Skip to content

Commit cb86df8

Browse files
makelovemakelove
committed
update ch37 data
1 parent 7057efc commit cb86df8

File tree

129 files changed

+26308
-15
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

129 files changed

+26308
-15
lines changed

‎ch32-介绍SIFT/sift.py‎

Lines changed: 31 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,31 @@
1+
# -*- coding: utf-8 -*-
2+
# @Time : 2017/7/13 下午2:23
3+
# @Author : play4fun
4+
# @File : sift.py
5+
# @Software: PyCharm
6+
7+
"""
8+
sift.py:尺度不变特征变换
9+
10+
关键点 极值点 定位
11+
12+
"""
13+
14+
import cv2
15+
import numpy as np
16+
17+
img = cv2.imread('../data/home.jpg')
18+
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
19+
20+
sift = cv2.xfeatures2d.SIFT_create()
21+
kp = sift.detect(gray, None)
22+
img = cv2.drawKeypoints(gray, kp, img)
23+
24+
# 计算关键点描述符
25+
# 使用函数 sift.compute() 来 计算 些关键点的描述符。例如
26+
# kp, des = sift.compute(gray, kp)
27+
kp, des = sift.detectAndCompute(gray,None)
28+
29+
cv2.imwrite('sift_keypoints.jpg', img)
30+
cv2.imshow('sift_keypoints.jpg', img)
31+
cv2.waitKey(0)

‎ch33-介绍SURF/surf.py‎

Lines changed: 57 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,57 @@
1+
# -*- coding: utf-8 -*-
2+
# @Time : 2017/7/13 下午3:13
3+
# @Author : play4fun
4+
# @File : surf.py
5+
# @Software: PyCharm
6+
7+
"""
8+
surf.py:SURF加速稳健特征，加速版的SIFT
9+
10+
积分图像的一大特点是 计算图像中某个窗 口内所有像素和时，计算量的大小与窗口大小无关
11+
12+
"""
13+
import cv2
14+
import matplotlib.pyplot as plt
15+
16+
img = cv2.imread('fly.png', 0)
17+
# Create SURF object. You can specify params here or later. # Here I set Hessian Threshold to 400
18+
# surf = cv2.SURF(400)
19+
surf = cv2.xfeatures2d.SURF_create(400)
20+
21+
# Find keypoints and descriptors directly
22+
kp, des = surf.detectAndCompute(img, None)
23+
len(kp) # 699
24+
25+
# 提高Hessian 的阈值
26+
# Check present Hessian threshold
27+
print(surf.getHessianThreshold())
28+
# 400.0
29+
# We set it to some 50000. Remember, it is just for representing in picture.
30+
# In actual cases, it is better to have a value 300-500
31+
surf.setHessianThreshold(50000)
32+
# Again compute keypoints and check its number.
33+
kp, des = surf.detectAndCompute(img, None)
34+
print(len(kp))
35+
# 47
36+
img2 = cv2.drawKeypoints(img, kp, None, (255, 0, 0), 4)
37+
38+
plt.imshow(img2)
39+
plt.show()
40+
41+
# 最后我们再看看关键点描述符的大小 如果是 64 维的就改成 128 维。
42+
43+
# Find size of descriptor
44+
print(surf.descriptorSize())
45+
# 64
46+
# That means flag, "extended" is False.
47+
print(surf.getExtended())
48+
# False
49+
# So we make it to True to get 128-dim descriptors.
50+
surf.extended = True
51+
kp, des = surf.detectAndCompute(img, None)
52+
print(surf.descriptorSize())
53+
# 128
54+
print(des.shape)
55+
# (47, 128)
56+
57+
# 接下来要做的就是匹配了 我们会在后讨论。

