juypter 基于keras训练的笑脸检测+口罩人脸识别

一、笑脸数据集（genki4k）正负样本的划分、模型训练和测试的过程（至少包括SVM、CNN），输出模型训练精度和测试精度（F1-score和ROC）

1.训练数据

1) 导入Keras

import keras
keras.__version__

'2.3.1'

2) 读取数据集，训练

import os, shutil
# The path to the directory where the original
# dataset was uncompressed
original_dataset_dir = 'smileface/genki4k'

# The directory where we will
# store our smaller dataset
base_dir = 'smileface/1'
os.mkdir(base_dir)

# Directories for our training,
# validation and test splits
train_dir = os.path.join(base_dir, 'train')
os.mkdir(train_dir)
validation_dir = os.path.join(base_dir, 'validation')
os.mkdir(validation_dir)
test_dir = os.path.join(base_dir, 'test')
os.mkdir(test_dir)

# Directory with our training smile pictures
train_smile_dir = os.path.join(train_dir, 'smiles')
os.mkdir(train_smile_dir)

# Directory with our training nosmile pictures
train_nosmile_dir = os.path.join(train_dir, 'no_smiles')
os.mkdir(train_nosmile_dir)

# Directory with our validation smile pictures
validation_smile_dir = os.path.join(validation_dir, 'smiles')
os.mkdir(validation_smile_dir)

# Directory with our validation nosmile pictures
validation_nosmile_dir = os.path.join(validation_dir, 'no_smiles')
os.mkdir(validation_nosmile_dir)

# Directory with our validation smile pictures
test_smile_dir = os.path.join(test_dir, 'smiles')
os.mkdir(test_smile_dir)

# Directory with our validation nosmile pictures
test_nosmile_dir = os.path.join(test_dir, 'no_smiles')
os.mkdir(test_nosmile_dir)

3）将笑脸数据存入对应文件夹

print('total training smile images:', len(os.listdir(train_smile_dir)))
print('total training nosmile images:', len(os.listdir(train_nosmile_dir)))
print('total validation smile images:', len(os.listdir(validation_smile_dir)))
print('total validation nosmile images:', len(os.listdir(validation_nosmile_dir)))
print('total test smile images:', len(os.listdir(test_smile_dir)))
print('total test nosmile images:', len(os.listdir(test_nosmile_dir)))

total training smile images: 2162
total training nosmile images: 1837
total validation smile images: 2162
total validation nosmile images: 1837
total test smile images: 2162
total test nosmile images: 1837

4) 构建小型卷积神经网络

from keras import layers
from keras import models

model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

了解征图的尺寸是如何随着每一层变化的

model.summary()

Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_5 (Conv2D)            (None, 148, 148, 32)      896       
_________________________________________________________________
max_pooling2d_5 (MaxPooling2 (None, 74, 74, 32)        0         
_________________________________________________________________
conv2d_6 (Conv2D)            (None, 72, 72, 64)        18496     
_________________________________________________________________
max_pooling2d_6 (MaxPooling2 (None, 36, 36, 64)        0         
_________________________________________________________________
conv2d_7 (Conv2D)            (None, 34, 34, 128)       73856     
_________________________________________________________________
max_pooling2d_7 (MaxPooling2 (None, 17, 17, 128)       0         
_________________________________________________________________
conv2d_8 (Conv2D)            (None, 15, 15, 128)       147584    
_________________________________________________________________
max_pooling2d_8 (MaxPooling2 (None, 7, 7, 128)         0         
_________________________________________________________________
flatten_2 (Flatten)          (None, 6272)              0         
_________________________________________________________________
dense_3 (Dense)              (None, 512)               3211776   
_________________________________________________________________
dense_4 (Dense)              (None, 1)                 513       
=================================================================
Total params: 3,453,121
Trainable params: 3,453,121
Non-trainable params: 0
_________________________________________________________________

from keras import optimizers
model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
              metrics=['acc'])

5) 数据预处理

from keras.preprocessing.image import ImageDataGenerator
# All images will be rescaled by 1./255
train_datagen = ImageDataGenerator(rescale=1./255)
test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
        # This is the target directory
        train_dir,
        # All images will be resized to 150x150
        target_size=(150, 150),
        batch_size=20,
        # Since we use binary_crossentropy loss, we need binary labels
        class_mode='binary')

validation_generator = test_datagen.flow_from_directory(
        validation_dir,
        target_size=(150, 150),
        batch_size=20,
        class_mode='binary')

Found 3999 images belonging to 2 classes.
Found 3999 images belonging to 2 classes.

