[人工智能]Python基于keras训练简单微笑识别

一、数据预处理

实验数据来自genki4k

提取含有完整人脸的图片

def init_file():
    num = 0
    bar = tqdm(os.listdir(read_path))
    for file_name in bar:
        bar.desc = "预处理图片: "
        # a图片的全路径
        img_path = (read_path + "/" + file_name)
        # 读入的图片的路径中含非英文
        img = cv2.imdecode(np.fromfile(img_path, 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

        # dlib检测
        dets = detector(img, 1)
        for k, d in enumerate(dets):
            if len(dets) > 1:
                continue
            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)
            # 保存图片
            cv2.imencode('.jpg', img_blank)[1].tofile(save_path + "/" + "file" + str(num) + ".jpg")

    logging.info("一共", len(os.listdir(read_path)), "个样本")
    logging.info("有效样本", num)

二、训练模型

创建模型

# 创建网络
def create_model():
    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'])
    return model

训练模型

# 训练模型
def train_model(model):
    # 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_generator(
        train_generator,
        steps_per_epoch=60,
        epochs=12,
        validation_data=validation_generator,
        validation_steps=30)

    # 保存模型
    save_path = "../output/model"
    if not os.path.exists(save_path):
        os.makedirs(save_path)
    model.save(save_path + "/smileDetect.h5")
    return history

训练结果

准确率

丢失率

训练过程

三、预测

通过读取摄像头内容进行预测

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

效果

四、源代码

pretreatment.py

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
import shutil
from tqdm import tqdm
import logging

# dlib预测器
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor('../resources/shape_predictor_68_face_landmarks.dat')
# 原图片路径
read_path = "../resources/genki4k/files"
# 提取人脸存储路径
save_path = "../output/genki4k/files"
if not os.path.exists(save_path):
    os.makedirs(save_path)

# 新的数据集
data_dir = '../resources/data'
if not os.path.exists(data_dir):
    os.makedirs(data_dir)

# 训练集
train_dir = data_dir + "/train"
if not os.path.exists(train_dir):
    os.makedirs(train_dir)
# 验证集
validation_dir = os.path.join(data_dir, 'validation')
if not os.path.exists(validation_dir):
    os.makedirs(validation_dir)
# 测试集
test_dir = os.path.join(data_dir, 'test')
if not os.path.exists(test_dir):
    os.makedirs(test_dir)


# 初始化训练数据
def init_data(file_list):
    # 如果不存在文件夹则新建
    for file_path in file_list:
        if not os.path.exists(file_path):
            os.makedirs(file_path)
        # 存在则清空里面所有数据
        else:
            for i in os.listdir(file_path):
                path = os.path.join(file_path, i)
                if os.path.isfile(path):
                    os.remove(path)


def init_file():
    num = 0
    bar = tqdm(os.listdir(read_path))
    for file_name in bar:
        bar.desc = "预处理图片: "
        # a图片的全路径
        img_path = (read_path + "/" + file_name)
        # 读入的图片的路径中含非英文
        img = cv2.imdecode(np.fromfile(img_path, 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

        # dlib检测
        dets = detector(img, 1)
        for k, d in enumerate(dets):
            if len(dets) > 1:
                continue
            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)
            # 保存图片
            cv2.imencode('.jpg', img_blank)[1].tofile(save_path + "/" + "file" + str(num) + ".jpg")

    logging.info("一共", len(os.listdir(read_path)), "个样本")
    logging.info("有效样本", num)


# 划分数据集
def divide_data(file_path, message, begin, end):
    files = ['file{}.jpg'.format(i) for i in range(begin, end)]
    bar = tqdm(files)
    bar.desc = message
    for file in bar:
        src = os.path.join(save_path, file)
        dst = os.path.join(file_path, file)
        shutil.copyfile(src, dst)


if __name__ == "__main__":
    init_file()

    positive_train_dir = os.path.join(train_dir, 'smile')
    negative_train_dir = os.path.join(train_dir, 'unSmile')
    positive_validation_dir = os.path.join(validation_dir, 'smile')
    negative_validation_dir = os.path.join(validation_dir, 'unSmile')
    positive_test_dir = os.path.join(test_dir, 'smile')
    negative_test_dir = os.path.join(test_dir, 'unSmile')
    file_list = [positive_train_dir, positive_validation_dir, positive_test_dir,
                 negative_train_dir, negative_validation_dir, negative_test_dir]

    init_data(file_list)

    divide_data(positive_train_dir, "划分训练集正样本", 1, 1001)
    divide_data(negative_train_dir, "划分训练集负样本", 2200, 3200)
    divide_data(positive_validation_dir, "划分验证集正样本", 1000, 1500)
    divide_data(negative_validation_dir, "划分验证集负样本", 3000, 3500)
    divide_data(positive_test_dir, "划分测试集正样本", 1500, 2000)
    divide_data(negative_test_dir, "划分测试集负样本", 2800, 3500)

train.py

import os
from keras import layers
from keras import models
from tensorflow import optimizers
import matplotlib.pyplot as plt
from keras.preprocessing.image import ImageDataGenerator

train_dir = "../resources/data/train"
validation_dir = "../resources/data/validation"


# 创建网络
def create_model():
    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'])
    return model


# 训练模型
def train_model(model):
    # 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_generator(
        train_generator,
        steps_per_epoch=60,
        epochs=12,
        validation_data=validation_generator,
        validation_steps=30)

    # 保存模型
    save_path = "../output/model"
    if not os.path.exists(save_path):
        os.makedirs(save_path)
    model.save(save_path + "/smileDetect.h5")
    return history


# 展示训练结果
def show_results(history):
    # 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']

    # 绘制结果
    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()


if __name__ == "__main__":
    model = create_model()

    history = train_model(model)

    show_results(history)

predict.py

import os
from keras import layers
from keras import models
from tensorflow import optimizers
import matplotlib.pyplot as plt
from keras.preprocessing.image import ImageDataGenerator

train_dir = "../resources/data/train"
validation_dir = "../resources/data/validation"


# 创建网络
# 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('../output/model/smileDetect.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()