TensorFlow和Keras完成JAFFE人脸表情识别

cut_save_face.py

#!/usr/bin/python
# coding:utf8
import cv2
import os
import numpy as np
import csv
def detect(img, cascade):
    """
    使用Haar特征检测分类器完成人脸检测
    :param img:
    :param cascade:
    :return:
    """
    # detectMultiScale检测出图片中所有的人脸，并将人脸用vector保存各个人脸的坐标、大小（用矩形表示），返回坐标。
    rects = cascade.detectMultiScale(img, scaleFactor=1.3, minNeighbors=4, minSize=(30, 30),
                                     flags=cv2.CASCADE_SCALE_IMAGE)
    if len(rects) == 0:
        return []
    rects[:, 2:] += rects[:, :2]
    return rects
cascade = cv2.CascadeClassifier(
    "E:/Anaconda3/envs/sklearn/Library/etc/haarcascades/haarcascade_frontalface_alt.xml")
f = "jaffe/"
fs = os.listdir(f)
data = np.zeros([213, 48 * 48], dtype=np.uint8)
label = np.zeros([213], dtype=int)
i = 0
for f1 in fs:
    tmp_path = os.path.join(f, f1)
    if not os.path.isdir(tmp_path):
        # print(tmp_path[len(f):])
        img = cv2.imread(tmp_path, 1)
        # BGR转换为灰度图
        dst = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
        rects = detect(dst, cascade)
        for x1, y1, x2, y2 in rects:
            cv2.rectangle(img, (x1 + 10, y1 + 20), (x2 - 10, y2), (0, 255, 255), 2)
            # 调整截取脸部区域大小
            img_roi = np.uint8([y2 - (y1 + 20), (x2 - 10) - (x1 + 10)])
            roi = dst[y1 + 20:y2, x1 + 10:x2 - 10]
            img_roi = roi
            re_roi = cv2.resize(img_roi, (48, 48))
            # 获得表情label
            img_label = tmp_path[len(f) + 3:len(f) + 5]
            # print(img_label)
            if img_label == 'AN':
                label[i] = 0
            elif img_label == 'DI':
                label[i] = 1
            elif img_label == 'FE':
                label[i] = 2
            elif img_label == 'HA':
                label[i] = 3
            elif img_label == 'SA':
                label[i] = 4
            elif img_label == 'SU':
                label[i] = 5
            elif img_label == 'NE':
                label[i] = 6
            else:
                print("get label error.......\n")
            # flatten返回一个折叠成一维的数组。但是该函数只能适用于numpy对象，即array或者mat，普通的list列表是不行的。
            data[i][0:48 * 48] = np.ndarray.flatten(re_roi)
            i = i + 1
            # cv2.imshow("src", dst)
            # cv2.imshow("img", img)
            # if cv2.waitKey() == 32:
            #     continue
with open(r"face.csv", "w") as csvfile:
    writer = csv.writer(csvfile)
    writer.writerow(['emotion', 'pixels'])
    for i in range(len(label)):
        data_list = list(data[i])
        b = " ".join(str(x) for x in data_list)
        # 在水平方向上平铺
        l = np.hstack([label[i], b])
        writer.writerow(l)

detectMultiScale(img, scaleFactor=1.3, minNeighbors=4, minSize=(30, 30),
                                     flags=cv2.CASCADE_SCALE_IMAGE)

参数image--待检测图片，一般为灰度图像加快检测速度；
参数scaleFactor--表示在前后两次相继的扫描中，搜索窗口的比例系数。默认为1.1即每次搜索窗口依次扩大10%;
参数minNeighbors--表示构成检测目标的相邻矩形的最小个数(默认为3个)。如果组成检测目标的小矩形的个数和小于 min_neighbors - 1 都会被排除。如果min_neighbors 为 0, 则函数不做任何操作就返回所有的被检候选矩形框，这种设定值一般用在用户自定义对检测结果的组合程序上；
参数flags--flags=0：可以取如下这些值：CASCADE_DO_CANNY_PRUNING=1,利用canny边缘检测来排除一些边缘很少或者很多的图像区域，CASCADE_SCALE_IMAGE=2,正常比例检测，CASCADE_FIND_BIGGEST_OBJECT=4,只检测最大的物体，CASCADE_DO_ROUGH_SEARCH=8 初略的检测

