SIFT图像匹配及其python实现

参考SIFT图像匹配及其python实现 - 知乎

特征点提取&生成描述

import cv2
import time

#  计算特征点提取&生成描述时间
start = time.time()
sift = cv2.xfeatures2d.SIFT_create()
#  使用SIFT查找关键点key points和描述符descriptors
kp1, des1 = sift.detectAndCompute(image1, None)
kp2, des2 = sift.detectAndCompute(image2, None)
end = time.time()
print("特征点提取&生成描述运行时间:%.2f秒"%(end-start))

kp_image1 = cv2.drawKeypoints(image1, kp1, None)
kp_image2 = cv2.drawKeypoints(image2, kp2, None)

plt.figure()
plt.imshow(kp_image1)
plt.savefig('kp_image1.png', dpi = 300)

plt.figure()
plt.imshow(kp_image2)
plt.savefig('kp_image2.png', dpi = 300)

#  查看关键点
print("关键点数目:", len(kp1))

for i in range(2):
    print("关键点", i)
    print("数据类型:", type(kp1[i]))
    print("关键点坐标:", kp1[i].pt)
    print("邻域直径:", kp1[i].size)
    print("方向:", kp1[i].angle)
    print("所在的图像金字塔的组:", kp1[i].octave)
    print("================")

#  查看描述
print("描述的shape:", des1.shape)
for i in range(2):
    print("描述", i)
    print(des1[i])

画出对应匹配

ratio = 0.85

#  计算匹配点匹配时间
start = time.time()

#  K近邻算法求取在空间中距离最近的K个数据点，并将这些数据点归为一类
matcher = cv2.BFMatcher()
raw_matches = matcher.knnMatch(des1, des2, k = 2)
good_matches = []
for m1, m2 in raw_matches:
    #  如果最接近和次接近的比值大于一个既定的值，那么我们保留这个最接近的值，认为它和其匹配的点为good_match
    if m1.distance < ratio * m2.distance:
        good_matches.append([m1])
end = time.time()
print("匹配点匹配运行时间:%.2f秒"%(end-start))

matches = cv2.drawMatchesKnn(image1, kp1, image2, kp2, good_matches, None, flags = 2)

plt.figure()
plt.imshow(matches)
plt.savefig('matches.png', dpi = 300)

#  匹配对的数目
print("匹配对的数目:", len(good_matches))
for i in range(2):
    print("匹配", i)
    print("数据类型:", type(good_matches[i][0]))
    print("描述符之间的距离:", good_matches[i][0].distance)
    print("查询图像中描述符的索引:", good_matches[i][0].queryIdx)
    print("目标图像中描述符的索引:", good_matches[i][0].trainIdx)
    print("================")

python+OpenCV笔记（三十五）：特征匹配——基于FLANN的匹配、基于FLANN进行单应性匹配_ReadyGo!!!的博客-CSDN博客_单应性匹配

Python opencv学习-16 SIFT+单应性矩阵+特征匹配（Flann）+目标检测_一头秀发的假程序猿的博客-CSDN博客_sift 目标检测