SIFI和ORB在尺度縮放、旋轉、仿射上的特徵點不變實驗代碼，並比較SIFI和ORB提取特徵點的速度

一、SIFI算法

1.驗證旋轉不變性

import numpy as np
import time
import cv2
from matplotlib import pyplot as plt

sift = cv2.xfeatures2d.SIFT_create()
# FLANN 參數設計
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
search_params = dict(checks=50)
flann = cv2.FlannBasedMatcher(index_params,search_params)

img1 = cv2.imread('1.jpg')
img2 = cv2.imread('2.jpg')
#使用cv2.imread()接口讀圖像，讀進來的是BGR格式以及[0～255]。所以要將img轉換爲RGB格式，不然後面顯示會有色差
img1 = cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)
gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY) #灰度處理圖像
#使用cv2.imread()接口讀圖像，讀進來的是BGR格式以及[0～255]。所以要將img轉換爲RGB格式，不然後面顯示會有色差
img2 = cv2.cvtColor(img2,cv2.COLOR_BGR2RGB)
gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)

start11 = time.time()
kp1, des1 = sift.detectAndCompute(img1,None)#des是描述符
kp2, des2 = sift.detectAndCompute(img2,None)
end11 = time.time()

print (des1.shape) #描述符數組維度 
print(len(kp1))   #關鍵點個數
print (des2.shape) #描述符數組維度 
print(len(kp2))   #關鍵點個數

print(img1.shape)
print(img2.shape)
hmerge = np.hstack((img1, img2)) #水平拼接
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge)
plt.title('original')
plt.axis('off')
plt.show()

hmerge = np.hstack((gray1, gray2)) #水平拼接
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge,cmap='gray')
plt.title('gray')
plt.axis('off')
plt.show()

img3 = cv2.drawKeypoints(img1,kp1,img1,color=(255,0,255))
img4 = cv2.drawKeypoints(img2,kp2,img2,color=(255,0,255))

hmerge = np.hstack((img3, img4)) #水平拼接
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge)
plt.title('keypoint')
plt.axis('off')
plt.show()

matches = flann.knnMatch(des1,des2,k=2)
matchesMask = [[0,0] for i in range(len(matches))]

good = []
for m,n in matches:
    if m.distance < 0.7*n.distance:
        
        good.append([m])

img5 = cv2.drawMatchesKnn(img1,kp1,img2,kp2,good,None,flags=2)

plt.figure(num=1,figsize=(16,16))
plt.imshow(img5)
plt.title('SIFT_rotation')
plt.axis('off')
plt.show()

(1085, 128)
1085
(1089, 128)
1089
(400, 600, 3)
(400, 600, 3)

2.驗證尺度不變性

img1 = cv2.imread("1.jpg")
img2 = cv2.imread("3.jpg")
img1 = cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)
img2 = cv2.cvtColor(img2,cv2.COLOR_BGR2RGB)
print(img1.shape)
print(img2.shape)
plt.figure(num=1,figsize=(12,12))
plt.imshow(img1,cmap='gray')
plt.title('original_img1')
plt.axis('off')
plt.show()
plt.figure(num=1,figsize=(12,12))
plt.imshow(img2,cmap='gray')
plt.title('original_img2')
plt.axis('off')
plt.show()

#灰度化處理，打印描述符數組維度和關鍵點個數
gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)
gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)

start12 = time.time()
kp1, des1 = sift.detectAndCompute(img1,None)
kp2, des2 = sift.detectAndCompute(img2,None)
end12 = time.time()

print (des1.shape) #描述符數組維度 
print(len(kp1))   #關鍵點個數
print (des1.shape) #描述符數組維度 
print(len(kp2))   #關鍵點個數

plt.figure(num=1,figsize=(12,12))
plt.imshow(gray1,cmap='gray')
plt.title('gray1')
plt.axis('off')
plt.show()
plt.figure(num=1,figsize=(12,12))
plt.imshow(gray2,cmap='gray')
plt.title('gray2')
plt.axis('off')
plt.show()

