李新乾的个人博客
0%

OpenCV(七)-滤波器

发表于 更新于 阅读次数： 1

一、图像滤波

滤波的作用：
一副图像==通过滤波器==得到==另一幅图像==
其中滤波器又称为==卷积核==，滤波的过程称为==卷积==

image-20241107155955301

image-20241107160007913

二、卷积相关概念

2.1、卷积核的大小

卷积核一般为==奇数==，如3x3,5x5,7x7等
一方面是增加==padding==的原因
另一方面是保证==锚点==在中间，防止位置发生偏移

在深度学习中，卷积核越大，看到的信息（==感受野==）越多，提取的==特征越好==，同时计算量也就越大

2.2、锚点

image-20241107160751205

2.3、边界扩充

当卷积核大于1且不进行边界扩充，输出尺寸将相应缩小

当卷积核以标准方式进行边界扩充，则输出数据的空间大小与输入相等

image-20241107160906228

image-20241107160916733

2.4、步长

image-20241107160945609

步长为2

三、实战图像卷积

image-20241107161122621

1
2
3
4
5
6
7
8
9
10
(function)
def filter2D(
    src: MatLike,
    ddepth: int,	# 位深
    kernel: MatLike,	# 卷积核
    dst: MatLike | None = ...,
    anchor: Point = ...,	# 锚点
    delta: float = ...,
    borderType: int = ...
) -> MatLike: ...

image-20241107162224860

1
2
3
4
5
6
7
8
9
10
11
12
import cv2
import numpy as np

img = cv2.imread('dog.jpeg')

kernal = np.ones((5,5), np.float32) / 25

dst = cv2.filter2D(img, -1, kernal)

cv2.imshow('dst', dst)
cv2.imshow('img', img)
cv2.waitKey(0)

可以看出滤波之后，线条变模糊了，眼睛变暗了

image-20241107162500753

四、方盒滤波与均值滤波

image-20241107162849123

image-20241107162945808

image-20241107162952541

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
(function)
def boxFilter(
    src: MatLike,
    ddepth: int,
    ksize: Size,
    dst: MatLike | None = ...,
    anchor: Point = ...,
    normalize: bool = ...,
    borderType: int = ...
) -> MatLike: ...

(function) def blur(
    src: MatLike,
    ksize: Size,
    dst: MatLike | None = ...,
    anchor: Point = ...,
    borderType: int = ...
) -> MatLike
1
2
3
4
5
6
7
8
9
10
11
12
import cv2
import numpy as np

img = cv2.imread('dog.jpeg')

dst = cv2.blur(img, (5,5))

cv2.imshow('dst', dst)
cv2.imshow('img', img)
cv2.waitKey(0)

# 效果图和上一节一样，均值滤波

五、高斯滤波

image-20241107163554151

image-20241107163625762

image-20241107163638106

中间比重高，四边权重低

1
2
3
4
5
6
7
8
9
10
def GaussianBlur(
    src: MatLike,
    ksize: Size,
    sigmaX: float,	# 越大越模糊
    dst: MatLike | None = ...,
    sigmaY: float = ...,
    borderType: int = ...
) -> MatLike: ...
    
# 高斯滤波解决的是噪点
1
2
3
4
5
6
7
8
9
10
import cv2
import numpy as np

img = cv2.imread('./gaussian.png')

dst = cv2.GaussianBlur(img, (5,5), sigmaX=1)

cv2.imshow('dst', dst)
cv2.imshow('img', img)
cv2.waitKey(0)

