jupyter

https://www.jianshu.com/p/17dabbc5936e(python 3.7)
mlTF: numpy tensorflow
ipython jupyter
opencv:https://www.pianshen.com/article/39091554701/

hello jupyter

https://blog.csdn.net/sinat_36502563/article/details/102302392

4.图片的几何变换

4.1图片的缩放

det. 是 determiner 的缩写,
1.调用API：
resize

2.算法：
2.1最近邻域插值法
原理

代码实现

2.2双线性插值法
原理

4.2图片移位

任务：读入一张图片，分别用api和原理实现图片位移

warpAffine
api算法分析
矩阵拆分、矩阵计算

#向左移动
import cv2
import numpy as np 
img= cv2.imread('1.png',1)
cv2.imshow('src',img)
imgInfo = img.shape
height = imgInfo[0]
width = imgInfo[1]
dst = np.zeros(img.shape,np.uint8)
for i in range(0,height):
    for j in range(0,width-100):
        dst[i,j+100] = img[i,j]
cv2.imshow('image',dst)
cv2.waitKey(0)

4.3图片镜像

import cv2
import numpy
img = imread('i.png')
imgInfo = img.shape 
height = imgInfo[0]
width = imgInfo[1]
# deep说明图片颜色的组成，RGB
deep = imgInfo[2]
newImgInfo = (height*2,width,deep)
dst = np.zeros(newImgInfo,np,uint8)
# 显示原图+镜像图片
for i in range(0,height):
    for j in range(0,width):
        dst[i,j]= img[i,j]
        #x不变  y=2*h-y-1         
        dst[height*2-i-1,j]=img[i,j]
# 画中间的那一条红线
for i in range(0,width):
    dst[height,i]=(0,0,255)#BGR
cv2.imshow('dst',dst)
cv2.waitKey(0)

小总结

for i,j就是原图片你想展现在新图片的部分
dst计算新图片相对于原图片的坐标

4.4图片缩放

用图片移位学的公式（api）将图片缩一半

import cv2
import numpy as np 
img = cv2.imread('1.png',1)
imgInfo = img.shape
height = imgInfo[0]
width = imgInfo[1]
matSacle = np.float32([[0.5,0,0],[0,0.5,0]])
dst = cv2.warpAffine(img,matScle,int(width/2),int(height/2))
cv2.imshow('dst',dst)
cv2.waitKey(0)

4.5图片的仿射变换

import cv2
import numpy as np
img= cv2.imread('i.png')
imgInfo = img.shape
height = imgInfo[0]
width = imgInfo[1]
#src 3->dst 3(左上角 左下角 右上角)
matSrc = np.float32([[0,0],[0,height-1],[width-1,0]])
matDst = np.float32([[50,50],[300,height-200],[width-300],100])
#矩阵组合
matAffine = cv2.getAffineTransform(matSrc,matDst) # mat 第1参数src 第2参数dst
dst = cv2.warpAffine(img,matAffine,(width,height))
cv2.imshow('dst',dst)
cv2.waitKey(0)

4.6图片旋转

#mat 1 center 2 angle 3 sc
#为什么要有缩放的系数(sc)：不然旋转超出范围了
matRotate = cv2.getRotationMatrix2D((height*0.5,width*0.5),45,0.5)
dst = cv2.warpAffine(img,matRotate,(height,width))

5图片特效

5.1灰度处理(最重要，基础，实时性)

1。imread

import cv2 
#0代表灰度图片 1代表彩色图片
img0 = cv2.imread('1.png',0)
img1 = cv2.imread('1.png',1)
print(img0.shape)
print(img1.shape)
cv2.imshow('src',img0)

2.cvtColor

import cv2
img = cv2.imread('1.png')
#颜色空间的转换 1data 2BGR gray
dst = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
cv2.imshow('dst',dst)

3.gray = (r+b+g)/3

import cv2
import numpy as np
img= cv2.imread('i.png')
imgInfo = img.shape
height = imgInfo[0]
width = imgInfo[1]
#RGB R=G=B =gray (R+G+B)/3
dst = np.zeros((height,width,3),np.uint8)
for i in range(0,height):
    for j in range(0,width):
        (b,g,r) = img[i,j]
        #uint8 进行相加计算可能会溢出，所以要转int
        gray = (int(b)+int(g)+int(r))/3
        dst[i,j] = np.uint8(gray)
cv2.imshow('dst',dst)
cv2.waitKey(0)