‎ch34-角点检测的FAST算法/fast.py‎

Lines changed: 45 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,45 @@
1+
# -*- coding: utf-8 -*-
2+
# @Time : 2017/7/13 下午3:31
3+
# @Author : play4fun
4+
# @File : fast.py
5+
# @Software: PyCharm
6+
7+
"""
8+
fast.py:FAST 特征检测器
9+
10+
效果很好。但是从实时处理的角度来看
11+
这些算法 不够快。
12+
一个最好例子就是 SLAM 同步定位与地图构建
13+
移动机器人 它们的计算资源非常有限。
14+
15+
"""
16+
17+
import numpy as np
18+
import cv2
19+
from matplotlib import pyplot as plt
20+
21+
img = cv2.imread('simple.jpg', 0)
22+
23+
# Initiate FAST object with default values
24+
fast = cv2.FastFeatureDetector_create()
25+
# find and draw the keypoints
26+
kp = fast.detect(img, None)
27+
img2 = cv2.drawKeypoints(img, kp, None, color=(255, 0, 0))
28+
29+
# Print all default params
30+
print("Threshold: ", fast.getThreshold())
31+
print("nonmaxSuppression: ", fast.getNonmaxSuppression())
32+
print("neighborhood: ", fast.getType())
33+
print("Total Keypoints with nonmaxSuppression: ", len(kp))
34+
cv2.imwrite('fast_true.png', img2)
35+
36+
# Disable nonmaxSuppression
37+
fast.setNonmaxSuppression(0)
38+
kp = fast.detect(img, None)
39+
print("Total Keypoints without nonmaxSuppression: ", len(kp))
40+
41+
img3 = cv2.drawKeypoints(img, kp, None, color=(255, 0, 0))
42+
cv2.imwrite('fast_false.png', img3)
43+
44+
#结果如下。第一幅图是使用了 非最大值抑制的结果
45+
# 第二幅没有使用非最大值抑制。

‎ch35-BRIEF/brief.py‎

Lines changed: 35 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,35 @@
1+
# -*- coding: utf-8 -*-
2+
# @Time : 2017/7/13 下午3:38
3+
# @Author : play4fun
4+
# @File : brief.py
5+
# @Software: PyCharm
6+
7+
"""
8+
brief.py:
9+
算法使用的是已经平滑后的图像
10+
11+
非常重要的一点是 BRIEF 是一种特征描述符 它不提供查找特征的方法。 所以我们不得不使用其他特征检测器 比如 SIFT 和 SURF 等。
12+
原始文献推荐 使用 CenSurE 特征检测器 种算法很快。而且 BRIEF 算法对 CenSurE 关键点的描述效果 比 SURF 关键点的描述更好。
13+
14+
简单来 BRIEF 是一种对特征点描述符运算和匹 的快速方法。 这种算法可以实现很高的识别率，除非出现平面内的大旋 。
15+
16+
"""
17+
18+
import numpy as np
19+
import cv2
20+
from matplotlib import pyplot as plt
21+
22+
img = cv2.imread('simple.jpg', 0)
23+
24+
# Initiate FAST detector
25+
star = cv2.xfeatures2d.StarDetector_create()
26+
# Initiate BRIEF extractor
27+
brief = cv2.xfeatures2d.BriefDescriptorExtractor_create()
28+
29+
# find the keypoints with STAR
30+
kp = star.detect(img, None)
31+
# compute the descriptors with BRIEF
32+
kp, des = brief.compute(img, kp)
33+
34+
print(brief.descriptorSize())
35+
print(des.shape)

‎ch36-ORB/orb.py‎

Lines changed: 43 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,43 @@
1+
# -*- coding: utf-8 -*-
2+
# @Time : 2017/7/13 下午3:45
3+
# @Author : play4fun
4+
# @File : orb.py
5+
# @Software: PyCharm
6+
7+
"""
8+
orb.py:
9+
SIFT 和 SURF 算法是有专利保护的 如果你 使用它们 就可能要花钱 。但是 ORB 不需要
10+
11+
ORB 基本是 FAST 关 点检测和 BRIEF 关 点描 器的结合体 并
12+
很多修改增强了性能。 先它使用 FAST 找到关 点 然后再使用 Harris 点检测对 些关 点 排序找到其中的前 N 个点。它也使用 字塔从而产 生尺度不变性特征。
13+
14+
数据的方差大的一个好处是 使得特征更容易分辨。
15+
16+
对于描述符，ORB使用BRIEF描述符。但我们已经看到，这个BRIEF的表现在旋转方面表现不佳。因此，ORB所做的是根据关键点的方向来“引导”。
17+
对于在位置(xi，yi)的n个二进制测试的任何特性集，定义一个包含这些像素坐标的2 n矩阵。然后利用补丁的方向，找到旋转矩阵并旋转S，以得到引导(旋转)版本s。
18+
19+
ORB将角度进行离散化，以增加2/30(12度)，并构造一个预先计算过的简短模式的查找表。只要键点的方向是一致的，就会使用正确的点集来计算它的描述符。
20+
21+
BRIEF有一个重要的属性，即每个比特的特性都有很大的方差，而平均值接近0.5。但是一旦它沿着键点方向移动，它就会失去这个属性并变得更加分散。高方差使特征更有区别，因为它对输入的响应不同。另一个可取的特性是让测试不相关，因为每个测试都将对结果有所贡献。为了解决所有这些问题，ORB在所有可能的二进制测试中运行一个贪婪的搜索，以找到那些既有高方差又接近0.5的，同时又不相关的。结果被称为rBRIEF。
22+
23+
对于描述符匹配，在传统的LSH上改进的多探测LSH是被使用的。这篇文章说，ORB比冲浪快得多，而且比冲浪还好。对于全景拼接的低功率设备，ORB是一个不错的选择。
24+
25+
"""
26+
27+
import numpy as np
28+
import cv2
29+
from matplotlib import pyplot as plt
30+
31+
img = cv2.imread('simple.jpg', 0)
32+
33+
# Initiate ORB detector
34+
orb = cv2.ORB_create()
35+
# find the keypoints with ORB
36+
kp = orb.detect(img, None)
37+
# compute the descriptors with ORB
38+
kp, des = orb.compute(img, kp)
39+
40+
# draw only keypoints location,not size and orientation
41+
img2 = cv2.drawKeypoints(img, kp, None, color=(0, 255, 0), flags=0)
42+
43+
plt.imshow(img2), plt.show()