可以看出右侧的噪点少了很多，但是边缘也变模糊了

image-20241107164528763

六、中值滤波

image-20241107164709281

优点：对==胡椒噪音==效果明显

1
2
3
4
5
def medianBlur(
    src: MatLike,
    ksize: int,
    dst: MatLike | None = ...
) -> MatLike: ...
1
2
3
4
5
6
7
8
9
10
import cv2
import numpy as np

img = cv2.imread('./papper.png')

dst = cv2.medianBlur(img, 5)

cv2.imshow('dst', dst)
cv2.imshow('img', img)
cv2.waitKey(0)

image-20241107165121046

七、双边滤波

image-20241107165219906

image-20241107165227513

image-20241107165309701

1
2
3
4
5
6
7
8
9
(function)
def bilateralFilter(
    src: MatLike,
    d: int,			# 直径，大小
    sigmaColor: float,
    sigmaSpace: float,
    dst: MatLike | None = ...,
    borderType: int = ...
) -> MatLike: ...
1
2
3
4
5
6
7
8
9
10
import cv2
import numpy as np

img = cv2.imread('./lena.png')

dst = cv2.bilateralFilter(img, 7, 20, 50)

cv2.imshow('dst', dst)
cv2.imshow('img', img)
cv2.waitKey(0)

可以看出，帽子这些颜色比较深，看作了边缘，保留的很好，
但是脸这里进行了优化

image-20241107165725441

八、高通滤波

8.1、Sobel（索贝尔）算子

image-20241107170107968

1
2
3
4
5
6
7
8
9
10
11
12
13
(function) def Sobel(
    src: MatLike,
    ddepth: int,	# 输出图像的深度（通常为 CV_8U 或 CV_64F）
    dx: int,		# x 方向的阶数（指定对图像在 x 方向的微分次数）
    dy: int,		# y 方向的阶数（指定对图像在 y 方向的微分次数）
    dst: MatLike | None = ...,
    ksize: int = ...,	# Sobel 核的大小，必须是 1, 3, 5, 7 等奇数
    scale: float = ...,
    delta: float = ...,
    borderType: int = ...
) -> MatLike

# 主要是用于边缘
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
import cv2
import numpy as np

img = cv2.imread('./chess.png')

# 索贝尔算子y方向边缘
d1 = cv2.Sobel(img, cv2.CV_64F, 1, 0, ksize=5)
# 索贝尔算子x方向边缘
d2 = cv2.Sobel(img, cv2.CV_64F, 0, 1, ksize=5)

dst = cv2.add(d1, d2)

cv2.imshow('img', img)
cv2.imshow('d1', d1)
cv2.imshow('d2', d2)
cv2.imshow('dst', dst)
cv2.waitKey(0)

image-20241107172706614

8.2、Scharr（沙尔）算子

image-20241107172817277

image-20241107172858112

image-20241107172925261

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
import cv2
import numpy as np

img = cv2.imread('./chess.png')

# 沙尔算子y方向边缘
d1 = cv2.Scharr(img, cv2.CV_64F, 1, 0)
# 沙尔算子x方向边缘
d2 = cv2.Scharr(img, cv2.CV_64F, 0, 1)

dst = cv2.add(d1, d2)

cv2.imshow('img', img)
cv2.imshow('d1', d1)
cv2.imshow('d2', d2)
cv2.imshow('dst', dst)
cv2.waitKey(0)

# 沙尔比较少用，在比较细小的边缘，索贝尔不行，它可以

image-20241107173136905

8.3、Laplacian（拉普拉斯）算子

image-20241107173307166

image-20241107173319212

1
2
3
4
5
6
7
8
9
10
11
import cv2
import numpy as np

img = cv2.imread('./chess.png')

# 拉普拉斯
ldst = cv2.Laplacian(img, cv2.CV_64F, ksize=5)

cv2.imshow('img', img)
cv2.imshow('ldst', ldst)
cv2.waitKey(0)

image-20241107173555100

九、边缘检测Canny

​ image-20241108090807306

image-20241108090830529

1
2
3
4
5
6
7
8
(function) def Canny(
    image: MatLike,
    threshold1: float,
    threshold2: float,
    edges: MatLike | None = ...,
    apertureSize: int = ...,
    L2gradient: bool = ...
) -> MatLike
1
2
3
4
5
6
7
8
9
10
import cv2
import numpy as np

img = cv2.imread('./lena.png')

dst = cv2.Canny(img, 100, 200)

cv2.imshow('img', img)
cv2.imshow('dst', dst)
cv2.waitKey(0)

image-20241108093520473

-------------本文结束感谢您的阅读-------------