生成器的输出

for data_batch, labels_batch in train_generator:
    print('data batch shape:', data_batch.shape)
    print('labels batch shape:', labels_batch.shape)
    break

data batch shape: (20, 150, 150, 3)
labels batch shape: (20,)

6) 训练模型

input_shape=(784),
history = model.fit_generator(
    train_generator,
    steps_per_epoch=50,
    epochs=15,
    validation_data=validation_generator,
    validation_steps=50)

WARNING:tensorflow:From C:\Users\xieru\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:422: The name tf.global_variables is deprecated. Please use tf.compat.v1.global_variables instead.

Epoch 1/15
50/50 [==============================] - 28s 560ms/step - loss: 0.6920 - acc: 0.5510 - val_loss: 0.7255 - val_acc: 0.5630
Epoch 2/15
50/50 [==============================] - 27s 547ms/step - loss: 0.6781 - acc: 0.5760 - val_loss: 0.6440 - val_acc: 0.5820
Epoch 3/15
50/50 [==============================] - 27s 545ms/step - loss: 0.6776 - acc: 0.5820 - val_loss: 0.7515 - val_acc: 0.5050
Epoch 4/15
50/50 [==============================] - 27s 539ms/step - loss: 0.6846 - acc: 0.5536 - val_loss: 0.6532 - val_acc: 0.6196
Epoch 5/15
50/50 [==============================] - 28s 552ms/step - loss: 0.6688 - acc: 0.6016 - val_loss: 0.6846 - val_acc: 0.6400
Epoch 6/15
50/50 [==============================] - 28s 555ms/step - loss: 0.6520 - acc: 0.6280 - val_loss: 0.6407 - val_acc: 0.6370
Epoch 7/15
50/50 [==============================] - 28s 557ms/step - loss: 0.6586 - acc: 0.6190 - val_loss: 0.6581 - val_acc: 0.6010
Epoch 8/15
50/50 [==============================] - 27s 547ms/step - loss: 0.6496 - acc: 0.6250 - val_loss: 0.6380 - val_acc: 0.6847
Epoch 9/15
50/50 [==============================] - 28s 562ms/step - loss: 0.6359 - acc: 0.6350 - val_loss: 0.6808 - val_acc: 0.5910
Epoch 10/15
50/50 [==============================] - 29s 571ms/step - loss: 0.6255 - acc: 0.6590 - val_loss: 0.7137 - val_acc: 0.6760
Epoch 11/15
50/50 [==============================] - 29s 581ms/step - loss: 0.6171 - acc: 0.6680 - val_loss: 0.6744 - val_acc: 0.7000
Epoch 12/15
50/50 [==============================] - 28s 568ms/step - loss: 0.6147 - acc: 0.6677 - val_loss: 0.5016 - val_acc: 0.6897
Epoch 13/15
50/50 [==============================] - 28s 562ms/step - loss: 0.5846 - acc: 0.7090 - val_loss: 0.4679 - val_acc: 0.6950
Epoch 14/15
50/50 [==============================] - 28s 557ms/step - loss: 0.5835 - acc: 0.7190 - val_loss: 0.5299 - val_acc: 0.7010
Epoch 15/15
50/50 [==============================] - 30s 609ms/step - loss: 0.5677 - acc: 0.7180 - val_loss: 0.6687 - val_acc: 0.6950

这里使用fit_generator方法来完成此操作，对于我们这样的数据生成器，它相当于fit方法。由于电脑性能原因，我训练的轮数较短。

保存模型

model.save('smile_and_nosmile.h5')

保存好模型之后便可以进行此模型的人脸识别了，不过精度不会太高，毕竟只使用了这么少量的图片

绘制模型的损失和准确性

import matplotlib.pyplot as plt

acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(len(acc))

