实验环境：anaconda、jupyter notebook

实验用到的包：opencv、matplotlib、numpy

注：opencv在3.4.2之后sift就不是免费的了

我用的是3.4.1.15版本

实验使用到的图片

一、sift函数获取特征值

读入图片

book = cv2.imread('book.png', cv2.IMREAD_GRAYSCALE)
pile = cv2.imread('pile.png', cv2.IMREAD_GRAYSCALE)plt.imshow(book,'gray')
plt.show()
plt.imshow(pile,'gray')
plt.show()

获取特征点和特征向量

# 计算特征点和特征向量
sift = cv2.xfeatures2d.SIFT_create()kp1,des1 = sift.detectAndCompute(book, None)
kp2,des2 = sift.detectAndCompute(pile, None)

一对一匹配

# 一对一
bf = cv2.BFMatcher(crossCheck=True)matches = bf.match(des1,des2)
matches = sorted(matches, key=lambda x : x.distance)res = cv2.drawMatches(book, kp1, pile, kp2, matches[:10],None, flags=2)plt.figure(figsize=(12,8))
plt.imshow(res,'gray')
plt.show()

显示前十个匹配

k对最佳匹配

# k对最佳匹配
bf = cv2.BFMatcher()
matches = bf.knnMatch(des1,des2, k=2)# 把距离小于阈值的记录下来
good = []
for m,n in matches:if m.distance < 0.75 * n.distance:good.append([m])res = cv2.drawMatchesKnn(book, kp1, pile, kp2,good,None, flags=2)plt.figure(figsize=(12,8))
plt.imshow(res,'gray')
plt.show()

二、图像拼接实验

读入图片

ml = cv2.imread('ml.png')
ml_gray = cv2.cvtColor(ml, cv2.COLOR_BGR2GRAY)
mr = cv2.imread('mr.png')
mr_gray = cv2.cvtColor(mr, cv2.COLOR_BGR2GRAY)plt.imshow(ml_gray, 'gray')
plt.show()plt.imshow(mr_gray, 'gray')
plt.show()

获取特征点和特征向量

# 获取特征点和特征向量
sift = cv2.xfeatures2d.SIFT_create()
kpl,desl = sift.detectAndCompute(ml_gray, None)
kpl_f = np.float32([kp.pt for kp in kpl])
kpr,desr = sift.detectAndCompute(mr_gray, None)
kpr_f = np.float32([kp.pt for kp in kpr])# 匹配并显示
bf = cv2.BFMatcher(crossCheck=True)matches = bf.match(desl,desr)
matches = sorted(matches, key=lambda x : x.distance)res = cv2.drawMatches(ml_gray, kpl, mr_gray, kpr, matches[:100],None, flags=2)plt.figure(figsize=(12,8))
plt.imshow(res,'gray')
plt.show()

拼接图片

拼接图片实质上就是把一张图片的一部分变化到匹配另一张图片后，把另一张图片覆盖到变化的部分上

matcher = cv2.BFMatcher()raw_matches = matcher.knnMatch(desr, desl, 2)H = Nonematches = []
for m in raw_matches:# 保留合适的特征值if len(m) == 2 and m[0].distance < m[1].distance * 0.75 :matches.append([m[0].trainIdx, m[0].queryIdx])# 配对大于4时，计算时间变换矩阵
if len(matches) > 4:# 获取配对的点坐标ptsl = np.float32([kpl_f[i] for (i,_) in matches])ptsr = np.float32([kpr_f[i] for (_,i) in matches])# 计算视角变换矩阵(H, status) = cv2.findHomography(ptsr, ptsl, cv2.RANSAC, 4)#对右图进行变换
result = cv2.warpPerspective(mr,H,(mr.shape[1] + ml.shape[1],mr.shape[0]))plt.title('before')
plt.imshow(cv2.cvtColor(result, cv2.COLOR_BGR2RGB))
plt.show() # 左图覆盖
result[0:ml.shape[0], 0:ml.shape[1]] = mlplt.title('after')
plt.imshow(cv2.cvtColor(result, cv2.COLOR_BGR2RGB))
plt.show()