show_facial_expression.py

#!/usr/bin/python
# coding:utf8
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
emotion = {0: 'Angry', 1: 'Disgust', 2: 'Fear', 3: 'Happy', 4: 'Sad', 5: 'Surprise', 6: 'Neutral'}
data = pd.read_csv(r'face.csv', dtype='a')
label = np.array(data['emotion'])
img_data = np.array(data['pixels'])
N_sample = label.size
Face_data = np.zeros((N_sample, 48 * 48))
Face_label = np.zeros((N_sample, 7), dtype=int)
# 显示人脸以及对应表情
for i in range(25):
    x = img_data[i]
    # 使用字符串创建矩阵。
    x = np.fromstring(x, dtype=float, sep=' ')
    x = x / x.max()
    img_x = np.reshape(x, (48, 48))
    plt.subplot(5, 5, i + 1)
    plt.axis('off')
    plt.title(emotion[int(label[i])])
    plt.imshow(img_x, plt.cm.gray)
plt.show()

count_facial_expression.py

#!/usr/bin/python
# coding:utf8
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
emotion = {0: 'Angry', 1: 'Disgust', 2: 'Fear', 3: 'Happy', 4: 'Sad', 5: 'Surprise', 6: 'Neutral'}
data = pd.read_csv(r'face.csv', dtype='a')
label = np.array(data['emotion'])
img_data = np.array(data['pixels'])
N_sample = label.size
emotions = np.zeros(7)
for i in label:
    for j in range(7):
        if int(i) == j:
            emotions[j] = emotions[j] + 1
print(emotions)
plt.bar(range(7), emotions, 0.5, color=['red', 'green', 'blue'])
plt.xlabel('emotions')
plt.xticks(range(7), ['Angry', 'Disgust', 'Fear', 'Happy', 'Sad', 'Surprise', 'Neutral'], rotation=0)
plt.ylabel('number')
plt.grid()
plt.show()

train_keras.py

#!/usr/bin/python
# coding:utf8
import numpy as np
import pandas as pd
from keras.layers import Dense, Conv2D, MaxPooling2D, Dropout, Flatten
from keras.models import Sequential
from keras.preprocessing.image import ImageDataGenerator
# 表情类别
emotion = {0: 'Angry', 1: 'Disgust', 2: 'Fear', 3: 'Happy', 4: 'Sad', 5: 'Surprise', 6: 'Neutral'}
# 读取数据
data = pd.read_csv(r'face.csv', dtype='a')
# 读取标签列表
label = np.array(data['emotion'])
# 图像列表
img_data = np.array(data['pixels'])
# 图像数量
N_sample = label.size
# (213, 2304)
Face_data = np.zeros((N_sample, 48 * 48))
# (213, 7)
Face_label = np.zeros((N_sample, 7), dtype=np.float)
for i in range(N_sample):
    x = img_data[i]
    x = np.fromstring(x, dtype=float, sep=' ')
    x = x / x.max()
    Face_data[i] = x
    Face_label[i, int(label[i])] = 1.0
# 训练数据数量
train_num = 200
# 测试数据数量
test_num = 13
# 训练数据
train_x = Face_data[0:train_num, :]
train_y = Face_label[0:train_num, :]
train_x = train_x.reshape(-1, 48, 48, 1)  # reshape
# 测试数据
test_x = Face_data[train_num: train_num + test_num, :]
test_y = Face_label[train_num: train_num + test_num, :]
test_x = test_x.reshape(-1, 48, 48, 1)  # reshape
# 序贯模型
model = Sequential()
model.add(Conv2D(32, (5, 5), activation='relu', input_shape=(48, 48, 1)))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(64, (5, 5), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dense(1024, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(7, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
# 扩增数据
datagen = ImageDataGenerator(
    featurewise_center=True,
    featurewise_std_normalization=True,
    rotation_range=20,
    width_shift_range=0.2,
    height_shift_range=0.2,
    horizontal_flip=True)
datagen.fit(train_x)
model.fit_generator(datagen.flow(train_x, train_y, batch_size=10), steps_per_epoch=len(train_x), epochs=20)
model.fit(train_x, train_y, batch_size=10, epochs=100)
score = model.evaluate(test_x, test_y, batch_size=10)
print("score:", score)
model.summary()

