一、卷积神经网络概述
卷积神经网络(Convolutional Neural Networks, CNN)是一类包含卷积计算且具有深度结构的前馈神经网络(Feedforward Neural Networks),是深度学习(deep learning)的代表算法之一 。卷积神经网络具有表征学习(representation learning)能力,能够按其阶层结构对输入信息进行平移不变分类(shift-invariant classification),因此也被称为“平移不变人工神经网络(Shift-Invariant Artificial Neural Networks, SIANN)” 。
对卷积神经网络的研究始于二十世纪80至90年代,时间延迟网络和LeNet-5是最早出现的卷积神经网络 ;在二十一世纪后,随着深度学习理论的提出和数值计算设备的改进,卷积神经网络得到了快速发展,并被应用于计算机视觉、自然语言处理等领域 [2] 。
卷积神经网络仿造生物的视知觉(visual perception)机制构建,可以进行监督学习和非监督学习,其隐含层内的卷积核参数共享和层间连接的稀疏性使得卷积神经网络能够以较小的计算量对格点化(grid-like topology)特征,例如像素和音频进行学习、有稳定的效果且对数据没有额外的特征工程(feature engineering)要求 。
二、图片预处理
-
打开微笑图片数据集
-
将人脸检测出来并对图片进行裁剪
import dlib # 人脸识别的库dlib
import numpy as np # 数据处理的库numpy
import cv2 # 图像处理的库OpenCv
import os
# dlib预测器
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')
# 读取图像的路径
path_read = "F:\\jupyter\\source\\picture\\genki4k\\files"
num=0
for file_name in os.listdir(path_read):
#aa是图片的全路径
aa=(path_read +"/"+file_name)
#读入的图片的路径中含非英文
img=cv2.imdecode(np.fromfile(aa, dtype=np.uint8), cv2.IMREAD_UNCHANGED)
#获取图片的宽高
img_shape=img.shape
img_height=img_shape[0]
img_width=img_shape[1]
# 用来存储生成的单张人脸的路径
path_save="F:\\jupyter\\source\\picture\\genki4k\\files1"
# dlib检测
dets = detector(img,1)
print("人脸数:", len(dets))
for k, d in enumerate(dets):
if len(dets)>1:
continue
num=num+1
# 计算矩形大小
# (x,y), (宽度width, 高度height)
pos_start = tuple([d.left(), d.top()])
pos_end = tuple([d.right(), d.bottom()])
# 计算矩形框大小
height = d.bottom()-d.top()
width = d.right()-d.left()
# 根据人脸大小生成空的图像
img_blank = np.zeros((height, width, 3), np.uint8)
for i in range(height):
if d.top()+i>=img_height:# 防止越界
continue
for j in range(width):
if d.left()+j>=img_width:# 防止越界
continue
img_blank[i][j] = img[d.top()+i][d.left()+j]
img_blank = cv2.resize(img_blank, (200, 200), interpolation=cv2.INTER_CUBIC)
cv2.imencode('.jpg', img_blank)[1].tofile(path_save+"\\"+"file"+str(num)+".jpg") # 正确方法
- 运行效果:
三、划分数据集
import os, shutil
# 原始数据集路径
original_dataset_dir = 'F:\\jupyter\\source\\picture\\genki4k\\files1'
# 新的数据集
base_dir = 'F:\\jupyter\\source\\picture\\genki4k\\files2'
os.mkdir(base_dir)
# 训练图像、验证图像、测试图像的目录
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)
train_smile_dir = os.path.join(train_dir, 'smile')
os.mkdir(train_smile_dir)
train_unsmile_dir = os.path.join(train_dir, 'unsmile')
os.mkdir(train_unsmile_dir)
validation_smile_dir = os.path.join(validation_dir, 'smile')
os.mkdir(validation_smile_dir)
validation_unsmile_dir = os.path.join(validation_dir, 'unsmile')
os.mkdir(validation_unsmile_dir)
test_smile_dir = os.path.join(test_dir, 'smile')
os.mkdir(test_smile_dir)
test_unsmile_dir = os.path.join(test_dir, 'unsmile')
os.mkdir(test_unsmile_dir)
# 复制1000张笑脸图片到train_c_dir
fnames = ['file{}.jpg'.format(i) for i in range(1,900)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(train_smile_dir, fname)
shutil.copyfile(src, dst)
fnames = ['file{}.jpg'.format(i) for i in range(900, 1350)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(validation_smile_dir, fname)
shutil.copyfile(src, dst)
# Copy next 500 cat images to test_cats_dir
fnames = ['file{}.jpg'.format(i) for i in range(1350, 1800)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(test_smile_dir, fname)