‎ch37-特征匹配/37.2-对ORB描述符进行蛮力匹配.py‎

Lines changed: 41 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,41 @@
1+
# -*- coding: utf-8 -*-
2+
# @Time : 2017/7/13 下午4:00
3+
# @Author : play4fun
4+
# @File : 37.2-对ORB描述符进行蛮力匹配.py
5+
# @Software: PyCharm
6+
7+
"""
8+
37.2-对ORB描述符进行蛮力匹配.py:
9+
匹配器对象是什么
10+
matches = bf.match(des1, des2) 返回值是一个 DMatch对象列表
11+
DMatch 对 具有下列属性
12+
• DMatch.distance - 描 符之 的 离。 小 好。
13+
• DMatch.trainIdx - 目标图像中描 符的索引。
14+
• DMatch.queryIdx - 查 图像中描 符的索引。
15+
• DMatch.imgIdx - 目标图像的索引。
16+
"""
17+
18+
import numpy as np
19+
import cv2
20+
import matplotlib.pyplot as plt
21+
22+
img1 = cv2.imread('box.png', 0) # queryImage
23+
img2 = cv2.imread('box_in_scene.png', 0) # trainImage
24+
25+
# Initiate ORB detector
26+
orb = cv2.ORB_create()
27+
# find the keypoints and descriptors with ORB
28+
kp1, des1 = orb.detectAndCompute(img1, None)
29+
kp2, des2 = orb.detectAndCompute(img2, None)
30+
31+
# create BFMatcher object
32+
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=True)
33+
# Match descriptors.
34+
matches = bf.match(des1, des2)
35+
36+
# Sort them in the order of their distance.
37+
matches = sorted(matches, key=lambda x: x.distance)
38+
# Draw first 10 matches.
39+
img3 = cv2.drawMatches(img1, kp1, img2, kp2, matches[:10], flags=2) # 前10个匹配
40+
41+
plt.imshow(img3), plt.show()

‎ch37-特征匹配/37.4-SIFT_match.py‎

Lines changed: 21 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -1,38 +1,44 @@
1-
#-*-coding:utf8-*-#
1+
# -*-coding:utf8-*-#
22
__author__ = 'play4fun'
33
"""
44
create time:15-11-9 下午1:28
5+
对 SIFT 描述符进行蛮力匹配和比值测试
56
"""
67

7-
8-
98
import numpy as np
109
import cv2
1110
from matplotlib import pyplot as plt
12-
img1 = cv2.imread('../data/box.png',0)
11+
12+
img1 = cv2.imread('../data/box.png', 0)
1313
# queryImage
14-
img2 = cv2.imread('../data/box_in_scene.png',0) # trainImage
14+
img2 = cv2.imread('../data/box_in_scene.png', 0) # trainImage
1515
# Initiate SIFT detector
1616
# sift = cv2.SIFT()
1717
sift = cv2.xfeatures2d.SIFT_create()
18-
#TODO
18+
1919