输出结果

Epoch 50/50
 10/200 [>.............................] - ETA: 1s - loss: 0.0015 - acc: 1.0000
 20/200 [==>...........................] - ETA: 1s - loss: 7.8377e-04 - acc: 1.0000
 30/200 [===>..........................] - ETA: 1s - loss: 0.0013 - acc: 1.0000
 40/200 [=====>........................] - ETA: 1s - loss: 0.0012 - acc: 1.0000
 50/200 [======>.......................] - ETA: 1s - loss: 0.0013 - acc: 1.0000
 60/200 [========>.....................] - ETA: 1s - loss: 0.0011 - acc: 1.0000
 70/200 [=========>....................] - ETA: 1s - loss: 0.0013 - acc: 1.0000
 80/200 [===========>..................] - ETA: 1s - loss: 0.0012 - acc: 1.0000
 90/200 [============>.................] - ETA: 0s - loss: 0.0011 - acc: 1.0000
100/200 [==============>...............] - ETA: 0s - loss: 0.0011 - acc: 1.0000
110/200 [===============>..............] - ETA: 0s - loss: 9.7286e-04 - acc: 1.0000
120/200 [=================>............] - ETA: 0s - loss: 8.9958e-04 - acc: 1.0000
130/200 [==================>...........] - ETA: 0s - loss: 8.3767e-04 - acc: 1.0000
140/200 [====================>.........] - ETA: 0s - loss: 9.1249e-04 - acc: 1.0000
150/200 [=====================>........] - ETA: 0s - loss: 9.3190e-04 - acc: 1.0000
160/200 [=======================>......] - ETA: 0s - loss: 9.0101e-04 - acc: 1.0000
170/200 [========================>.....] - ETA: 0s - loss: 8.6291e-04 - acc: 1.0000
180/200 [==========================>...] - ETA: 0s - loss: 8.2539e-04 - acc: 1.0000
190/200 [===========================>..] - ETA: 0s - loss: 7.8394e-04 - acc: 1.0000
200/200 [==============================] - 2s 9ms/step - loss: 7.8767e-04 - acc: 1.0000
10/13 [======================>.......] - ETA: 0s
13/13 [==============================] - 0s 5ms/step
_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
conv2d_1 (Conv2D)            (None, 44, 44, 32)        832
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 22, 22, 32)        0
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 18, 18, 64)        51264
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 9, 9, 64)          0
_________________________________________________________________
flatten_1 (Flatten)          (None, 5184)              0
_________________________________________________________________
dense_1 (Dense)              (None, 1024)              5309440
_________________________________________________________________
dropout_1 (Dropout)          (None, 1024)              0
_________________________________________________________________
dense_2 (Dense)              (None, 7)                 7175
=================================================================
Total params: 5,368,711
Trainable params: 5,368,711
Non-trainable params: 0
_________________________________________________________________