shutil.copyfile(src, dst)
fnames = ['file{}.jpg'.format(i) for i in range(2127,3000)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(train_unsmile_dir, fname)
shutil.copyfile(src, dst)
# Copy next 500 dog images to validation_dogs_dir
fnames = ['file{}.jpg'.format(i) for i in range(3000,3878)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(validation_unsmile_dir, fname)
shutil.copyfile(src, dst)
# Copy next 500 dog images to test_dogs_dir
fnames = ['file{}.jpg'.format(i) for i in range(3000,3878)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(test_unsmile_dir, fname)
shutil.copyfile(src, dst)
- 运行结果:
四、CNN提取人脸识别笑脸和非笑脸
#创建模型
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'))
model.summary()#查看
- 运行结果:
- 归一化处理
#归一化
from tensorflow import optimizers
model.compile(loss='binary_crossentropy',
optimizer=optimizers.RMSprop(learning_rate=1e-4),
metrics=['acc'])
from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale=1./255)
validation_datagen=ImageDataGenerator(rescale=1./255)
test_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_directory(
# 目标文件目录
train_dir,
#所有图片的size必须是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')
test_generator = test_datagen.flow_from_directory(test_dir,
target_size=(150, 150),
batch_size=20,
class_mode='binary')
for data_batch, labels_batch in train_generator:
print('data batch shape:', data_batch.shape)
print('labels batch shape:', labels_batch)
break
-
运行结果:
-
数据增强
#数据增强
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')
#数据增强后图片变化
import matplotlib.pyplot as plt
# 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)]
img_path = fnames[3]
img = image.load_img(img_path, target_size=(150, 150))
x = image.img_to_array(img)
x = x.reshape((1,) + x.shape)
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()
运行结果:
- 创建网络
#创建网络
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(learning_rate=1e-4),
metrics=['acc'])
#归一化处理
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,)
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(
train_generator,
steps_per_epoch=56,
epochs=100,
validation_data=validation_generator,
validation_steps=42)
model.save('smileAndUnsmile1.h5')
#数据增强过后的训练集与验证集的精确度与损失度的图形
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()
- 单张图片测试
# 单张图片进行判断 是笑脸还是非笑脸
import cv2
from keras.preprocessing import image
from keras.models import load_model
import numpy as np
#加载模型
model = load_model('smileAndUnsmile1.h5')
#本地图片路径
img_path='test1.jpg'
img = image.load_img(img_path, target_size=(150, 150))
img_tensor = image.img_to_array(img)/255.0
img_tensor = np.expand_dims(img_tensor, axis=0)
prediction =model.predict(img_tensor)
print(prediction)
if prediction[0][0]>0.5:
result='非笑脸'
else:
result='笑脸'
print(result)
测试图片:
测试结果:
- 摄像头实时测试
#检测视频或者摄像头中的人脸
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('smileAndUnsmile1.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)
if prediction[0][0]>0.5:
result='unsmile'
else:
result='smile'
cv2.putText(img, result, (left,top), font, 2, (0, 255, 0), 2, cv2.LINE_AA)
cv2.imshow('Video', 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()