机器学习-人脸识别

人脸识别在现在高科技的社会中，随处可见，作为一个初学者小白，根据老师上课的内容，完善了一个小型的人脸识别案例

import cv2
import os
import numpy as np

CASE_PATH="C:\\Users\Administrator\Anaconda3\Lib\site-packages\cv2\data\haarcascade_frontalface_default.xml"

RAM_IMAGE_DIR ='./me/'   
DATASET_DIR='./images/'  #数据集

face_cascade=cv2.CascadeClassifier(CASE_PATH)

def save_faces(img,name,x,y,width,height):
    image=img[y:y+height,x:x+width]
    if img.size ==  0:#对于这种情况就不要用imwriter
        cv2.imwrite(name,image)

image_list=os.listdir(RAM_IMAGE_DIR)#列出文件夹下面所有的文件
count=166

for image_path in image_list:
    fullpath=RAM_IMAGE_DIR+image_path
    image=cv2.imread(fullpath)

    gray=cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)
    faces=face_cascade.detectMultiScale(gray,scaleFactor=1.2,minNeighbors=5,minSize=(5,5))
    
    for(x,y,width,height)in faces:
       
       save_faces(image,DATASET_DIR+image_path,x,y-30,width,height+30)
    count+=1

 前面的都是通用代码

对读取的图片进行处理

#为了方便后期卷积数据的输入，对切割出来的图片尺寸进行调整
def resize_without_deformation(image,size=(100,100)):
    height,width,_=image.shape
    longest_edge=max(height,width)
    top,bottom,left,right=0,0,0,0
    if height

#读取图片
def read_image(size=None):
    data_x,data_y=[],[]
    
    image_list=os.listdir(DATASET_DIR)#列出文件夹下面所有的文件

    name=''
    number=-1
    for image_path in image_list:
        fullpath=DATASET_DIR+image_path
        
        myName=image_path.split('_')[0]
        if myName != name:
            name=myName
            number+=1
            
        try:
            im=cv2.imread(fullpath)
            
        except IOError as e:
            print(e)
        except:
            print('Unknown Error!')
        else:
            if size is None:
                size=(100,100)
            im=resize_without_deformation(im,size)
            data_x.append(np.asarray(im,dtype=np.int8))
            data_y.append(number)

    return data_x,data_y  

 接下来构建网络模型

from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense
from tensorflow.keras.models import Sequential, load_model
from tensorflow.keras import optimizers
from tensorflow.keras.preprocessing.image import ImageDataGenerator

IMAGE_SIZE = 100
raw_images,raw_labels = read_image(size=(IMAGE_SIZE,IMAGE_SIZE))
raw_images,raw_labels = np.asarray(raw_images,dtype=np.float32),np.asarray(raw_labels,dtype = np.float32)

from keras.utils import np_utils
ont_hot_labels = np_utils.to_categorical(raw_labels)

from sklearn.model_selection import train_test_split
train_input,test_input,train_output,test_output= train_test_split(raw_images,ont_hot_labels,test_size=0.3)

train_input /= 255.0
test_input /= 255.0

 模型输出

import tensorflow as tf
face_recongnition_model = tf.keras.models.Sequential()
face_recongnition_model.add(tf.keras.layers.Conv2D(32,(3,3),input_shape=(IMAGE_SIZE,IMAGE_SIZE,3),activation='relu',padding ='same'))
face_recongnition_model.add(tf.keras.layers.Conv2D(32,(3,3),activation='relu',padding ='same'))
face_recongnition_model.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))
face_recongnition_model.add(tf.keras.layers.Dropout(0.2))

face_recongnition_model.add(tf.keras.layers.Conv2D(64,(3,3),activation='relu',padding ='same'))
face_recongnition_model.add(tf.keras.layers.Conv2D(64,(3,3),activation='relu',padding ='same'))
face_recongnition_model.add(tf.keras.layers.MaxPooling2D(pool_size=(2, 2)))
face_recongnition_model.add(tf.keras.layers.Dropout(0.2))

face_recongnition_model.add(tf.keras.layers.Flatten())
face_recongnition_model.add(tf.keras.layers.Dense(512, activation = 'relu'))
face_recongnition_model.add(tf.keras.layers.Dropout(0.4))
face_recongnition_model.add(tf.keras.layers.Dense(len(ont_hot_labels[0]), activation = 'sigmoid'))

face_recongnition_model.summary()

learning_rate=0.01
decay=1e-6  #学习率衰减因子，用来随着迭代次数不断减小学习率，防止出现震荡
momentum=0.8  #引入冲量，可以在学习率较小的时候加速学习，又可以在学习率较大的时候减速
"""损失函数使用交叉熵，交叉熵函数随着输出核期望的差距越来越大,
曲线就会越来越陡峭,对权值的惩罚力度也会越来越大，其他的损失函数也是可以的"""
nesterov=True  
#sgd=SGD(lr=learning_rate,decay=decay,momentum=momentum,nesterov=nesterov)

face_recongnition_model.compile(loss='categorical_crossentropy',
                                optimizer=optimizers.SGD(lr=learning_rate,decay=decay,momentum=momentum,nesterov=nesterov),metrics=['acc'])

#开始训练，训练100次（epochs），每次训练分几个批次，每批（batch_size）20个，shuffle用来打乱样本顺序
batch_size=20 #每批训练数据量的大小
epochs=100
face_recongnition_model.fit(train_input,train_output,epochs=epochs,batch_size=batch_size,shuffle=True)

#训练完成后在测试集上评估结果并保存模型供以后加载使用
print(face_recongnition_model.evaluate(test_input,test_output,verbose=0))
MODEL_PATH='face_model.h5'
face_recongnition_model.save(MODEL_PATH)

 

 

from tensorflow import keras
#from keras.models import load_model
from tensorflow.keras.models import load_model
#加载卷积神经网络模型
face_recongnition_model=keras.Sequential()
MODEL_PATH='face_model.h5'
face_recognition_model=load_model(MODEL_PATH)


image=cv2.imread("test//Abhishek Bachan_20.jpg",1)
gray=cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)

#人脸检测：
faces=face_cascade.detectMultiScale(gray,scaleFactor=1.1,minNeighbors=5,minSize=(30,30))

for(x,y,width,height) in faces:
    img =image[y:y+height,x:x+width]
    img =resize_without_deformation(img)
    
    img=img.reshape((1,100,100,3))
    img =np.asarray(img,dtype=np.float32)
    img /=255.0
    
    result =face_recognition_model.predict_classes(img)

    cv2.rectangle(image,(x,y),(x+width,y+height),(0,255,0),2)
    font=cv2.FONT_HERSHEY_SIMPLEX
   
    if result[0]==8:
        cv2.putText(image,'头像',(x,y-2),font,0.7,(0,255,0),2)
    else:
        cv2.putText(image,'No.%d' % result[0], (x, y-2), font, 0.7,(0, 255, 0),2)
        
cv2.imshow('',image)
cv2.waitKey(0)

 最后识别的人脸能用彩色方框标注出来