train_tensorflow.py

#!/usr/bin/python
# coding:utf8
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import tensorflow as tf
def conv_pool_layer(data, weights_size, biases_size):
    """
    卷积神经网络的层
    :param data:
    :param weights_size:
    :param biases_size:
    :return:
    """
    weights = tf.Variable(tf.truncated_normal(weights_size, stddev=0.1))
    biases = tf.Variable(tf.constant(0.1, shape=biases_size))
    conv2d = tf.nn.conv2d(data, weights, strides=[1, 1, 1, 1], padding='SAME')
    relu = tf.nn.relu(conv2d + biases)
    return tf.nn.max_pool(relu, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
def linear_layer(data, weights_size, biases_size):
    weights = tf.Variable(tf.truncated_normal(weights_size, stddev=0.1))
    biases = tf.Variable(tf.constant(0.1, shape=biases_size))
    return tf.add(tf.matmul(data, weights), biases)
def convolutional_neural_network(x, keep_prob):
    """
    卷积神经网络
    :param x:
    :param keep_prob:
    :return:
    """
    x_image = tf.reshape(x, [-1, 48, 48, 1])
    h_pool1 = conv_pool_layer(x_image, [5, 5, 1, 32], [32])
    h_pool2 = conv_pool_layer(h_pool1, [5, 5, 32, 64], [64])
    h_pool2_flat = tf.reshape(h_pool2, [-1, 12 * 12 * 64])
    h_fc1 = tf.nn.relu(linear_layer(h_pool2_flat, [12 * 12 * 64, 1024], [1024]))
    h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
    return tf.nn.softmax(linear_layer(h_fc1_drop, [1024, class_sum], [class_sum]))
def batch_data(x, y, batch, num):
    ind = np.arange(num)
    index = ind[batch * batch_size:(batch + 1) * batch_size]
    batch_x = x[index, :]
    batch_y = y[index, :]
    return batch_x, batch_y
if __name__ == "__main__":
    # 表情类别
    emotion = {0: 'Angry', 1: 'Disgust', 2: 'Fear', 3: 'Happy', 4: 'Sad', 5: 'Surprise', 6: 'Neutral'}
    # 数据
    data = pd.read_csv(r'face.csv', dtype='a')
    # 表情标签
    label = np.array(data['emotion'])
    # 图片数据
    img_data = np.array(data['pixels'])
    # 图片数量
    N_sample = label.size
    # 图片矩阵
    Face_data = np.zeros((N_sample, 48 * 48))
    # 标签矩阵
    Face_label = np.zeros((N_sample, 7), dtype=int)
    # 遍历将图片和标签构建为矩阵形式
    for i in range(N_sample):
        x = img_data[i]
        x = np.fromstring(x, dtype=float, sep=' ')
        x = x / x.max()
        Face_data[i] = x
        Face_label[i, int(label[i])] = 1
    # 参数
    dropout = 0.5
    class_sum = 7
    # dropout减轻过拟合问题
    keep_prob = tf.placeholder(tf.float32)
    x = tf.placeholder(tf.float32, [None, 48 * 48])
    y = tf.placeholder(tf.float32, [None, class_sum])
    # 构建卷积神经网络
    pred = convolutional_neural_network(x, keep_prob)
    # 取前200个作为训练数据，后13个为测试数据
    train_num = 200
    test_num = 13
    train_x = Face_data[0:train_num, :]
    train_y = Face_label[0:train_num, :]
    test_x = Face_data[train_num: train_num + test_num, :]
    test_y = Face_label[train_num: train_num + test_num, :]
    batch_size = 20
    train_batch_num = int(train_num / batch_size)
    test_batch_num = test_num / batch_size
    # 训练和评估模型
    cross_entropy = -tf.reduce_sum(y * tf.log(pred))
    train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
    correct_pred = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_pred, "float"))
    total_train_loss = []
    total_train_acc = []
    total_test_loss = []
    total_test_acc = []
    train_epoch = 50
    with tf.Session() as sess:
        sess.run(tf.initialize_all_variables())
        for epoch in range(0, train_epoch):
            Total_train_loss = 0
            Total_train_acc = 0
            for train_batch in range(0, train_batch_num):
                batch_x, batch_y = batch_data(train_x, train_y, train_batch, train_num)
                # 优化操作
                sess.run(train_step, feed_dict={x: batch_x, y: batch_y, keep_prob: dropout})
                if train_batch % batch_size == 0:
                    # 计算损失和准确率
                    loss, acc = sess.run([cross_entropy, accuracy], feed_dict={x: batch_x, y: batch_y, keep_prob: 1.})
                    print("Epoch: " + str(epoch + 1) + ", Batch: " + str(train_batch) + ", Loss= " + "{:.3f}".format(
                        loss) + ", Training Accuracy= " + "{:.3f}".format(acc))
                    Total_train_loss = Total_train_loss + loss
                    Total_train_acc = Total_train_acc + acc
            total_train_loss.append(Total_train_loss)
            total_train_acc.append(Total_train_acc)
    plt.subplot(2, 1, 1)
    plt.ylabel('Train loss')
    plt.plot(total_train_loss, 'r')
    plt.subplot(2, 1, 2)
    plt.ylabel('Train accuracy')
    plt.plot(total_train_acc, 'r')
    plt.savefig("loss_acc.png")
    plt.show()

