圖像特征

• 特征理解
• 特征檢測
• 特征描述

Harris角點檢測

• cv2.cornerHarris(img, blockSize, ksize, k, borderType=…)
  • img：輸入圖像，數據類型為float32
  • blockSize：角點檢測中要考慮的領域大小
  • ksize：Sobe求導中使用的窗口大小
  • k：Harris角點檢測方程中的自由參數，取值參數為 [0.04,0.06]
  • borderType：邊界類型

import numpy as np
import cv2
from matplotlib import pyplot as plt# img = cv2.imread('./resource/opencv/image/chessboard.png', cv2.IMREAD_COLOR)
img = cv2.imread('./resource/opencv/image/pattern.png', cv2.IMREAD_COLOR)gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)gray = np.float32(gray)# 輸入圖像必須是float32，最后一個參數在0.04到0.05之間
dst = cv2.cornerHarris(gray, 2, 3, 0.05)
dst = cv2.dilate(dst, None)img[dst>0.01*dst.max()] = [0, 0, 255]cv2.imshow('dst', img)
cv2.waitKey(0)
cv2.destroyAllWindows()

亞像素級精度的角點檢測

• cv2.cornerSubPix(img, corners, winSize, zeroZone, criteria)
  最大精度的角點檢測，首先要找到 Harris角點，然后將角點的重心傳給這個函數進行修正。

import numpy as np
import cv2
from matplotlib import pyplot as pltimg = cv2.imread('./resource/opencv/image/subpixel.png', cv2.IMREAD_COLOR)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)gray = np.float32(gray)
dst = cv2.cornerHarris(gray, 2, 3, 0.04)
dst = cv2.dilate(dst, None)
ret, dst = cv2.threshold(dst, 0.01*dst.max(), 255, 0)
dst = np.uint8(dst)ret, labels, stats, centroids = cv2.connectedComponentsWithStats(dst)criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 100, 0.001)corners = cv2.cornerSubPix(gray, np.float32(centroids), (5,5), (-1, -1), criteria)res = np.hstack((centroids, corners))res = np.int0(res)
img[res[:,1],res[:,0]]=[0,0,255]
img[res[:,3],res[:,2]]=[0,255,0]cv2.imshow('img', img)
cv2.waitKey(0)
cv2.destroyAllWindows()

Harris 角點用紅色像素標出，綠色像素是修正后的角點。

Shi-Tomasi角點檢測

• cv2.goodFeatureToTrack()

import numpy as np
import cv2
from matplotlib import pyplot as pltimg = cv2.imread('./resource/opencv/image/shitomasi_block.jpg', cv2.IMREAD_COLOR)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)corners = cv2.goodFeaturesToTrack(gray, 25, 0.01, 10)corners = np.int0(corners)for i in corners:x,y = i.ravel()cv2.circle(img, (x,y), 3, 255, -1)plt.imshow(img)
plt.show()

SIFT(Scale-Invariant Feature Transfrom)

• SIFT，即尺度不變特征變換（Scale-invariant feature transform，SIFT），是用于圖像處理領域的一種描述。這種描述具有尺度不變性，可在圖像中檢測出關鍵點，是一種局部特征描述子。

• cv2.SIFT_create()

  • kp = sift.detect(img, None)：查找特征點
  • kp, des = sift.compute(img, kp)：計算特征點
  • kp, des = sift.detectAndCompute(img, None) ：直接找到特征點并計算描述符

• cv2.drawKeypoints(img, kp, out_img, flags=cv2.DRAW_MATCHES_FLAGS_NOT_DRAW_SINGLE_POINTS)：畫特征點

