目录
一、数据集准备
- Genki4k数据集下载
下载地址:https://inc.ucsd.edu/mplab/398.php - 划分数据集
将Genki4k数据集分为测试集与训练集。
并按标签分开存放。
二、基于卷积神经网络训练模型
1. 构建模型
构建代码:
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'))
查看征图的尺寸是如何随着每一层变化的
model.summary()
2. 图像数据预处理
from tensorflow.keras import optimizers
from tensorflow.keras.preprocessing.image import ImageDataGenerator
import os, shutil
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')
model.compile(loss='binary_crossentropy',
optimizer=optimizers.RMSprop(lr=1e-4),
metrics=['acc'])
base_dir = 'E:\\smile-detection-master\\datasets'
train_dir = os.path.join(base_dir, 'train_folder')
test_dir = os.path.join(base_dir, 'test_folder')
# All images will be rescaled by 1./255
# All images will be rescaled by 1./255
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')
test_generator = test_datagen.flow_from_directory(
test_dir,
target_size=(150, 150),
batch_size=32,
class_mode='binary')
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
3. 训练
代码:
history = model.fit_generator(
train_generator,
steps_per_epoch=100,
epochs=100,
validation_data=test_generator,
validation_steps=50)
保存模型:
model.save('E:\\smile-detection-master\\models\\smiles_and_unsmiles_small_2.h5')
4. 绘制模型的损失图和准确性图像
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()
5. 使用该模型进行微笑识别
#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('D:/smile1/smiles_and_unsmiles_small_2.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='smile'
else:
result='unsmile'
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(5) & 0xFF == ord('q'):
break
video.release()
cv2.destroyAllWindows()
这里采用上面训练出来的模型
smiles_and_unsmiles_small_1.h5
笑脸
识别:
非笑脸
识别:
三、使用OpenCV自带的微笑识别库
1.代码编写
import cv2
face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades+'haarcascade_frontalface_default.xml')
eye_cascade = cv2.CascadeClassifier(cv2.data.haarcascades+'haarcascade_eye.xml')
smile_cascade = cv2.CascadeClassifier(cv2.data.haarcascades+'haarcascade_smile.xml')
# 调用摄像头摄像头
cap = cv2.VideoCapture(0)
while(True):
# 获取摄像头拍摄到的画面
ret, frame = cap.read()
faces = face_cascade.detectMultiScale(frame, 1.3, 2)
img = frame
for (x,y,w,h) in faces:
# 画出人脸框,颜色自己定义,画笔宽度微
img = cv2.rectangle(img,(x,y),(x+w,y+h),(255,225,0),2)
# 框选出人脸区域,在人脸区域而不是全图中进行人眼检测,节省计算资源
face_area = img[y:y+h, x:x+w]
## 人眼检测
# 用人眼级联分类器引擎在人脸区域进行人眼识别,返回的eyes为眼睛坐标列表
eyes = eye_cascade.detectMultiScale(face_area,1.3,10)
for (ex,ey,ew,eh) in eyes:
#画出人眼框,绿色,画笔宽度为1
cv2.rectangle(face_area,(ex,ey),(ex+ew,ey+eh),(0,255,0),1)
## 微笑检测
# 用微笑级联分类器引擎在人脸区域进行人眼识别,返回的eyes为眼睛坐标列表
smiles = smile_cascade.detectMultiScale(face_area,scaleFactor= 1.16,minNeighbors=65,minSize=(25, 25),flags=cv2.CASCADE_SCALE_IMAGE)
for (ex,ey,ew,eh) in smiles:
#画出微笑框,(BGR色彩体系),画笔宽度为1
cv2.rectangle(face_area,(ex,ey),(ex+ew,ey+eh),(0,0,255),1)
cv2.putText(img,'Smile',(x,y-7), 3, 1.2, (222, 0, 255), 2, cv2.LINE_AA)
# 实时展示效果画面
cv2.imshow('frame2',img)
# 每5毫秒监听一次键盘动作
if cv2.waitKey(5) & 0xFF == ord('q'):
break
# 最后,关闭所有窗口
cap.release()
cv2.destroyAllWindows()
2.运行测试:
笑脸
识别:
非笑脸
识别:
四、基于Dlib笑脸识别
1.代码编写
import sys
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
class face_emotion():
def __init__(self):
# 使用特征提取器get_frontal_face_detector
self.detector = dlib.get_frontal_face_detector()
# dlib的68点模型,使用作者训练好的特征预测器
self.predictor = dlib.shape_predictor(
"shape_predictor_68_face_landmarks.dat")
# 建cv2摄像头对象,这里使用电脑自带摄像头,如果接了外部摄像头,则自动切换到外部摄像头
self.cap = cv2.VideoCapture(0)
# 设置视频参数,propId设置的视频参数,value设置的参数值
self.cap.set(3, 480)