输出结果

Epoch: 1, Batch: 0, Loss= 119.417, Training Accuracy= 0.150
Epoch: 2, Batch: 0, Loss= 49.221, Training Accuracy= 0.250
Epoch: 3, Batch: 0, Loss= 39.813, Training Accuracy= 0.200
Epoch: 4, Batch: 0, Loss= 43.067, Training Accuracy= 0.300
Epoch: 5, Batch: 0, Loss= 28.889, Training Accuracy= 0.500
Epoch: 6, Batch: 0, Loss= 18.173, Training Accuracy= 0.700
Epoch: 7, Batch: 0, Loss= 18.474, Training Accuracy= 0.600
Epoch: 8, Batch: 0, Loss= 15.813, Training Accuracy= 0.800
Epoch: 9, Batch: 0, Loss= 12.878, Training Accuracy= 0.850
Epoch: 10, Batch: 0, Loss= 13.250, Training Accuracy= 0.800
Epoch: 11, Batch: 0, Loss= 8.750, Training Accuracy= 0.950
Epoch: 12, Batch: 0, Loss= 10.424, Training Accuracy= 0.850
Epoch: 13, Batch: 0, Loss= 7.592, Training Accuracy= 1.000
Epoch: 14, Batch: 0, Loss= 7.146, Training Accuracy= 0.950
Epoch: 15, Batch: 0, Loss= 7.377, Training Accuracy= 1.000
Epoch: 16, Batch: 0, Loss= 5.423, Training Accuracy= 1.000
Epoch: 17, Batch: 0, Loss= 6.173, Training Accuracy= 1.000
Epoch: 18, Batch: 0, Loss= 4.069, Training Accuracy= 1.000
Epoch: 19, Batch: 0, Loss= 4.163, Training Accuracy= 1.000
Epoch: 20, Batch: 0, Loss= 3.650, Training Accuracy= 1.000
Epoch: 21, Batch: 0, Loss= 3.317, Training Accuracy= 1.000
Epoch: 22, Batch: 0, Loss= 4.195, Training Accuracy= 1.000
Epoch: 23, Batch: 0, Loss= 2.729, Training Accuracy= 1.000
Epoch: 24, Batch: 0, Loss= 2.448, Training Accuracy= 1.000
Epoch: 25, Batch: 0, Loss= 2.614, Training Accuracy= 1.000
Epoch: 26, Batch: 0, Loss= 2.424, Training Accuracy= 1.000
Epoch: 27, Batch: 0, Loss= 2.707, Training Accuracy= 1.000
Epoch: 28, Batch: 0, Loss= 2.072, Training Accuracy= 1.000
Epoch: 29, Batch: 0, Loss= 1.726, Training Accuracy= 1.000
Epoch: 30, Batch: 0, Loss= 1.701, Training Accuracy= 1.000
Epoch: 31, Batch: 0, Loss= 1.598, Training Accuracy= 1.000
Epoch: 32, Batch: 0, Loss= 1.381, Training Accuracy= 1.000
Epoch: 33, Batch: 0, Loss= 1.826, Training Accuracy= 1.000
Epoch: 34, Batch: 0, Loss= 1.227, Training Accuracy= 1.000
Epoch: 35, Batch: 0, Loss= 1.320, Training Accuracy= 1.000
Epoch: 36, Batch: 0, Loss= 1.110, Training Accuracy= 1.000
Epoch: 37, Batch: 0, Loss= 0.875, Training Accuracy= 1.000
Epoch: 38, Batch: 0, Loss= 1.214, Training Accuracy= 1.000
Epoch: 39, Batch: 0, Loss= 0.982, Training Accuracy= 1.000
Epoch: 40, Batch: 0, Loss= 0.982, Training Accuracy= 1.000
Epoch: 41, Batch: 0, Loss= 0.681, Training Accuracy= 1.000
Epoch: 42, Batch: 0, Loss= 0.839, Training Accuracy= 1.000
Epoch: 43, Batch: 0, Loss= 0.777, Training Accuracy= 1.000
Epoch: 44, Batch: 0, Loss= 0.671, Training Accuracy= 1.000
Epoch: 45, Batch: 0, Loss= 0.859, Training Accuracy= 1.000
Epoch: 46, Batch: 0, Loss= 0.529, Training Accuracy= 1.000
Epoch: 47, Batch: 0, Loss= 0.707, Training Accuracy= 1.000
Epoch: 48, Batch: 0, Loss= 0.491, Training Accuracy= 1.000
Epoch: 49, Batch: 0, Loss= 0.500, Training Accuracy= 1.000
Epoch: 50, Batch: 0, Loss= 0.415, Training Accuracy= 1.000