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()

plt.figure()

plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
plt.show()

8) 数据增强

数据增强采用的方法是从现有的训练样本中生成更多的训练数据，方法是通过一系列随机变换来“增强”样本，从而产生看上去可信的图像。我们的目标是在训练时，我们的模型不会两次看到完全相同的图像。这有助于将模型暴露于数据的更多方面，并更好地泛化。

datagen = ImageDataGenerator(
      rotation_range=40,
      width_shift_range=0.2,
      height_shift_range=0.2,
      shear_range=0.2,
      zoom_range=0.2,
      horizontal_flip=True,
      fill_mode='nearest')

# This is module with image preprocessing utilities
from keras.preprocessing import image

fnames = [os.path.join(train_smile_dir, fname) for fname in os.listdir(train_smile_dir)]

# We pick one image to "augment"
img_path = fnames[3]

# Read the image and resize it
img = image.load_img(img_path, target_size=(150, 150))

# Convert it to a Numpy array with shape (150, 150, 3)
x = image.img_to_array(img)

# Reshape it to (1, 150, 150, 3)
x = x.reshape((1,) + x.shape)

# The .flow() command below generates batches of randomly transformed images.
# It will loop indefinitely, so we need to `break` the loop at some point!
i = 0
for batch in datagen.flow(x, batch_size=1):
    plt.figure(i)
    imgplot = plt.imshow(image.array_to_img(batch[0]))
    i += 1
    if i % 4 == 0:
        break

plt.show()

如果我们使用这种数据增加配置训练一个新的网络，我们的网络将永远不会看到两次相同的输入。然而，它看到的输入仍然是高度相关的，因为它们来自少量的原始图像——我们不能产生新的信息，我们只能混合现有的信息。因此，这可能还不足以完全消除过度拟合。

为了进一步对抗过拟合，我们还将在我们的模型中增加一个Dropout层，就在密集连接分类器之前：

model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
                        input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dropout(0.5))
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

model.compile(loss='binary_crossentropy',
              optimizer=optimizers.RMSprop(lr=1e-4),
              metrics=['acc'])

用数据增强和退出训练网络

train_datagen = ImageDataGenerator(
    rescale=1./255,
    rotation_range=40,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,)

# Note that the validation data should not be augmented!
test_datagen = ImageDataGenerator(rescale=1./255)

train_generator = train_datagen.flow_from_directory(
        # This is the target directory
        train_dir,
        # All images will be resized to 150x150
        target_size=(150, 150),
        batch_size=32,
        # Since we use binary_crossentropy loss, we need binary labels
        class_mode='binary')

validation_generator = test_datagen.flow_from_directory(
        validation_dir,
        target_size=(150, 150),
        batch_size=32,
        class_mode='binary')

history = model.fit_generator(
      train_generator,
      steps_per_epoch=30,
      epochs=30,
      validation_data=validation_generator,
      validation_steps=25)