img3 = cv2.drawKeypoints(img1,kp1,img1,color=(255,0,255))
img4 = cv2.drawKeypoints(img2,kp2,img2,color=(255,0,255))

plt.figure(num=1,figsize=(12,12))
plt.imshow(img3)
plt.title('keypoint_img3')
plt.axis('off')
plt.show()
plt.figure(num=1,figsize=(12,12))
plt.imshow(img4)
plt.title('keypoint_img4')
plt.axis('off')
plt.show()

matches = flann.knnMatch(des1,des2,k=2)
matchesMask = [[0,0] for i in range(len(matches))]

good = []
for m,n in matches:
    if m.distance < 0.7*n.distance:
        good.append([m])

img5 = cv2.drawMatchesKnn(img1,kp1,img2,kp2,good,None,flags=2)

plt.figure(num=1,figsize=(16,16))
plt.imshow(img5)
plt.title('SIFT_size')
plt.axis('off')
plt.show()

3.驗證仿射不變性 對原圖進行仿射變換並輸出

img = cv2.imread('1.jpg')
img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
rows, cols, ch = img.shape
 
pts1 = np.float32([[0, 0], [cols - 1, 0], [0, rows - 1]])
pts2 = np.float32([[cols * 0.2, rows * 0.1], [cols * 0.9, rows * 0.2], [cols * 0.1, rows * 0.9]])
 
M = cv2.getAffineTransform(pts1, pts2)
dst = cv2.warpAffine(img, M, (cols, rows))
plt.figure(num=1,figsize=(16,16))
plt.imshow(dst)
plt.title('affine_img')
plt.axis('off')
plt.show()

print(img.shape)
print(dst.shape)
hmerge = np.hstack((img, dst))  # 兩張圖片橫向合併（便於對比顯示）
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge)
plt.title('original')
plt.axis('off')
plt.show()

gray1 = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) #灰度處理圖像
gray2 = cv2.cvtColor(dst, cv2.COLOR_BGR2GRAY) #灰度處理圖像

start13 = time.time()
kp1, des1 = sift.detectAndCompute(img,None)#des是描述符
kp2, des2 = sift.detectAndCompute(dst,None)#des是描述符
end13 = time.time()

print (des1.shape) #描述符數組維度 
print(len(kp1))   #關鍵點個數
print (des2.shape) #描述符數組維度 
print(len(kp2))   #關鍵點個數

hmerge = np.hstack((gray1, gray2))  # 兩張圖片橫向合併（便於對比顯示）
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge,cmap='gray')
plt.title('gray')
plt.axis('off')
plt.show()

img3 = cv2.drawKeypoints(img,kp1,img,color=(255,0,255))
img4 = cv2.drawKeypoints(dst,kp2,dst,color=(255,0,255))

hmerge = np.hstack((img3, img4)) #水平拼接
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge)
plt.title('keypoint')
plt.axis('off')
plt.show()

matches = flann.knnMatch(des1,des2,k=2)
matchesMask = [[0,0] for i in range(len(matches))]

good = []
for m,n in matches:
    if m.distance < 0.7*n.distance:
        good.append([m])

img5 = cv2.drawMatchesKnn(img,kp1,dst,kp2,good,None,flags=2)

plt.figure(num=1,figsize=(16,16))
plt.imshow(img5)
plt.title('SIFT_affine')
plt.axis('off')
plt.show()