train_tensorboard.py

#!/usr/bin/python
# coding:utf8
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import tensorflow as tf
def conv_pool_layer(data, weights_size, biases_size):
    weights = tf.Variable(tf.truncated_normal(weights_size, stddev=0.1))
    biases = tf.Variable(tf.constant(0.1, shape=biases_size))
    conv2d = tf.nn.conv2d(data, weights, strides=[1, 1, 1, 1], padding='SAME')
    relu = tf.nn.relu(conv2d + biases)
    return tf.nn.max_pool(relu, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
def linear_layer(data, weights_size, biases_size):
    weights = tf.Variable(tf.truncated_normal(weights_size, stddev=0.1))
    biases = tf.Variable(tf.constant(0.1, shape=biases_size))
    return tf.add(tf.matmul(data, weights), biases)
def convolutional_neural_network(x, keep_prob):
    with tf.name_scope('input'):
        x_image = tf.reshape(x, [-1, 48, 48, 1])
    with tf.name_scope('conv1'):
        h_pool1 = conv_pool_layer(x_image, [5, 5, 1, 32], [32])
    with tf.name_scope('conv2'):
        h_pool2 = conv_pool_layer(h_pool1, [5, 5, 32, 64], [64])
    h_pool2_flat = tf.reshape(h_pool2, [-1, 12 * 12 * 64])
    with tf.name_scope('fc3'):
        h_fc1 = tf.nn.relu(linear_layer(h_pool2_flat, [12 * 12 * 64, 1024], [1024]))
    with tf.name_scope('dropout4'):
        h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
    with tf.name_scope('softmax5'):
        out = tf.nn.softmax(linear_layer(h_fc1_drop, [1024, class_sum], [class_sum]))
    return out
def batch_data(x, y, batch, num):
    ind = np.arange(num)
    index = ind[batch * batch_size:(batch + 1) * batch_size]
    batch_x = x[index, :]
    batch_y = y[index, :]
    return batch_x, batch_y
if __name__ == "__main__":
    emotion = {0: 'Angry', 1: 'Disgust', 2: 'Fear', 3: 'Happy', 4: 'Sad', 5: 'Surprise', 6: 'Neutral'}
    data = pd.read_csv(r'face.csv', dtype='a')
    label = np.array(data['emotion'])
    img_data = np.array(data['pixels'])
    N_sample = label.size
    Face_data = np.zeros((N_sample, 48 * 48))
    Face_label = np.zeros((N_sample, 7), dtype=int)
    for i in range(N_sample):
        x = img_data[i]
        x = np.fromstring(x, dtype=float, sep=' ')
        x = x / x.max()
        Face_data[i] = x
        Face_label[i, int(label[i])] = 1
    # 参数
    dropout = 0.5
    class_sum = 7
    # dropout减轻过拟合问题
    keep_prob = tf.placeholder(tf.float32)
    x = tf.placeholder(tf.float32, [None, 48 * 48])
    y = tf.placeholder(tf.float32, [None, class_sum])
    pred = convolutional_neural_network(x, keep_prob)
    # 取前200个作为训练数据，后13个为测试数据
    train_num = 200
    test_num = 13
    train_x = Face_data[0:train_num, :]
    train_y = Face_label[0:train_num, :]
    test_x = Face_data[train_num: train_num + test_num, :]
