随机生成数组，使用scipy中的io.savemat()保存
文件格式是.mat，标准的二进制文件

---

import scipy.io as io

保存

# result是图片的ndarray
io.savemat('./moon_clean.mat',mdict = {'moon':np.real(result)})

使用io.loadmat()读取数据

io.loadmat('./moon_clean.mat')['moon']

47.png

---

读写图片使用scipy中misc.imread()/imsave()

from scipy import misc

import warnings

warnings.filterwarnings('ignore')

cat = misc.imread('./cat.jpg')
plt.imshow(cat)

5.04.png

misc旋转、resize、imfilter

cat2 = misc.imrotate(cat,90)
plt.imshow(cat2)

19.png

'''size : int, float or tuple
        * int   - Percentage of current size.
        * float - Fraction of current size.
        * tuple - Size of the output image (height, width).
'''
cat3 = misc.imresize(cat,0.2)
# cat4 = misc.imresize(cat,(200,200))
plt.imshow(cat4)
cat3.shape
plt.imshow(cat3)

8.png

'''ftype : str
        The filter that has to be applied. Legal values are:
        'blur', 'contour', 'detail', 'edge_enhance', 'edge_enhance_more',
        'emboss', 'find_edges', 'smooth', 'smooth_more', 'sharpen'.'''
# 原理：卷积操作
# 卷积神经网络tensorflow
cat6 = misc.imfilter(cat,'find_edges')
plt.imshow(cat6)

2.png

音乐处理

11.png

from scipy.io import wavfile

like = wavfile.read('./邓紫棋-喜欢你.wav')
like

like[1].shape

out:(10776444, 2)
歌曲秒数时间

like[1].shape[0]/44100

out244.36380952380952

231.png

lost = wavfile.read('./林俊杰-爱不会绝迹.wav')
lost

2323.png

歌曲拼接，前一分钟和后一分钟

mix = np.concatenate([like[1][:44100*60],lost[1][-44100*60:]])

mix.shape

out(5292000, 2)

保存文件

wavfile.write('./mix.wav',44100,mix)

scipy不能处理mp3

pip install pydub

42.png

读取Mp3

jing = pydub.AudioSegment.from_mp3('./难念的经.mp3')
jing

you = pydub.AudioSegment.from_mp3('./一路上有你.mp3')
you

jing[:1000*30] + you[:1000*30]

data = you.get_array_of_samples()

type(data)

array.array

data_raw = np.array(data.tolist())
data_raw.shape

(25337088,)

# 此数据就是wav格式用到的数据
data_raw = data_raw.reshape(-1,2)
data_raw.shape

(12668544, 2)

you.frame_rate

wavfile.write('./一路有你.wav',44100,data_raw.astype(np.int16))

使用ffmpeg操作视频

# ffmpeg -i law.mp4 -f mp3 -vn law.mp3
# 获取音频
import subprocess

cmd = 'ffmpeg -i law.mp4 -f mp3 -vn law2.mp3'
# 执行成功，subprocess返回结果是0
subprocess.call(cmd)

# 从视频中获取无声视频
cmd = 'ffmpeg -i 1.mp4 -vcodec copy -an 2.mp4'

subprocess.call(cmd)

# 合并视频和音频
cmd = 'ffmpeg -i law.mp3  -i 2.mp4 out.mp4'

subprocess.call(cmd)

图片处理

使用scipy.misc.face(gray=True)获取图片，使用ndimage移动坐标、旋转图片、切割图片、缩放图片
导包，读取图片显示图片

import numpy as np
# 移动坐标、旋转图片、切割图片、缩放图片、滤波（不清晰的图片，清晰化，取出‘噪声’）
from scipy import ndimage

from scipy import misc

import matplotlib.pyplot as plt
%matplotlib inline

# misc模块中，有一张图片
# 样本图片
face = misc.face()
plt.imshow(face)

3.png

shift移动坐标

face.shape

(768, 1024, 3)

# jpg 0~255 0:纯黑，255纯白
face2 = ndimage.shift(face,shift=[100,200,0],cval=255)
plt.imshow(face2)

35.png

# mode : {'reflect', 'constant', 'nearest', 'mirror', 'wrap'
# mirror reflect 镜像
# nearest 最近的像素进行填充
# wrap使用移出去的像素进行填充
face3 = ndimage.shift(face,shift=[0,-400,0],mode='wrap')
plt.imshow(face3)

we.png

rotate旋转图片

face4 = ndimage.rotate(face,angle = 60)
plt.imshow(face4)

21.png

face5 = misc.imrotate(face,60)
plt.imshow(face5)

3243.png

zoom缩放图片

# zoom : float or sequence
face6 = ndimage.zoom(face,zoom = 1/3)
face6.shape
# 合法：黑白二维；彩色三维（第三维，三个数RGB，四个数ARGB）
plt.imshow(face6.reshape(256,-1),cmap = plt.cm.gray)

2145.png

# zoom : float or sequence
face6 = ndimage.zoom(face,zoom = 4/3)
face6.shape
# 合法：黑白二维；彩色三维（第三维，三个数RGB，四个数ARGB）
# plt.imshow(face6.reshape(256,-1),cmap = plt.cm.gray)

(1024, 1365, 4)

face6[:,:,0] = 0.1
plt.imshow(face6)

dsfsdf.png

face6

47.png

# jpg类型透明度：0 ~ 255
# png类型图片 0 ~ 1
# 彩色图片三维，第三维四个数据：RGBA（alpha，透明）
face6[:,:,3] = 20
plt.imshow(face6)
# 机器学习：图片辨识、语音识别、

42.png

face.shape

(768, 1024, 3)

face6 = ndimage.zoom(face,zoom = [200/768,300/1024,1])
face6.shape
plt.imshow(face6)

54.png

切割图片

# numpy切片操作
face7 = face[:512,-512:]
plt.imshow(face7)

02.png

图片进行过滤
添加噪声，对噪声图片使用ndimage中的高斯滤波、中值滤波、signal中维纳滤波进行处理
使图片变清楚

moon = plt.imread('./moonlanding.png')
plt.figure(figsize=(12,9))
plt.imshow(moon,cmap = 'gray')

11.png

# sigma高斯核，高斯分布，正态分布，sigma离散程度
moon2  = ndimage.gaussian_filter(moon,sigma=2)
plt.figure(figsize=(12,9))
plt.imshow(moon2,cmap = 'gray')

55.png

moon2 = ndimage.median_filter(moon,size = 5)
plt.figure(figsize=(12,9))
plt.imshow(moon2,cmap = 'gray')

09.png

moon2 = ndimage.median_filter(moon,size = 10)
plt.figure(figsize=(12,9))
plt.imshow(moon2,cmap = 'gray')

9.png

moon2 = ndimage.median_filter(moon,size = 50)
plt.figure(figsize=(12,9))
plt.imshow(moon2,cmap = 'gray')

qw32.png

signal维纳滤波mysize：滤镜尺寸的标量

from scipy import signal

moon3 = signal.wiener(moon,mysize=5)
plt.figure(figsize=(12,9))
plt.imshow(moon3,cmap = 'gray')

xcb10.png

moon3 = signal.wiener(moon,mysize=20)
plt.figure(figsize=(12,9))
plt.imshow(moon3,cmap = 'gray')
# 

jhgj46.png