二、ORB算法

1.驗證旋轉不變性

import numpy as np
import cv2
from matplotlib import pyplot as plt

orb=cv2.ORB_create()
# FLANN 參數設計
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
search_params = dict(checks=50)
flann = cv2.FlannBasedMatcher(index_params,search_params)

img1 = cv2.imread('1.jpg')
#使用cv2.imread()接口讀圖像，讀進來的是BGR格式以及[0～255]。所以要將img轉換爲RGB格式，不然後面顯示會有色差
img1 = cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)
gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY) #灰度處理圖像

img2 = cv2.imread('2.jpg')
#使用cv2.imread()接口讀圖像，讀進來的是BGR格式以及[0～255]。所以要將img轉換爲RGB格式，不然後面顯示會有色差
img2 = cv2.cvtColor(img2,cv2.COLOR_BGR2RGB)
gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)

start21 = time.time()
kp1, des1 = orb.detectAndCompute(img1,None)#des是描述符
kp2, des2 = orb.detectAndCompute(img2,None)
end21 = time.time()


print (des1.shape) #描述符數組維度 
print(len(kp1))   #關鍵點個數
print (des2.shape) #描述符數組維度 
print(len(kp2))   #關鍵點個數

print(img1.shape)
print(img2.shape)
hmerge = np.hstack((img1, img2)) #水平拼接
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge)
plt.title('original')
plt.axis('off')
plt.show()

hmerge = np.hstack((gray1, gray2)) #水平拼接
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge,cmap='gray')
plt.title('gray')
plt.axis('off')
plt.show()

img3 = cv2.drawKeypoints(img1,kp1,img1,color=(255,0,255))
img4 = cv2.drawKeypoints(img2,kp2,img2,color=(255,0,255))

hmerge = np.hstack((img3, img4)) #水平拼接
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge)
plt.title('keypoint')
plt.axis('off')
plt.show()

# 獲得一個暴力匹配器的對象
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
# 利用匹配器 匹配兩個描述符的相近成都
maches = bf.match(des1,des2)
# 按照相近程度 進行排序
maches = sorted(maches, key=lambda x: x.distance)
# 畫出匹配項
img5 = cv2.drawMatches(img1,kp1,img2,kp2, maches[:], None, flags=2)

plt.figure(num=1,figsize=(16,16))
plt.imshow(img5)
plt.title('ORB_rotation')
plt.axis('off')
plt.show()

運行結果：

2.驗證尺度不變性

img1 = cv2.imread("1.jpg")
img2 = cv2.imread("3.jpg")
img1 = cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)
img2 = cv2.cvtColor(img2,cv2.COLOR_BGR2RGB)
print(img1.shape)
print(img2.shape)
plt.figure(num=1,figsize=(12,12))
plt.imshow(img1,cmap='gray')
plt.title('original_img1')
plt.axis('off')
plt.show()
plt.figure(num=1,figsize=(12,12))
plt.imshow(img2,cmap='gray')
plt.title('original_img2')
plt.axis('off')
plt.show()

gray1 = cv2.cvtColor(img1, cv2.COLOR_BGR2GRAY)



gray2 = cv2.cvtColor(img2, cv2.COLOR_BGR2GRAY)

start22 = time.time()
kp1, des1 = orb.detectAndCompute(img1,None)
kp2, des2 = orb.detectAndCompute(img2,None)
end22 = time.time()

print (des1.shape) #描述符數組維度 
print(len(kp1))   #關鍵點個數
print (des1.shape) #描述符數組維度 
print(len(kp2))   #關鍵點個數

plt.figure(num=1,figsize=(12,12))
plt.imshow(gray1,cmap='gray')
plt.title('gray1')
plt.axis('off')
plt.show()
plt.figure(num=1,figsize=(12,12))
plt.imshow(gray2,cmap='gray')
plt.title('gray2')
plt.axis('off')
plt.show()

img3 = cv2.drawKeypoints(img1,kp1,img1,color=(255,0,255))
img4 = cv2.drawKeypoints(img2,kp2,img2,color=(255,0,255))

plt.figure(num=1,figsize=(12,12))
plt.imshow(img3)
plt.title('keypoint_img3')
plt.axis('off')
plt.show()
plt.figure(num=1,figsize=(12,12))
plt.imshow(img4)
plt.title('keypoint_img4')
plt.axis('off')
plt.show()