  • img : 輸入圖像
  • kp：圖像特征點
  • out_img：輸出圖像
  • flags：
    cv2.DRAW_MATCHES_FLAGS_DEFAULT
    cv2.DRAW_MATCHES_FLAGS_DRAW_OVER_OUTIMG
    cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS
    cv2.DRAW_MATCHES_FLAGS_NOT_DRAW_SINGLE_POINTS

import numpy as np
import cv2# 讀取圖片
# img = cv2.imread('./resource/opencv/image/home.jpg')
img = cv2.imread('./resource/opencv/image/AverageMaleFace.jpg')
key_points = img.copy()# 實例化SIFT算法
sift = cv2.SIFT_create()# 得到特征點
kp = sift.detect(img, None)
print(np.array(kp).shape)# 繪制特征點
cv2.drawKeypoints(img, kp, key_points, flags=cv2.DRAW_MATCHES_FLAGS_NOT_DRAW_SINGLE_POINTS)# 圖片展示
cv2.imshow("key points", key_points)
cv2.waitKey(0)
cv2.destroyAllWindows()# 保存圖片
# cv2.imwrite("key_points.jpg", key_points)# 計算特征
kp, des = sift.compute(img, kp)# 調試輸出
print(des.shape)
print(des[0])cv2.imshow('kp', key_points)
cv2.waitKey(0)
cv2.destroyAllWindows()

SURF(Speeded-Up Robust Features)

• 文章前面介紹了使用 SIFT 算法進行關鍵點檢測和描述。但是這種算法的執行速度比較慢，人們需要速度更快的算法。在 2006 年Bay,H.,Tuytelaars,T. 和 Van Gool,L 共同提出了 SURF（加速穩健特征）算法。跟它的名字一樣，這是個算法是加速版的 SIFT。
• 與 SIFT 相同 OpenCV 也提供了 SURF 的相關函數。首先我們要初始化一個 SURF 對象，同時設置好可選參數： 64/128 維描述符， Upright/Normal 模式等。所有的細節都已經在文檔中解釋的很明白了。就像我們在SIFT 中一樣，我們可以使用函數 SURF.detect()， SURF.compute() 等來進行關鍵點攙著和描述。

img = cv2.imread(‘fly.png’, 0)
surf = cv2.SURF(400)
kp, des = surf.detectAndCompute(img, None)
len(kp) # 699
print(surf.hessianThreshold)
surf.hessianThreshold = 50000
kp, des = surf.detectAndCompute(img,None)
print(len(kp)) # 47
不檢測關鍵點的方向
print(surf.upright) #False
surf.upright = True

FAST算法

import numpy as np
import cv2
from matplotlib import pyplot as pltimg = cv2.imread('./resource/opencv/image/fly.jpg', cv2.IMREAD_GRAYSCALE)# fast = cv2.FastFeatureDetector_create(threshold=100, nonmaxSuppression=False, type=cv2.FAST_FEATURE_DETECTOR_TYPE_5_8)
fast = cv2.FastFeatureDetector_create(threshold=400)
kp = fast.detect(img, None)
img2 = cv2.drawKeypoints(img, kp, img.copy(), color=(0, 0, 255), flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)cv2.imshow('fast', img2)
cv2.waitKey(0)
cv2.destroyAllWindows()

BRIEF(Binary Robust Independent Elementary Features)算法

• BRIEF(Binary Robust Independent Elementary Features)

import numpy as np
import cv2
from matplotlib import pyplot as pltimg = cv2.imread('./resource/opencv/image/fly.jpg', cv2.IMREAD_GRAYSCALE)# Initiate STAR detector
star = cv2.FeatureDetector_create("STAR")
# Initiate BRIEF extractor
brief = cv2.DescriptorExtractor_create("BRIEF")
# find the keypoints with STAR
kp = star.detect(img,None)
# compute the descriptors with BRIEF
kp, des = brief.compute(img, kp)
print(brief.getInt('bytes'))
print(des.shape)

ORB (Oriented FAST and Rotated BRIEF)算法

import numpy as np
import cv2
from matplotlib import pyplot as pltimg = cv2.imread('./resource/opencv/image/fly.jpg', cv2.IMREAD_GRAYSCALE)# ORB_create(nfeatures=..., scaleFactor=..., nlevels=..., edgeThreshold=..., firstLevel=..., WTA_K=..., scoreType=..., patchSize=..., fastThreshold=...)
orb = cv2.ORB_create()kp = orb.detect(img, None)kp, des = orb.compute(img, kp)img2 = cv2.drawKeypoints(img, kp, img.copy(), color=(255, 0, 0), flags=0)
plt.imshow(img2)
plt.show()