    test_y = Face_label[train_num: train_num + test_num, :]
    batch_size = 20
    train_batch_num = int(train_num / batch_size)
    test_batch_num = test_num / batch_size
    # 训练和评估模型
    with tf.name_scope('cross_entropy'):
        cross_entropy = -tf.reduce_sum(y * tf.log(pred))
        tf.summary.histogram("cross_entropy", cross_entropy)
        # tf.summary.scalar("cross_entropy", cross_entropy)
    with tf.name_scope('minimize'):
        train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
    with tf.name_scope('accuracy'):
        with tf.name_scope('correct_pred'):
            correct_pred = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
        with tf.name_scope('accuracy'):
            accuracy = tf.reduce_mean(tf.cast(correct_pred, "float"))
            tf.summary.histogram("accuracy", accuracy)
        # 输出包含单个标量值的摘要协议缓冲区
        tf.summary.scalar('accuracy', accuracy)
    total_train_loss = []
    total_train_acc = []
    total_test_loss = []
    total_test_acc = []
    train_epoch = 50
    # 合并在默认图形中收集的所有摘要
    merged = tf.summary.merge_all()
    with tf.Session() as sess:
        writer = tf.summary.FileWriter("tmp/logs", sess.graph)
        sess.run(tf.global_variables_initializer())
        for epoch in range(0, train_epoch):
            Total_train_loss = 0
            Total_train_acc = 0
            for train_batch in range(0, train_batch_num):
                batch_x, batch_y = batch_data(train_x, train_y, train_batch, train_num)
                # 优化操作
                # sess.run(train_step, feed_dict={x: batch_x, y: batch_y, keep_prob: dropout})
                summary, _ = sess.run([merged, train_step], feed_dict={x: batch_x, y: batch_y, keep_prob: dropout})
                writer.add_summary(summary, train_batch)
                if train_batch % batch_size == 0:
                    # 计算损失和准确率
                    loss, acc = sess.run([cross_entropy, accuracy], feed_dict={x: batch_x, y: batch_y, keep_prob: 1.})
                    print("Epoch: " + str(epoch + 1) + ", Batch: " + str(train_batch) +
                          ", Loss= " + "{:.3f}".format(loss) +
                          ", Training Accuracy= " + "{:.3f}".format(acc))
                    Total_train_loss = Total_train_loss + loss
                    Total_train_acc = Total_train_acc + acc
            total_train_loss.append(Total_train_loss)
            total_train_acc.append(Total_train_acc)
        writer.close()
    plt.subplot(2, 1, 1)
    plt.ylabel('Train loss')
    plt.plot(total_train_loss, 'r')
    plt.subplot(2, 1, 2)
    plt.ylabel('Train accuracy')
    plt.plot(total_train_acc, 'r')
    plt.savefig("face_loss_acc.png")
    plt.show()