# 獲得一個暴力匹配器的對象
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
# 利用匹配器 匹配兩個描述符的相近成都
maches = bf.match(des1,des2)
# 按照相近程度 進行排序
maches = sorted(maches, key=lambda x: x.distance)
# 畫出匹配項
img5 = cv2.drawMatches(img1,kp1,img2,kp2, maches[: 30], None, flags=2)

plt.figure(num=1,figsize=(16,16))
plt.imshow(img5)
plt.title('ORB_size')
plt.axis('off')
plt.show()

運行結果：

3.驗證仿射不變性 對原圖進行仿射變換並輸出

img = cv2.imread('1.jpg')
img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
rows, cols, ch = img.shape
 
pts1 = np.float32([[0, 0], [cols - 1, 0], [0, rows - 1]])
pts2 = np.float32([[cols * 0.2, rows * 0.1], [cols * 0.9, rows * 0.2], [cols * 0.1, rows * 0.9]])
 
M = cv2.getAffineTransform(pts1, pts2)
dst = cv2.warpAffine(img, M, (cols, rows))
plt.figure(num=1,figsize=(16,16))
plt.imshow(dst)
plt.title('affine_img')
plt.axis('off')
plt.show()

print(img.shape)
print(dst.shape)
hmerge = np.hstack((img, dst))  # 兩張圖片橫向合併（便於對比顯示）
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge)
plt.title('original')
plt.axis('off')
plt.show()

gray1 = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) #灰度處理圖像
gray2 = cv2.cvtColor(dst, cv2.COLOR_BGR2GRAY) #灰度處理圖像

start23 = time.time()
kp1, des1 = orb.detectAndCompute(img,None)#des是描述符
kp2, des2 = orb.detectAndCompute(dst,None)#des是描述符
end23 = time.time()

print (des1.shape) #描述符數組維度 
print(len(kp1))   #關鍵點個數
print (des2.shape) #描述符數組維度 
print(len(kp2))   #關鍵點個數

hmerge = np.hstack((gray1, gray2))  # 兩張圖片橫向合併（便於對比顯示）
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge,cmap='gray')
plt.title('gray')
plt.axis('off')
plt.show()

img3 = cv2.drawKeypoints(img,kp1,img,color=(255,0,255))
img4 = cv2.drawKeypoints(dst,kp2,dst,color=(255,0,255))

hmerge = np.hstack((img3, img4)) #水平拼接
plt.figure(num=1,figsize=(16,16))
plt.imshow(hmerge)
plt.title('keypoint')
plt.axis('off')
plt.show()

# 獲得一個暴力匹配器的對象
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
# 利用匹配器 匹配兩個描述符的相近成都
maches = bf.match(des1,des2)
# 按照相近程度 進行排序
maches = sorted(maches, key=lambda x: x.distance)
# 畫出匹配項
img5 = cv2.drawMatches(img,kp1,dst,kp2, maches[:], None, flags=2)

plt.figure(num=1,figsize=(16,16))
plt.imshow(img5)
plt.title('ORB_affine')
plt.axis('off')
plt.show()

運行結果：

三、比較SIFT和ORB的尺度旋轉，縮放，仿射上運行時間

print("SIFI的旋轉的體徵點提取速度",end11-start11)
print("SIFI的縮放的體徵點提取速度",end12-start12)
print("SIFI的仿射的體徵點提取速度",end13-start13)

print("ORB的旋轉的體徵點提取速度",end21-start21)
print("ORB的縮放的體徵點提取速度",end22-start22)
print("ORB的仿射的體徵點提取速度",end23-start23)

SIFI的旋轉的體徵點提取速度 0.09574317932128906
SIFI的縮放的體徵點提取速度 0.06183433532714844
SIFI的仿射的體徵點提取速度 0.09873533248901367
ORB的旋轉的體徵點提取速度 0.012612581253051758
ORB的縮放的體徵點提取速度 0.008951902389526367
ORB的仿射的體徵點提取速度 0.011967182159423828

由此可見，ORB提取特徵點的速度比SIFI好