2020
# find the keypoints and descriptors with SIFT
21-
kp1, des1 = sift.detectAndCompute(img1,None)
22-
kp2, des2 = sift.detectAndCompute(img2,None)
21+
kp1, des1 = sift.detectAndCompute(img1, None)
22+
kp2, des2 = sift.detectAndCompute(img2, None)
23+
2324
# BFMatcher with default params
2425
bf = cv2.BFMatcher()
25-
matches = bf.knnMatch(des1,des2, k=2)
26+
matches = bf.knnMatch(des1, des2, k=2)
27+
2628
# Apply ratio test
2729
# 比值测试，首先获取与 A距离最近的点 B （最近）和 C （次近），
2830
# 只有当 B/C 小于阀值时（0.75）才被认为是匹配，
29-
#因为假设匹配是一一对应的，真正的匹配的理想距离为0
31+
# 因为假设匹配是一一对应的，真正的匹配的理想距离为0
3032
good = []
31-
for m,n in matches:
32-
if m.distance < 0.75*n.distance:
33+
for m, n in matches:
34+
if m.distance < 0.75 * n.distance:
3335
good.append([m])
3436

3537
# cv2.drawMatchesKnn expects list of lists as matches.
36-
img3=np.ndarray([2,2])
37-
img3 = cv2.drawMatchesKnn(img1,kp1,img2,kp2,good[:10],img3,flags=2)
38-
plt.imshow(img3),plt.show()
38+
# img3 = np.ndarray([2, 2])
39+
# img3 = cv2.drawMatchesKnn(img1, kp1, img2, kp2, good[:10], img3, flags=2)
40+
41+
# cv2.drawMatchesKnn expects list of lists as matches.
42+
img3 = cv2.drawMatchesKnn(img1,kp1,img2,kp2,good,flags=2)
43+
44+
plt.imshow(img3), plt.show()

‎ch37-特征匹配/37.5-FLANN匹配器.py‎

Lines changed: 55 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,55 @@
1+
# -*- coding: utf-8 -*-
2+
# @Time : 2017/7/13 下午4:22
3+
# @Author : play4fun
4+
# @File : 37.5-FLANN匹配器.py
5+
# @Software: PyCharm
6+
7+
"""
8+
37.5-FLANN匹配器.py:
9+
FLANN 是快速最近邻搜索包 Fast_Library_for_Approximate_Nearest_Neighbors 的简称。
10+
它是一个对大数据集和高维特征进行最近邻搜索的算法的集合
11+
而且这些算法 已经被优化 了。
12+
在面对大数据集时它的效果 好于 BFMatcher
13+
14+
"""
15+
16+
17+
18+
import numpy as np
19+
import cv2
20+
from matplotlib import pyplot as plt
21+
22+
img1 = cv2.imread('box.png', 0) # queryImage
23+
img2 = cv2.imread('box_in_scene.png', 0) # trainImage
24+
25+
# Initiate SIFT detector
26+
# sift = cv2.SIFT()
27+
sift = cv2.xfeatures2d.SIFT_create()
28+
29+
# find the keypoints and descriptors with SIFT
30+
kp1, des1 = sift.detectAndCompute(img1, None)
31+
kp2, des2 = sift.detectAndCompute(img2, None)
32+
33+
# FLANN parameters
34+
FLANN_INDEX_KDTREE = 0
35+
index_params = dict(algorithm=FLANN_INDEX_KDTREE, trees=5)
36+
search_params = dict(checks=50) # or pass empty dictionary
37+
38+
flann = cv2.FlannBasedMatcher(index_params, search_params)
39+
matches = flann.knnMatch(des1, des2, k=2)
40+
# Need to draw only good matches, so create a mask
41+
matchesMask = [[0, 0] for i in range(len(matches))]
42+
43+
# ratio test as per Lowe's paper
44+
for i, (m, n) in enumerate(matches):
45+
if m.distance < 0.7 * n.distance:
46+
matchesMask[i] = [1, 0]
47+
48+
draw_params = dict(matchColor=(0, 255, 0),
49+
singlePointColor=(255, 0, 0),
50+
matchesMask=matchesMask,
51+
flags=0)
52+
53+
img3 = cv2.drawMatchesKnn(img1, kp1, img2, kp2, matches, None, **draw_params)
54+
55+
plt.imshow(img3, ), plt.show()

‎data/._messi5.jpg‎

-4 KB
Binary file not shown.

‎data/H1to3p.xml‎

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,11 @@
1+
<?xml version="1.0"?>
2+
<opencv_storage>
3+
<H13 type_id="opencv-matrix">
4+
<rows>3</rows>
5+
<cols>3</cols>
6+
<dt>d</dt>
7+
<data>
8+
7.6285898e-01 -2.9922929e-01 2.2567123e+02
9+
3.3443473e-01 1.0143901e+00 -7.6999973e+01
10+
3.4663091e-04 -1.4364524e-05 1.0000000e+00 </data></H13>
11+
</opencv_storage>

0 commit comments

Comments
 (0)