4.gray = r0.299+g0.587+b*0.114

import cv2
import numpy as np
img= cv2.imread('i.png')
imgInfo = img.shape
height = imgInfo[0]
width = imgInfo[1]
#RGB R=G=B =gray (R+G+B)/3
dst = np.zeros((height,width,3),np.uint8)
for i in range(0,height):
    for j in range(0,width):
        (b,g,r) = img[i,j]
        #uint8 进行相加计算可能会溢出，所以要转int
        gray = int(b)*0.114+int(g)*0.587+int(r)*0.299
        dst[i,j] = np.uint8(gray)
cv2.imshow('dst',dst)
cv2.waitKey(0)

4.算法优化（实时性）

定点-》浮点 +-/ >>

浮点运算转为定点运算（4再/4，即左移两位再右移两位）

定点运算转为移位运算

5.2颜色反转

6机器学习

6.1视频与图片的分解和合成

视频的分解

1 load 2 info 3 parse(解码) 4 imshow 5 imwrite

import cv2
cap = cv2.VideoCapture('1.mp4')
isOpened = cap.isOpended
print(isOpened)
fps = cap.get(cv2.CAP_PROP_FPS)
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
print(fps,width,height)
i = 0
while(isOpened):
    if i  ==10:
        break
    else:
        i = i+1 
    (flag,frame) = cap.read()#读取每一张flag frame
    fileName = 'image'+str(i)+'jpg'
    print(fileName)
    if flag == True:
        cv2.imwrite(fileName,frame,[cv2.IMWRITE_JPEG_QUALITY,100]
print('end!')

图片的合成

videoWrite = cv2.VideoWriter(‘2.mp4’,-1,5,size)—>size
接着循环遍历图片加到视频里

import cv2 
img = cv2.imread('image1.jpg')
imgInfo = img.shape
size = (imgInfo[1],imgInfo[0])
print(size)
videoWrite = cv2.VideoWriter('2.mp4',-1,5,size)
for i in range(1,11):
    fileName = 'image'+str(i)+'.jpg'
    img = cv2.imread(fileName)
    videoWrite.write(img)
print('end!')

6.2Haar特征+adaboost分类器

Haar特征

原理

Haar特征的计算量

快速计算的方法

计算特征->要计算矩形方框中的像素->有快速计算的方法：


结论：一个任意的方框，都可以由它相邻的ABCD四个矩形通过加减运算得来

adaboost分类器

原理

adaboost分类器的训练

Haar_adaboost人脸识别

import cv2
import numpy as np 
face_xml = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
eye_xml = cv2.CascadeClassifier('haarcascade_eye.xml')
img = cv2.imread('face.png')
cv2.imshow('src',img)
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
# detect faces 1 (gray)data 2 scale 3 5(目标大小，人脸最小不能小于5个像素)
faces = face_xml.detectMultiScale(gray,1.3,5)
print('faces=',len(faces))
#draw给人脸画方框
for (x,y,w,h) in faces:
    cv2.rectangle(img,(x,y),(x+w,y+h),(0,0,255),2)#最后一个参数：线条宽度
    #detect eyes
    roi_face = gray[y:y+h,x:x+w]
    roi_color = img[y:y+h,x:x+w]
    eyes = eye_xml.detectMultiScale(roi_face)
    print('eyes=',len(eyes))
    #draw 给眼睛画方框
    for (e_x,e_y,e_w,e_h) in eyes:
        cv2.rectangle( roi_color,(e_x,e_y),(e_x+e_w,e_y+e_h),(0,0,255),2)
cv2.imshow('dst',img)
cv2.waitKey(0)

6.3Hog特征+SVM

SVM(分类、监督学习)

