Python OpenCV特征檢測(cè)之特征匹配方式詳解

更新時(shí)間：2021年12月31日 15:21:21 作者：機(jī)器視覺(jué)小學(xué)徒

OpenCV中提供了兩種技術(shù)用于特征匹配，分別為Brute-Force匹配器和基于FLANN的匹配器。本文將為大家詳細(xì)介紹一下這兩種匹配方式，需要的可以參考一下

前言

獲得圖像的關(guān)鍵點(diǎn)后，可通過(guò)計(jì)算得到關(guān)鍵點(diǎn)的描述符。關(guān)鍵點(diǎn)描述符可用于圖像的特征匹配。通常，在計(jì)算圖A是否包含圖B的特征區(qū)域時(shí)，將圖A稱做訓(xùn)練圖像，將圖B稱為查詢圖像。圖A的關(guān)鍵點(diǎn)描述符稱為訓(xùn)練描述符，圖B的關(guān)鍵點(diǎn)描述符稱為查詢描述符。

一、暴力匹配器

暴力匹配器使用描述符進(jìn)行特征比較。在比較時(shí)，暴力匹配器首先在查詢描述符中取一個(gè)關(guān)鍵點(diǎn)的描述符，將其與訓(xùn)練描述符中的所有關(guān)鍵點(diǎn)描述符進(jìn)行比較，每次比較后會(huì)給出一個(gè)距離值，距離最小的值對(duì)應(yīng)最佳匹配結(jié)果。所有描述符比較完后，匹配器返回匹配結(jié)果列表。

OpenCV的cv2.BFMatcher_create()函數(shù)用于創(chuàng)建暴力匹配器，其基本格式如下：

bf = cv2.BFMatcher_create([normType[, crossCheck]])

bf為返回的暴力匹配器對(duì)象

normType為距離測(cè)量類型, 默認(rèn)為cv2.NORM_L2, 通常, SIFT描述符使用cv2.NORM_L1或cv2.NORM_L2, ORB描述符使用cv2.NORM_HAMMING

crossCheck默認(rèn)為False, 匹配器為每個(gè)查詢描述符找到k個(gè)距離最近的匹配描述符, 為T(mén)rue時(shí), 只返回滿足交叉驗(yàn)證條件的匹配結(jié)果

暴力匹配器對(duì)象的match()方法返回每個(gè)關(guān)鍵點(diǎn)的最佳匹配結(jié)果，其基本格式如下：

ms = bf.match(des1, des2)

ms為返回的結(jié)果, 它是一個(gè)DMatch對(duì)象列表, 每個(gè)DMatch對(duì)象表示關(guān)鍵點(diǎn)的一個(gè)匹配結(jié)果, 其dintance屬性表示距離, 距離值越小匹配度越高

des1為查詢描述符

des2為訓(xùn)練描述符

獲得匹配結(jié)果后，可調(diào)用cv2.drawMatches()函數(shù)或cv2.drawMatchesKnn()函數(shù)繪制匹配結(jié)果圖像，其基本格式如下：

outImg = cv2.drawMatches(img1, keypoints1, img2, keypoints2, matches1to2[, matchColor[, singlePointColor[, matchesMask[, flags]]]])

outImg為返回的繪制結(jié)果圖像, 圖像中查詢圖像與訓(xùn)練圖像中匹配的關(guān)鍵點(diǎn)個(gè)兩點(diǎn)之間的連線為彩色

img1為查詢圖像

keypoints1為img1的關(guān)鍵點(diǎn)

img2為訓(xùn)練圖像

keypoints2為img2的關(guān)鍵點(diǎn)

matches1to2為img1與img2的匹配結(jié)果

matchColor為關(guān)鍵點(diǎn)和鏈接線的顏色, 默認(rèn)使用隨機(jī)顏色

singlePointColor為單個(gè)關(guān)鍵點(diǎn)的顏色, 默認(rèn)使用隨機(jī)顏色

matchesMask為掩膜, 用于決定繪制哪些匹配結(jié)果, 默認(rèn)為空, 表示繪制所有匹配結(jié)果

flags為標(biāo)志, 可設(shè)置為下列參數(shù)值:

cv2.DrawMatchesFlags_DEFAUL:默認(rèn)方式, 繪制兩個(gè)源圖像、匹配項(xiàng)和單個(gè)關(guān)鍵點(diǎn), 沒(méi)有圍繞關(guān)鍵點(diǎn)的圓以及關(guān)鍵點(diǎn)的大小和方向

cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS:不會(huì)繪制單個(gè)關(guān)鍵點(diǎn)

cv2.DrawMatchesFlags_DRAW_RICH_KEYPOINTS:在關(guān)鍵點(diǎn)周圍繪制具有關(guān)鍵點(diǎn)大小和方向的圓圈

# 暴力匹配器、ORB描述符和match()方法
import cv2

img1 = cv2.imread("xhu1.jpg", cv2.IMREAD_GRAYSCALE)
img2 = cv2.imread("xhu2.jpg", cv2.IMREAD_GRAYSCALE)

orb = cv2.ORB_create()
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)

bf = cv2.BFMatcher_create(cv2.NORM_HAMMING, crossCheck = False)
ms = bf.match(des1, des2)
ms = sorted(ms, key = lambda x:x.distance)
img3 = cv2.drawMatches(img1, kp1, img2, kp2, ms[:20], None, flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)

cv2.imshow("Xhu1", img1)
cv2.imshow("Xhu2", img2)
cv2.imshow("Matches", img3)

cv2.waitKey(0)
cv2.destroyAllWindows()

# 暴力匹配器、SIFT描述符和match()方法
import cv2

img1 = cv2.imread("xhu1.jpg", cv2.IMREAD_GRAYSCALE)
img2 = cv2.imread("xhu2.jpg", cv2.IMREAD_GRAYSCALE)

sift=cv2.SIFT_create()
kp1, des1 = sift.detectAndCompute(img1, None)
kp2, des2 = sift.detectAndCompute(img2, None)

bf = cv2.BFMatcher_create(cv2.NORM_L1, crossCheck = False)
ms = bf.match(des1, des2)
ms = sorted(ms, key = lambda x:x.distance)
img3 = cv2.drawMatches(img1, kp1, img2, kp2, ms[:20], None, flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)

# ms = np.expand_dims(ms,1)
# img3 = cv2.drawMatchesKnn(img1, kp1, img2, kp2, ms[:20], None, flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)

cv2.imshow("Xhu1", img1)
cv2.imshow("Xhu2", img2)
cv2.imshow("Matches", img3)

cv2.waitKey(0)
cv2.destroyAllWindows()

暴力匹配器對(duì)象的knnMatch()方法可返回指定數(shù)量的最佳匹配結(jié)果，其基本格式如下：

ms = knnMatch(des1, des2, k=n)

ms為返回的匹配結(jié)果, 每個(gè)列表元素是一個(gè)子列表, 它包含了由參數(shù)k指定個(gè)數(shù)的DMatch對(duì)象

des1為查詢描述符

des2為訓(xùn)練描述符

k為返回的最佳匹配個(gè)數(shù)

# 暴力匹配器、ORB描述符和knnMatch()方法
import cv2

img1 = cv2.imread("xhu1.jpg", cv2.IMREAD_GRAYSCALE)
img2 = cv2.imread("xhu2.jpg", cv2.IMREAD_GRAYSCALE)

orb = cv2.ORB_create()
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)

bf = cv2.BFMatcher_create(cv2.NORM_HAMMING, crossCheck = False)
ms = bf.knnMatch(des1, des2, k=2)

