在《使用numpy处理图片——模糊处理》一文中，我们介绍了如何使用scipy库进行滤镜处理。本文我们将通过9宫格的形式，展现不同参数时滤镜效果。

black_tophat

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagegenerate('lena.png', 'black_tophat.png', ndimage.black_tophat, 1, 91, 10)

对应的size（ndimage.black_tophat第二个参数）的值

1	11	21
31	41	51
61	71	81

white_tophat

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagedef func(*args):return ndimage.white_tophat(args[0], args[1])generate('lena.png', 'white_tophat.png', func, 1, 91, 10)

对应的size（ndimage.white_tophat第二个参数）的值

1	11	21
31	41	51
61	71	81

convolve

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagedef func(*args):weights = np.eye(args[1])return ndimage.convolve(args[0], weights)generate('lena.png', 'convolve.png', func, 1, 10, 1)

对应的weights（ndimage.convolve第二个参数）的维度是

1	2	3
4	5	6
7	8	9

correlate

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagedef func(*args):weights = np.eye(args[1])return ndimage.correlate(args[0], weights)generate('lena.png', 'correlate.png', func, 1, 10, 1)

对应的weights（ndimage.correlate第二个参数）的维度是

1	2	3
4	5	6
7	8	9

gaussian_filter

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagegenerate('lena.png', 'gaussian_filter.png', ndimage.gaussian_filter, 1, 10, 1)

对应的sigma（ndimage.gaussian_filter第二个参数）的值

1	2	3
4	5	6
7	8	9

gaussian_laplace

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagegenerate('lena.png', 'gaussian_laplace.png', ndimage.gaussian_laplace, 0.2, 1.9, 0.2)

对应的sigma（ndimage.black_tophat第二个参数）的值

0.2	0.4	0.6
0.8	1.0	1.2
1.4	1.6	1.8

maximum_filter

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagegenerate('lena.png', 'maximum_filter.png', ndimage.maximum_filter, 1, 10, 1)

对应的size（ndimage.maximum_filter第二个参数）的值

1	2	3
4	5	6
7	8	9

median_filter

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagegenerate('lena.png', 'median_filter.png', ndimage.median_filter, 1, 10, 1)

对应的size（ndimage.median_filter第二个参数）的值

1	2	3
4	5	6
7	8	9

minimum_filter

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagegenerate('lena.png', 'minimum_filter.png', ndimage.minimum_filter, 1, 10, 1)

对应的size（ndimage.minimum_filter第二个参数）的值

1	2	3
4	5	6
7	8	9

percentile_filter

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagedef func(*args):return ndimage.percentile_filter(args[0], percentile=args[1], size=args[1])generate('lena.png', 'percentile_filter.png', func, 1, 10, 1)

对应的percentile和size（ndimage.percentile_filter第二、三个参数）的值

1	2	3
4	5	6
7	8	9

prewitt

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagedef func(*args):return ndimage.prewitt(args[0])generate('lena.png', 'prewitt.png', func, 1, 2, 1)

rank_filter

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagedef func(*args):return ndimage.rank_filter(args[0], rank=args[1], size=args[1]*2)generate('lena.png', 'rank_filter.png', func, 1, 10, 1)

对应的rank（ndimage.rank_filter第二个参数）的值

1	2	3
4	5	6
7	8	9

对应的size（ndimage.rank_filter第三个参数）的值

2	4	6
8	10	12
14	16	18

sobel

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagedef func(*args):return ndimage.sobel(args[0])generate('lena.png', 'sobel.png', func, 1, 2, 1)

spline_filter

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagedef func(*args):return ndimage.spline_filter(args[0], args[1]).astype(np.uint8)generate('lena.png', 'spline_filter.png', func, 2, 5, 1)

对应的size（ndimage.black_tophat第二个参数）的值

2	3	4

uniform_filter

import sys 
sys.path.append("..") 
from frame import *
import scipy.ndimage as ndimagedef func(*args):return ndimage.uniform_filter(args[0], args[1])generate('lena.png', 'uniform_filter.png', func, 1, 10, 1)

对应的size（ndimage.uniform_filter第二个参数）的值

1	2	3
4	5	6
7	8	9

基础代码

# frame.py
import numpy as np
from PIL import Image
import scipy.ndimage as ndimagedef generate(image_from, image_to, filter, start = 1, end = 10, step = 1):source = np.array(Image.open(image_from))colorDim3List = np.dsplit(source, 3)red = colorDim3List[0].reshape(source.shape[0], source.shape[1])green = colorDim3List[1].reshape(source.shape[0], source.shape[1])blue = colorDim3List[2].reshape(source.shape[0], source.shape[1])def inline_filter(red, green, blue, some_value):redFilter = filter(red, some_value)greenFilter = filter(green, some_value)blueFilter = filter(blue, some_value)return np.dstack((redFilter, greenFilter, blueFilter))varrays = []harrays = []hindex = 0for i in np.arange(start, end, step):filter3D = inline_filter(red, green, blue, i)harrays.append(filter3D)hindex += 1if hindex % 3 == 0:varrays.append(np.hstack(harrays))harrays = []hindex = 0if varrays == []:varrays.append(np.hstack(harrays))full3D = np.vstack(varrays)Image.fromarray(full3D).save(image_to)

代码仓库

https://github.com/f304646673/scipy-ndimage-example/tree/main/