB_create()
 8 # find the keypoints and descriptors with ORB
 9 kp1, des1 = orb.detectAndCompute(img1,None)
10 kp2, des2 = orb.detectAndCompute(img2,None)
11 
12 # create BFMatcher object
13 bf = cv.BFMatcher(cv.NORM_HAMMING, crossCheck=True)
14 # Match descriptors.
15 matches = bf.match(des1,des2)
16 # Sort them in the order of their distance.
17 matches = sorted(matches, key = lambda x:x.distance)
18 # Draw first 10 matches.
19 img3 = cv.drawMatches(img1,kp1,img2,kp2,matches[:20],None, flags=2)
20 plt.imshow(img3),plt.show()

 1 from Stitcher import Stitcher
 2 import cv2
 3 
 4 # 读取拼接图片
 5 imageA = cv2.imread("image/3.png")
 6 imageB = cv2.imread("image/4.png")
 7 
 8 # 把图片拼接成全景图
 9 stitcher = Stitcher()
10 (result, vis) = stitcher.stitch([imageA, imageB], showMatches=True)
11 
12 # 显示所有图片
13 cv2.imshow("Image A", imageA)
14 cv2.imshow("Image B", imageB)
15 cv2.imshow("Keypoint Matches", vis)
16 cv2.imshow("Result", result)
17 cv2.waitKey(0)
18 cv2.destroyAllWindows()

  1 import numpy as np
  2 import cv2
  3 
  4 class Stitcher:
  5 
  6     #拼接函数
  7     def stitch(self, images, ratio=0.75, reprojThresh=4.0,showMatches=False):
  8         #获取输入图片
  9         (imageB, imageA) = images
 10         #检测A、B图片的SIFT关键特征点，并计算特征描述子
 11         (kpsA, featuresA) = self.detectAndDescribe(imageA)
 12         (kpsB, featuresB) = self.detectAndDescribe(imageB)
 13 
 14         # 匹配两张图片的所有特征点，返回匹配结果
 15         M = self.matchKeypoints(kpsA, kpsB, featuresA, featuresB, ratio, reprojThresh)
 16 
 17         # 如果返回结果为空，没有匹配成功的特征点，退出算法
 18         if M is None:
 19             return None
 20 
 21         # 否则，提取匹配结果
 22         # H是3x3视角变换矩阵      
 23         (matches, H, status) = M
 24         # 将图片A进行视角变换，result是变换后图片
 25         result = cv2.warpPerspective(imageA, H, (imageA.shape[1] + imageB.shape[1], imageA.shape[0]))
 26         # 将图片B传入result图片最左端
 27         result[0:imageB.shape[0], 0:imageB.shape[1]] = imageB
 28 
 29         # 检测是否需要显示图片匹配
 30         if showMatches:
 31             # 生成匹配图片
 32             vis = self.drawMatches(imageA, imageB, kpsA, kpsB, matches, status)
 33             # 返回结果
 34             return (result, vis)
 35 
 36         # 返回匹配结果
 37         return result
 38 
 39     def detectAndDescribe(self, image):
 40         # 将彩色图片转换成灰度图
 41         gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
 42 
 43         # 建立SIFT生成器
 44         descriptor = cv2.xfeatures2d.SIFT_create()
 45         # 检测SIFT特征点，并计算描述子
 46         (kps, features) = descriptor.detectAndCompute(image, None)
 47 
 48         # 将结果转换成NumPy数组
 49         kps = np.float32([kp.pt for kp in kps])
 50 
 51         # 返回特征点集，及对应的描述特征
 52         return (kps, features)
 53 
 54     def matchKeypoints(self, kpsA, kpsB, featuresA, featuresB, ratio, reprojThresh):
 55         # 建立暴力匹配器
 56         matcher = cv2.DescriptorMatcher_create("BruteForce")
 57   
 58         # 使用KNN检测来自A、B图的SIFT特征匹配对，K=2
 59         rawMatches = matcher.knnMatch(featuresA, featuresB, 2)
 60 
 61         matches = []
 62         for m in rawMatches:
 63             # 当最近距离跟次近距离的比值小于ratio值时，保留此匹配对
 64             if len(m) == 2 and m[0].distance < m[1].distance * ratio:
 65             # 存储两个点在featuresA, featuresB中的索引值
 66                 matches.append((m[0].trainIdx, m[0].queryIdx))
 67 
 68         # 当筛选后的匹配对大于4时，计算视角变换矩阵
 69         if len(matches) > 4:
 70             # 获取匹配对的点坐标
 71             ptsA = np.float32([kpsA[i] for (_, i) in matches])
 72             ptsB = np.float32([kpsB[i] for (i, _) in matches])
 73 
 74             # 计算视角变换矩阵
 75             (H, status) = cv2.findHomography(ptsA, ptsB, cv2.RANSAC, reprojThresh)
 76 
 77             # 返回结果
 78             return (matches, H, status)
 79 
 80         # 如果匹配对小于4时，返回None
 81         return None
 82 
 83     def drawMatches(self, imageA, imageB, kpsA, kpsB, matches, status):
 84         # 初始化可视化图片，将A、B图左右连接到一起
 85         (hA, wA) = imageA.shape[:2]
 86         (hB, wB) = imageB.shape[:2]
 87         vis = np.zeros((max(hA, hB), wA + wB, 3), dtype="uint8")
 88         vis[0:hA, 0:wA] = imageA
 89         vis[0:hB, wA:] = imageB
 90 
 91         # 联合遍历，画出匹配对
 92         for ((trainIdx, queryIdx), s) in zip(matches, status):
 93             # 当点对匹配成功时，画到可视化图上
 94             if s == 1:
 95                 # 画出匹配对
 96                 ptA = (int(kpsA[queryIdx][0]), int(kpsA[queryIdx][1]))
 97                 ptB = (int(kpsB[trainIdx][0]) + wA, int(kpsB[trainIdx][1]))
 98                 cv2.line(vis, ptA, ptB, (0, 255, 0), 1)
 99 
100         # 返回可视化结果
101         return vis