# 截图screenshoot的计数器
self.cnt = 0
def learning_face(self):
# 眉毛直线拟合数据缓冲
line_brow_x = []
line_brow_y = []
# cap.isOpened() 返回true/false 检查初始化是否成功
while (self.cap.isOpened()):
# cap.read()
# 返回两个值:
# 一个布尔值true/false,用来判断读取视频是否成功/是否到视频末尾
# 图像对象,图像的三维矩阵
flag, im_rd = self.cap.read()
# 每帧数据延时1ms,延时为0读取的是静态帧
k = cv2.waitKey(1)
# 取灰度
img_gray = cv2.cvtColor(im_rd, cv2.COLOR_RGB2GRAY)
# 使用人脸检测器检测每一帧图像中的人脸。并返回人脸数rects
faces = self.detector(img_gray, 0)
# 待会要显示在屏幕上的字体
font = cv2.FONT_HERSHEY_SIMPLEX
# 如果检测到人脸
if (len(faces) != 0):
# 对每个人脸都标出68个特征点
for i in range(len(faces)):
# enumerate方法同时返回数据对象的索引和数据,k为索引,d为faces中的对象
for k, d in enumerate(faces):
# 用红色矩形框出人脸
cv2.rectangle(im_rd, (d.left(), d.top()),
(d.right(), d.bottom()), (0, 0, 255))
# 计算人脸热别框边长
self.face_width = d.right() - d.left()
# 使用预测器得到68点数据的坐标
shape = self.predictor(im_rd, d)
# 圆圈显示每个特征点
for i in range(68):
cv2.circle(im_rd, (shape.part(i).x, shape.part(
i).y), 2, (222, 222, 0), -1, 8)
#cv2.putText(im_rd, str(i), (shape.part(i).x, shape.part(i).y), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
# (255, 255, 255))
# 分析任意n点的位置关系来作为表情识别的依据
mouth_width = (shape.part(
54).x - shape.part(48).x) / self.face_width # 嘴巴咧开程度
mouth_higth = (shape.part(
66).y - shape.part(62).y) / self.face_width # 嘴巴张开程度
# 通过两个眉毛上的10个特征点,分析挑眉程度和皱眉程度
brow_sum = 0 # 高度之和
frown_sum = 0 # 两边眉毛距离之和
for j in range(17, 21):
brow_sum += (shape.part(j).y - d.top()) + \
(shape.part(j + 5).y - d.top())
frown_sum += shape.part(j + 5).x - shape.part(j).x
line_brow_x.append(shape.part(j).x)
line_brow_y.append(shape.part(j).y)
# self.brow_k, self.brow_d = self.fit_slr(line_brow_x, line_brow_y) # 计算眉毛的倾斜程度
tempx = np.array(line_brow_x)
tempy = np.array(line_brow_y)
z1 = np.polyfit(tempx, tempy, 1) # 拟合成一次直线
# 拟合出曲线的斜率和实际眉毛的倾斜方向是相反的
self.brow_k = -round(z1[0], 3)
brow_hight = (brow_sum / 10) / \
self.face_width # 眉毛高度占比
brow_width = (frown_sum / 5) / \
self.face_width # 眉毛距离占比
# 眼睛睁开程度
eye_sum = (shape.part(41).y - shape.part(37).y + shape.part(40).y - shape.part(38).y +
shape.part(47).y - shape.part(43).y + shape.part(46).y - shape.part(44).y)
eye_hight = (eye_sum / 4) / self.face_width
# 分情况讨论
# 张嘴,可能是开心或者惊讶
if round(mouth_higth >= 0.03):
if eye_hight >= 0.056:
cv2.putText(im_rd, "amazing", (d.left(), d.bottom() + 20), cv2.FONT_HERSHEY_SIMPLEX,
0.8,
(0, 0, 255), 2, 4)
else:
cv2.putText(im_rd, "smile", (d.left(), d.bottom() + 20), cv2.FONT_HERSHEY_SIMPLEX, 0.8,
(0, 0, 255), 2, 4)
# 没有张嘴,可能是正常和生气
else:
if self.brow_k <= -0.3:
cv2.putText(im_rd, "angry", (d.left(), d.bottom() + 20), cv2.FONT_HERSHEY_SIMPLEX, 0.8,
(0, 0, 255), 2, 4)
else:
cv2.putText(im_rd, "nosmile", (d.left(), d.bottom() + 20), cv2.FONT_HERSHEY_SIMPLEX, 0.8,
(0, 0, 255), 2, 4)
# 标出人脸数
cv2.putText(im_rd, "Faces: " + str(len(faces)),
(20, 50), font, 1, (0, 0, 255), 1, cv2.LINE_AA)
else:
# 没有检测到人脸
cv2.putText(im_rd, "No Face", (20, 50), font,
1, (0, 0, 255), 1, cv2.LINE_AA)
im_rd = cv2.putText(im_rd, "Q: quit", (20, 450),
font, 0.8, (0, 0, 255), 1, cv2.LINE_AA)
# 按下q键退出
if (k == ord('q')):
break
# 窗口显示
cv2.imshow("camera", im_rd)
# 释放摄像头
self.cap.release()
# 删除建立的窗口
cv2.destroyAllWindows()
if __name__ == "__main__":
my_face = face_emotion()
my_face.learning_face()
2. 运行测试
笑脸
识别:
非笑脸
识别:
五、总结
本文利用Genki4k人脸微笑数据集,分别通过搭建卷积神经网络模型、Dlib,以及OpenCV自带的微笑识别库,完成人脸实时的微笑识别,
六、参考
https://blog.csdn.net/lxzysx/article/details/107028257
https://blog.csdn.net/qq_41133375/article/details/107141898