opencv-python图像配准(匹配和叠加)的实现

import cv2 as cv
import numpy as np
import matplotlib.pyplot as plt
original_image = cv.imread("Image A.jpg")
target_image = cv.imread("Image B.jpg")
# 生成透视矩阵
src_points = np.array([[957, 1655], [2177, 1170], [2676, 24], [2487, 1931]], dtype=np.float32)
den_points = np.array([[687, 1150], [2000, 996], [2757, 18], [2098, 1819]], dtype=np.float32)
# getPerspectiveTransform可以得到从点集src_points到点集den_points的透视变换矩阵
T = cv.getPerspectiveTransform(src_points, den_points)
# 进行透视变换
# 注意透视变换第三个参数为变换后图片大小，格式为（高度，宽度）
warp_imgae = cv.warpPerspective(original_image, T, (target_image.shape[1], target_image.shape[0]))
plt.imshow(warp_imgae)
plt.show()

# 设置原始和目标特征点
src_more_point = np.float32([[957, 1655], [2177, 1170], [620, 2586], [1280, 2316], [2487, 1931], [937, 758], [2676, 24]]).reshape(-1, 1, 2)
den_more_point = np.float32([[687, 1150], [2000, 996], [121, 1974], [927, 1886], [2098, 1819], [899, 280], [2757, 18]]).reshape(-1, 1, 2)
# 调用库函数计算特征矩阵
# cv.findHomography第三个参数为计算单位矩阵所用的方法，0为常规算法，cv.RANSAC为基于RANSAC的鲁棒算法，cv.LMEDS为最小中值
# 鲁棒算法,cv.RHO基于PROSAC的鲁棒算法.第四个参数取值范围在1到10，绝一个点对的阈值。原图像的点经过变换后点与目标图像上对应
# 点的误差.返回值中H为变换矩阵.mask是掩模，在线的点.
H, status = cv.findHomography(src_more_point, den_more_point, cv.RANSAC, 5.0)
# 进行透视变换
warped_more_point_image = cv.warpPerspective(original_image, H, (target_image.shape[1], target_image.shape[0]))

# 用AKAZE库函数进行自动特征检测，AKAZE与SIFT等属于相似的 特征检测，但是有一些不同
akaze = cv.AKAZE_create()
# Find the keypoints and descriptors with SIFT
kp1, des1 = akaze.detectAndCompute(original_image_gray, None)
kp2, des2 = akaze.detectAndCompute(target_image_gray, None)

bf = cv.BFMatcher()
matches = bf.knnMatch(des1, des2, k=2)
good_matches = []
for m, n in matches:
    if m.distance  0.75 * n.distance:
        good_matches.append([m])

# 画出符合条件的匹配点的连线
img3 = cv.drawMatchesKnn(original_image_gray, kp1, target_image_gray, kp2, good_matches, None, flags=cv.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)
cv.imwrite('matches.jpg', img3)


src_automatic_points = np.float32([kp1[m[0].queryIdx].pt for m in good_matches]).reshape(-1, 1, 2)
den_automatic_points = np.float32([kp2[m[0].trainIdx].pt for m in good_matches]).reshape(-1, 1, 2)

# 调用库函数计算特征矩阵
H, status = cv.findHomography(src_more_point, den_more_point, cv.RANSAC, 5.0)
# 进行透视变换
warped_automatic_image = cv.warpPerspective(original_image, H, (target_image.shape[1], target_image.shape[0]))

# 绘制图像
my_draw(warped_automatic_image, tip='automatic')