特征匹配

OpenCV 中的特征匹配

• 蠻力（ Brute-Force）匹配
• FLANN 匹配

Brute-Force 蠻力匹配

對 ORB 描述符進行蠻力匹配

import numpy as np
import cv2
from matplotlib import pyplot as pltimg1 = cv2.imread('./resource/opencv/image/box.png', 0)
img2 = cv2.imread('./resource/opencv/image/box_in_scene.png', 0)orb = cv2.ORB_create()kp1, des1 = orb.detectAndCompute(img1, None)
kp2, des2 = orb.detectAndCompute(img2, None)bf = cv2.BFMatcher_create(cv2.NORM_HAMMING, crossCheck=True)matches = bf.match(des1, des2)# matches = bf:match(des1; des2) 返回值是一個 DMatch 對象列表。這個
# DMatch 對象具有下列屬性：
# ? DMatch.distance - 描述符之間的距離。越小越好。
# ? DMatch.trainIdx - 目標圖像中描述符的索引。
# ? DMatch.queryIdx - 查詢圖像中描述符的索引。
# ? DMatch.imgIdx - 目標圖像的索引。# 距離排序
matches = sorted(matches, key = lambda x:x.distance)# 畫出前30匹配
img3 = cv2.drawMatches(img1, kp1, img2, kp2, matches[:30], None, flags=2)cv2.imshow('img', img3)
cv2.waitKey(0)
cv2.destroyAllWindows()

對 SIFT 描述符進行蠻力匹配和比值測試

現在我們使用 BFMatcher.knnMatch() 來獲得 k 對最佳匹配。在本例中我們設置 k = 2，這樣我們就可以使用 D.Lowe 文章中的比值測試了。

import numpy as np
import cv2
from matplotlib import pyplot as pltimg1 = cv2.imread('./resource/opencv/image/box.png', 0)
img2 = cv2.imread('./resource/opencv/image/box_in_scene.png', 0)sift = cv2.SIFT_create()
kp1, des1 = sift.detectAndCompute(img1, None)
kp2, des2 = sift.detectAndCompute(img2, None)bf = cv2.BFMatcher_create()
matches = bf.knnMatch(des1, des2, k=2)good = []
for m,n in matches:if m.distance < 0.75*n.distance:good.append([m])# drawMatchesKnn(img1, keypoints1, img2, keypoints2, matches1to2, outImg, matchColor=..., singlePointColor=..., matchesMask=..., flags: int = ...)
img3 = cv2.drawMatchesKnn(img1, kp1, img2, kp2, good[:100], None, flags=2)
plt.imshow(img3)
plt.show()

FLANN 匹配

FLANN 是快速最近鄰搜索包（Fast_Library_for_Approximate_Nearest_Neighbors）的簡稱。它是一個對大數據集和高維特征進行最近鄰搜索的算法的集合，而且這些算法都已經被優化過了。在面對大數據集時它的效果要好于 BFMatcher。我們來對第二個例子使用 FLANN 匹配看看它的效果。

import numpy as np
import cv2
from matplotlib import pyplot as pltimg1 = cv2.imread('./resource/opencv/image/box.png', 0)
img2 = cv2.imread('./resource/opencv/image/box_in_scene.png', 0)sift = cv2.SIFT_create()
kp1, des1 = sift.detectAndCompute(img1, None)
kp2, des2 = sift.detectAndCompute(img2, None)flann = cv2.FlannBasedMatcher_create()
matches = flann.knnMatch(des1, des2, k=2)matchesMask = [[0,0] for i in range(len(matches))]for i, (m, n) in enumerate(matches):if m.distance < 0.7*n.distance:matchesMask[i] = [1,0]draw_params = dict(matchColor = (0, 255, 0),singlePointColor = (255, 0, 0),matchesMask = matchesMask,flags = 0)img3 = cv2.drawMatchesKnn(img1, kp1, img2, kp2, matches, None, **draw_params)
plt.imshow(img3)
plt.show()