Found 3999 images belonging to 2 classes.
Found 3999 images belonging to 2 classes.
Epoch 1/30
30/30 [==============================] - 26s 853ms/step - loss: 0.6824 - acc: 0.5656 - val_loss: 0.6342 - val_acc: 0.5725
Epoch 2/30
30/30 [==============================] - 25s 836ms/step - loss: 0.6790 - acc: 0.5552 - val_loss: 0.6786 - val_acc: 0.5725
Epoch 3/30
30/30 [==============================] - 25s 839ms/step - loss: 0.6871 - acc: 0.5558 - val_loss: 0.6456 - val_acc: 0.5975
Epoch 4/30
30/30 [==============================] - 25s 847ms/step - loss: 0.6800 - acc: 0.5740 - val_loss: 0.6984 - val_acc: 0.5788
Epoch 5/30
30/30 [==============================] - 25s 838ms/step - loss: 0.6841 - acc: 0.5506 - val_loss: 0.6941 - val_acc: 0.5695
Epoch 6/30
30/30 [==============================] - 26s 858ms/step - loss: 0.6797 - acc: 0.5625 - val_loss: 0.6757 - val_acc: 0.5875
Epoch 7/30
30/30 [==============================] - 26s 856ms/step - loss: 0.6849 - acc: 0.5635 - val_loss: 0.6757 - val_acc: 0.5888
Epoch 8/30
30/30 [==============================] - 26s 859ms/step - loss: 0.6797 - acc: 0.5844 - val_loss: 0.6609 - val_acc: 0.5600
Epoch 9/30
30/30 [==============================] - 26s 856ms/step - loss: 0.6791 - acc: 0.5625 - val_loss: 0.6076 - val_acc: 0.6075
Epoch 10/30
30/30 [==============================] - 25s 828ms/step - loss: 0.6730 - acc: 0.5719 - val_loss: 0.6391 - val_acc: 0.5682
Epoch 11/30
30/30 [==============================] - 25s 849ms/step - loss: 0.6791 - acc: 0.5740 - val_loss: 0.7491 - val_acc: 0.5150
Epoch 12/30
30/30 [==============================] - 25s 829ms/step - loss: 0.6762 - acc: 0.5892 - val_loss: 0.6416 - val_acc: 0.5875
Epoch 13/30
30/30 [==============================] - 25s 833ms/step - loss: 0.6809 - acc: 0.5813 - val_loss: 0.6868 - val_acc: 0.5962
Epoch 14/30
30/30 [==============================] - 25s 836ms/step - loss: 0.6680 - acc: 0.5854 - val_loss: 0.6515 - val_acc: 0.6100
Epoch 15/30
30/30 [==============================] - 26s 851ms/step - loss: 0.6757 - acc: 0.5938 - val_loss: 0.6267 - val_acc: 0.6245
Epoch 16/30
30/30 [==============================] - 25s 846ms/step - loss: 0.6706 - acc: 0.5892 - val_loss: 0.6812 - val_acc: 0.5825
Epoch 17/30
30/30 [==============================] - 25s 830ms/step - loss: 0.6764 - acc: 0.5823 - val_loss: 0.6789 - val_acc: 0.6237
Epoch 18/30
30/30 [==============================] - 25s 841ms/step - loss: 0.6729 - acc: 0.5798 - val_loss: 0.7280 - val_acc: 0.6025
Epoch 19/30
30/30 [==============================] - 25s 832ms/step - loss: 0.6756 - acc: 0.5938 - val_loss: 0.6452 - val_acc: 0.6125
Epoch 20/30
30/30 [==============================] - 25s 835ms/step - loss: 0.6722 - acc: 0.5917 - val_loss: 0.5814 - val_acc: 0.6058
Epoch 21/30
30/30 [==============================] - 25s 841ms/step - loss: 0.6654 - acc: 0.6156 - val_loss: 0.6956 - val_acc: 0.5900
Epoch 22/30
30/30 [==============================] - 25s 830ms/step - loss: 0.6787 - acc: 0.5704 - val_loss: 0.6486 - val_acc: 0.6025
Epoch 23/30
30/30 [==============================] - 25s 842ms/step - loss: 0.6671 - acc: 0.6062 - val_loss: 0.6033 - val_acc: 0.6212
Epoch 24/30
30/30 [==============================] - 25s 827ms/step - loss: 0.6582 - acc: 0.6083 - val_loss: 0.7576 - val_acc: 0.5975
Epoch 25/30
30/30 [==============================] - 25s 835ms/step - loss: 0.6678 - acc: 0.5844 - val_loss: 0.7062 - val_acc: 0.6045
Epoch 26/30
30/30 [==============================] - 25s 835ms/step - loss: 0.6756 - acc: 0.5860 - val_loss: 0.5878 - val_acc: 0.6162
Epoch 27/30
30/30 [==============================] - 25s 841ms/step - loss: 0.6727 - acc: 0.5719 - val_loss: 0.6711 - val_acc: 0.6187
Epoch 28/30
30/30 [==============================] - 25s 844ms/step - loss: 0.6657 - acc: 0.5969 - val_loss: 0.6661 - val_acc: 0.5950
Epoch 29/30
30/30 [==============================] - 25s 828ms/step - loss: 0.6669 - acc: 0.6031 - val_loss: 0.6573 - val_acc: 0.6250
Epoch 30/30
30/30 [==============================] - 25s 841ms/step - loss: 0.6563 - acc: 0.6246 - val_loss: 0.7193 - val_acc: 0.5695

model.save('smile_and_nosmile_1.h5')