# 應(yīng)用比例測(cè)試選擇要使用的匹配結(jié)果
good = []
for m, n in ms:
    if m.distance < 0.75 * n.distance:
		good.append(m)

img3 = cv2.drawMatches(img1, kp1, img2, kp2, good[:20], None, flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)
# good = np.expand_dims(good,1)
#img3 = cv2.drawMatchesKnn(img1, kp1, img2, kp2, good[:20], None, flags=cv2.DrawMatchesFlags_NOT_DRAW_SINGLE_POINTS)

cv2.imshow("Xhu1", img1)
cv2.imshow("Xhu2", img2)
cv2.imshow("Matches", img3)


cv2.waitKey(0)
cv2.destroyAllWindows()

二、FLANN匹配器

FLANN(Fast Library for Approximate Nearest Neignbors)為近似最近鄰的快速庫(kù)，F(xiàn)LANN特征匹配算法比其它的最近鄰算法更快。

在創(chuàng)建FLANN匹配器時(shí)，需要傳遞兩參數(shù)：index_params和search_params。

index_params用來(lái)指定索引樹(shù)的算法類型和數(shù)量。SIFT算法可以使用下面的代碼來(lái)設(shè)置。

FLANN_INDEX_KDTREE = 1
index_params = dict(algorithm = FLANN_INDEX_KDTREE,
                    trees= 5)

ORB算法可以使用下面的代碼來(lái)設(shè)置。

FLANN_INDEX_LSH = 6
index_params = dict(algorithm = FLANN_INDEX_LSH,
                    table_number = 6,
                    key_size = 12,
                    multi_probe_level = 1)

search_params用于指定索引樹(shù)的遍歷次數(shù)，遍歷次數(shù)越多，匹配結(jié)果越精細(xì)，通常設(shè)置為50即可，如下所示：

search_params = dict(check = 50)

# FLANN匹配器、ORB描述符
import cv2

img1 = cv2.imread("xhu1.jpg", cv2.IMREAD_GRAYSCALE)
img2 = cv2.imread("xhu2.jpg", cv2.IMREAD_GRAYSCALE)

orb = cv2.ORB_create()
kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)

# 定義FLANN參數(shù)
FLANN_INDEX_LSH = 6
index_params = dict(algorithm = FLANN_INDEX_LSH,
                    table_number = 6,
                    key_size = 12,
                    multi_probe_level = 1)
search_params = dict(check = 50)

flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.match(des1, des2)
draw_params = dict(matchColor = (0,255,0),
                   singlePointColor = (255,0,0),
                   matchesMask = None,
                   flags = cv2.DrawMatchesFlags_DEFAULT)

img3 = cv2.drawMatches(img1, kp1, img2, kp2, matches[:20], None, **draw_params)

cv2.imshow("Xhu1", img1)
cv2.imshow("Xhu2", img2)
cv2.imshow("Matches", img3)


cv2.waitKey(0)
cv2.destroyAllWindows()

# FLANN匹配器、SIFT描述符
import cv2

img1 = cv2.imread("xhu1.jpg", cv2.IMREAD_GRAYSCALE)
img2 = cv2.imread("xhu2.jpg", cv2.IMREAD_GRAYSCALE)

sift = cv2.SIFT_create()
kp1, des1 = sift.detectAndCompute(img1, None)
kp2, des2 = sift.detectAndCompute(img2, None)

# 定義FLANN參數(shù)
FLANN_INDEX_KDTREE = 1
index_params = dict(algorithm = FLANN_INDEX_KDTREE,
                    trees = 5)
search_params = dict(check = 50)

flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.match(des1, des2)
draw_params = dict(matchColor = (0,255,0),
                   singlePointColor = (255,0,0),
                   matchesMask = None,
                   flags = cv2.DrawMatchesFlags_DEFAULT)

img3 = cv2.drawMatches(img1, kp1, img2, kp2, matches[:20], None, **draw_params)

cv2.imshow("Xhu1", img1)
cv2.imshow("Xhu2", img2)
cv2.imshow("Matches", img3)


cv2.waitKey(0)
cv2.destroyAllWindows()