小案例（给出身高体重，判断男or女）

Hog特征

Hog_SVM小狮子的识别

数字识别的案例

样本准备http://yann.lecun.com/exdb/mnist/

knn最近邻域法

1.Load Data

#本质：knn test 样本K个 max10 8个1 -》1
import tensorflow as tf
import numpy as np
import random
from tensorflow.examples.tutorials.mnist import input_data
# 1.load data 1 fileName 2 one_hot: 1 0000
mnist = input_data.read_data_sets('MNIST_data',one_hot=True)
# 2.属性设置
trainNum = 55000
testNum = 10000
trainSize = 500
testSize = 5
#在trainSize中找到4个与testSize最像的图片
k=4
#data分解
#获取范围在 0-trainNum 中的 trainSize个随机值 （replace不可重复）
trainIndex = np.random.choice(trainNum,trainSize,replace = False)#训练图片的下标（随机选取）
testIndex = np.random.choice(testNum,testSize,replace = False)
trainData = mnist.train.images[trainIndex]#训练图片
trainLabel = mnist.train.labels[trainIndex]#训练标签
testData = mnist.test.images[testIndex]#测试图片
testLabel = mnist.test.labels[testIndex]#测试标签
print('trainData.shape=',trainData.shape)#(500, 784)  一共500张训练图片，大小(宽*高)28*28 =784（一张图片784像素）
print('trainLabel.shape=',trainLabel.shape)#(500, 10)
print('testData.shape=',testData.shape)#(5, 784)
print('testLabel.shape=',testLabel.shape)#(5, 10)
print('testLabel=',testLabel)
#tf input 
trainDataInput = tf.placeholder(shape=[None,784],dtype= tf.float32)
trainLabelInput = tf.placeholder(shape=[None,10],dtype = tf.float32)
testDataInput = tf.placeholder(shape=[None,784],dtype= tf.float32)
testLabelInput = tf.placeholder(shape=[None,10],dtype = tf.float32)

\

2.knn test train distance

3.knn 对每一个测试图片，在500个训练图中，挑选k个（4个）最近（最像）的图片

4.用label对应找出k个（4个）最近（最像）的图片 对应的数字值

#knn distance 
#增加维度  5*784(2D) 变为5*1*784(3D)
#原因：5张测试784 分别与500张训练784 相减  即 5 500 784（3D） 5*500*784
f1 = tf.expand_dims(testDataInput,1)#维度扩展
f2 = tf.subtract(trainDataInput,f1)
f3 = tf.reduce_sum(tf.abs(f2),reduction_indices=2)#完成数据的累加 多维指定784的维数相加
#选取最接近的4张
f4 = tf.negative(f3)#取反功能
f5,f6 = tf.nn.top_k(f4,k=4)#选取f4最大的4个值  取反  即 f3最小的4个值
#由index获取具体图片内的数字值  f6 index -> trainLabelInput
f7 = tf.gather(trainLabelInput,f6)
#竖直方向的累加
f8 = tf.reduce_sum(f7,reduction_indices =1 )
#选取在某一个最大的值，获得下标index
#f9 ->test5 image ->5num
f9= tf.argmax(f8,dimension=1)
with tf.Session() as sess:
    p1 = sess.run(f1,feed_dict={testDataInput:testData[0:5]})
    print('p1.shape=',p1.shape)# (5, 1, 784)
    p2 = sess.run(f2,feed_dict={trainDataInput:trainData,testDataInput:testData[0:5]})
    print('p2.shape=',p2.shape)#(5, 500, 784)
    p3 = sess.run(f3,feed_dict={trainDataInput:trainData,testDataInput:testData[0:5]})
    print('p3.shape=',p3.shape)#(5, 500)
    print('p3[0,0]=',p3[0,0])#p3[0,0]= 114.79216
    p4 = sess.run(f4,feed_dict={trainDataInput:trainData,testDataInput:testData[0:5]})
    print('p4.shape=',p4.shape)
    print('p4[0,0]=',p4[0,0])#p4[0,0]= -114.79216
    p5,p6 = sess.run((f5,f6),feed_dict={trainDataInput:trainData,testDataInput:testData[0:5]})
    print('p5.shape=',p5.shape)#(5, 4)  每张测试图片（5张）分别对应4张最近训练图片
    print('p6.shape=',p6.shape)#(5, 4)
    print('p5[0,0]=',p5[0,0]) #   -45.580387
    print('p6[0,0]=',p6[0,0])#  196 (p6是下标index、)
    p7 = sess.run(f7,feed_dict={trainDataInput:trainData,testDataInput:testData[0:5],trainLabelInput:trainLabel})
    print('p7.shape=',p7.shape) #(5, 4, 10)
    print('p7=',p7)
    p8 = sess.run(f8,feed_dict={trainDataInput:trainData,testDataInput:testData[0:5],trainLabelInput:trainLabel})
    print('p8.shape=',p8.shape) # (5, 10)
    print('p8=',p8)
    p9 = sess.run(f9,feed_dict={trainDataInput:trainData,testDataInput:testData[0:5],trainLabelInput:trainLabel})
    print('p9.shape=',p9.shape) # (5, )
    print('p9=',p9)

5.检测概率统计

    p10 = np.argmax(testLabel[0:5],axis = 1)
    print('p10[]=',p10)
j = 0
for i in range(0,5):
    if p10[i] == p9[i]:
        j = j + 1
print('ac=',j*100/5)

cnn