再查看模型的损失和准确性

acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(len(acc))

plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()

plt.figure()

plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
plt.show()

import cv2
from keras.preprocessing import image
from keras.models import load_model
import numpy as np
import dlib
from PIL import Image
model = load_model('smile_and_nosmile_1.h5')
detector = dlib.get_frontal_face_detector()
video=cv2.VideoCapture(0)
font = cv2.FONT_HERSHEY_SIMPLEX
def rec(img):
    gray=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
    dets=detector(gray,1)
    if dets is not None:
        for face in dets:
            left=face.left()
            top=face.top()
            right=face.right()
            bottom=face.bottom()
            cv2.rectangle(img,(left,top),(right,bottom),(0,255,0),2)
            img1=cv2.resize(img[top:bottom,left:right],dsize=(150,150))
            img1=cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)
            img1 = np.array(img1)/255.
            img_tensor = img1.reshape(-1,150,150,3)
            prediction =model.predict(img_tensor)    
            print(prediction)
            if prediction[0][0]<0.5:
                result='no_smiles'
            else:
                result='smiles'
            cv2.putText(img, result, (left,top), font, 2, (0, 255, 0), 2, cv2.LINE_AA)
        cv2.imshow('smile detector', img)
while video.isOpened():
    res, img_rd = video.read()
    if not res:
        break
    rec(img_rd)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break
video.release()
cv2.destroyAllWindows()

[[0.373922]]
[[0.37300688]]
[[0.36894965]]
[[0.37052]]
[[0.36977816]]
[[0.36832398]]
[[0.36462715]]
[[0.37091285]]
[[0.37168673]]
[[0.3695552]]
[[0.37056762]]
[[0.3709046]]
[[0.3715356]]
[[0.3716728]]
[[0.37013638]]
[[0.3702656]]
[[0.37569472]]
[[0.3751066]]
[[0.37662053]]
[[0.37343165]]
[[0.3769425]]
[[0.3725439]]
[[0.369594]]
[[0.37417337]]
[[0.3706985]]
[[0.37367338]]
[[0.37272263]]
[[0.37169605]]
[[0.37007022]]
[[0.37382162]]
[[0.36791837]]
[[0.37170732]]
[[0.37154534]]
[[0.37069517]]
[[0.3701343]]

在这里插入图片描述

二、将数据集换为口罩数据集，并进行训练，实现佩戴口罩的人脸识别

1.口罩数据集下载

训练人脸口罩数据集和训练笑脸数据集一样，只需改一下和相应的变量名和数据集。

import cv2
from keras.preprocessing import image
from keras.models import load_model
import numpy as np
import dlib
from PIL import Image
model = load_model('mask_and_nomask.h5')
detector = dlib.get_frontal_face_detector()
video=cv2.VideoCapture(0)
font = cv2.FONT_HERSHEY_SIMPLEX
def rec(img):
    gray=cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
    dets=detector(gray,1)
    if dets is not None:
        for face in dets:
            left=face.left()
            top=face.top()
            right=face.right()
            bottom=face.bottom()
            cv2.rectangle(img,(left,top),(right,bottom),(0,255,0),2)
            img1=cv2.resize(img[top:bottom,left:right],dsize=(150,150))
            img1=cv2.cvtColor(img1,cv2.COLOR_BGR2RGB)
            img1 = np.array(img1)/255.
            img_tensor = img1.reshape(-1,150,150,3)
            prediction =model.predict(img_tensor)    
            print(prediction)
            if prediction[0][0]>0.5:
                result='nomask'
            else:
                result='mask'
            cv2.putText(img, result, (left,top), font, 2, (0, 255, 0), 2, cv2.LINE_AA)
        cv2.imshow('mask detector', img)
while video.isOpened():
    res, img_rd = video.read()
    if not res:
        break
    rec(img_rd)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break
video.release()
cv2.destroyAllWindows()

参考链接: Python — 笑脸检测+口罩人脸识别.