TensorFlow和Keras完成JAFFE人脸表情识别的更多相关文章

CVPR 2020几篇论文内容点评：目标检测跟踪，人脸表情识别，姿态估计，实例分割等
CVPR 2020几篇论文内容点评:目标检测跟踪,人脸表情识别,姿态估计,实例分割等 CVPR 2020中选论文放榜后,最新开源项目合集也来了. 本届CPVR共接收6656篇论文,中选1470篇,&q ...
【Gabor】基于多尺度多方向Gabor融合+分块直方图的表情识别
Topic:表情识别Env: win10 + Pycharm2018 + Python3.6.8Date: 2019/6/23~25 by hw_Chen2018 ...
42 在Raspberry Pi上安装dlib表情识别
https://www.jianshu.com/p/848014d8dea9 https://www.pyimagesearch.com/2017/05/01/install-dlib-raspber ...
基于深度学习的人脸性别识别系统（含UI界面，Python代码）
摘要:人脸性别识别是人脸识别领域的一个热门方向,本文详细介绍基于深度学习的人脸性别识别系统,在介绍算法原理的同时,给出Python的实现代码以及PyQt的UI界面.在界面中可以选择人脸图片.视频进行检 ...
keras系列︱人脸表情分类与识别：opencv人脸检测+Keras情绪分类（四）
引自:http://blog.csdn.net/sinat_26917383/article/details/72885715 人脸识别热门,表情识别更加.但是表情识别很难,因为人脸的微表情很多,本节 ...
深度学习项目——基于卷积神经网络（CNN）的人脸在线识别系统
基于卷积神经网络(CNN)的人脸在线识别系统本设计研究人脸识别技术,基于卷积神经网络构建了一套人脸在线检测识别系统,系统将由以下几个部分构成: 制作人脸数据集.CNN神经网络模型训练.人脸检测.人脸 ...
机器学习： Tensor Flow with CNN 做表情识别
我们利用 TensorFlow 构造 CNN 做表情识别,我们用的是FER-2013 这个数据库, 这个数据库一共有 35887 张人脸图像,这里只是做一个简单到仿真实验,为了计算方便,我们用其中到 ...
UWP通过机器学习加载ONNX进行表情识别
首先我们先来说说这个ONNX ONNX是一种针对机器学习所设计的开放式的文件格式,用于存储训练好的模型.它使得不同的人工智能框架(如Pytorch, MXNet)可以采用相同格式存储模型数据并交互. ...
Python学习案例之视频人脸检测识别
前言上一篇博文与大家分享了简单的图片人脸识别技术,其实在实际应用中,很多是通过视频流的方式进行识别,比如人脸识别通道门禁考勤系统.人脸动态跟踪识别系统等等. 案例这里我们还是使用 opencv 中 ...

随机推荐

java学习之路--多线程实现的方法
1 继承Thread类继承Thread类的方法尽管被我列为一种多线程实现方式,但Thread本质上也是实现了Runnable接口的一个实例,它代表一个线程的实例,并且,启动线程的唯一方法就是通过Th ...
FTPService工具类
package com.vcredit.ddcash.server.commons.net; import com.vcredit.ddcash.server.commons.model.FtpPar ...
kubernetes组成
kubernetes组成 k8s主要包括: kubectl 客户端命令行工具: 将接收的命令,发送给kube-apiserver,作为对整个平台操作的入口. kube-apiserver REST A ...
逆向工程之修改关键CALL返回值_破解视频转换专家
1)注册软件随便输入注册名注册码 2)进入软件根目录,发送到PEID查壳 3)发现无壳 4)发送到OD 4.1)右键菜单选择智能搜索 4.2)找到关键信息点注册 4.3)找到关键信息点双击进入汇编,向 ...
C和C指针小记(十四)-字符串、字符和字节
1.字符串 C语言没有字符串数据类型,因为字符串以字符串常量的形式出现或存储于字符数组中. 字符串常量和适用于那些程序不会对他们进行修改的字符串. 所有其他字符串都必须存储于字符串数组或动态分配的内存 ...
PowerPoint使用技巧
1.右键Group两个元素,可以一起移动: 2.Insert 屏幕输入功能: 3.录制旁白: 4.录制完旁白之后可以生成视频: 5.如果不确定所有引用的组件是否可以在别的机器上使用,可以导出只CD,生 ...
caffe中的卷积
https://www.zhihu.com/question/28385679 如上,将三维的操作转换到二维上面去做,然后调用GEMM库进行矩阵间的运算得到最后结果. 两个矩阵相乘,需要中间的那个维度 ...
为什么分布式数据库中不使用uuid作为主键？
分布式数据库当然也有主键的需求,但是为什么不直接使用uuid作为主键呢?作为曾经被这个问题困惑过的人,试着回答一下 1. UUID生成速率低下 Java的UUID依赖于SecureRandom.nex ...
PHP中new self()和new static()的区别探究
1.new static()是在PHP5.3版本中引入的新特性. 2.无论是new static()还是new self(),都是new了一个新的对象. 3.这两个方法new出来的对象有什么区别呢,说 ...
python中的双冒号作用
Python序列切片地址可以写为[开始:结束:步长],其中的开始和结束可以省略. 1. range(n)生成[0,n)区间整数 2. 开始start省略时,默认从第0项开始 3. 结尾省略的时候,默认 ...

TensorFlow和Keras完成JAFFE人脸表情识别

TensorFlow和Keras完成JAFFE人脸表情识别的更多相关文章

随机推荐

热门专题