TensorFlow基础，TensorFlow简单人脸识别

TensorFlow基础，TensorFlow简单人脸识别

数据材料：

这是一个小型的人脸数据库，一共有40个人，每个人有10张照片作为样本数据。这些图片都是黑白照片，意味着这些图片都只有灰度0-255，没有rgb三通道。于是我们需要对这张大图片切分成一个个的小脸。整张图片大小是1190 × 942，一共有20 × 20张照片。那么每张照片的大小就是（1190 / 20）× （942 / 20）= 57 × 47 （大约，以为每张图片之间存在间距）。

问题解决：

10类样本，利用CNN训练可以分类10类数据的神经网络，与手写字符识别类似

#coding=utf-8

#http://www.jianshu.com/p/3e5ddc44aa56

#tensorflow 1.3.1

#python 3.6

import os

import numpy as np

import tensorflow as tf

import matplotlib.pyplot as plt

import matplotlib.image as mpimg

import matplotlib.patches as patches

import numpy

from PIL import Image

#获取dataset

def load_data(dataset_path):

    img = Image.open(dataset_path)

    # 定义一个20 × 20的训练样本，一共有40个人，每个人都10张样本照片

    img_ndarray = np.asarray(img, dtype='float64') / 256

    #img_ndarray = np.asarray(img, dtype='float32') / 32

    # 记录脸数据矩阵，57 * 47为每张脸的像素矩阵

    faces = np.empty((400, 57 * 47))

    for row in range(20):

        for column in range(20):

            faces[20 * row + column] = np.ndarray.flatten(

                img_ndarray[row * 57: (row + 1) * 57, column * 47 : (column + 1) * 47]

            )

    label = np.zeros((400, 40))

    for i in range(40):

        label[i * 10: (i + 1) * 10, i] = 1

    # 将数据分成训练集，验证集，测试集

    train_data = np.empty((320, 57 * 47))

    train_label = np.zeros((320, 40))

    vaild_data = np.empty((40, 57 * 47))

    vaild_label = np.zeros((40, 40))

    test_data = np.empty((40, 57 * 47))

    test_label = np.zeros((40, 40))

    for i in range(40):

        train_data[i * 8: i * 8 + 8] = faces[i * 10: i * 10 + 8]

        train_label[i * 8: i * 8 + 8] = label[i * 10: i * 10 + 8]

        vaild_data[i] = faces[i * 10 + 8]

        vaild_label[i] = label[i * 10 + 8]

        test_data[i] = faces[i * 10 + 9]

        test_label[i] = label[i * 10 + 9]

    train_data = train_data.astype('float32')

    vaild_data = vaild_data.astype('float32')

    test_data = test_data.astype('float32')

    return [

        (train_data, train_label),

        (vaild_data, vaild_label),

        (test_data, test_label)

    ]

def convolutional_layer(data, kernel_size, bias_size, pooling_size):

    kernel = tf.get_variable("conv", kernel_size, initializer=tf.random_normal_initializer())

    bias = tf.get_variable('bias', bias_size, initializer=tf.random_normal_initializer())

    conv = tf.nn.conv2d(data, kernel, strides=[1, 1, 1, 1], padding='SAME')

    linear_output = tf.nn.relu(tf.add(conv, bias))

    pooling = tf.nn.max_pool(linear_output, ksize=pooling_size, strides=pooling_size, padding="SAME")

    return pooling

def linear_layer(data, weights_size, biases_size):

    weights = tf.get_variable("weigths", weights_size, initializer=tf.random_normal_initializer())

    biases = tf.get_variable("biases", biases_size, initializer=tf.random_normal_initializer())

    return tf.add(tf.matmul(data, weights), biases)

def convolutional_neural_network(data):

    # 根据类别个数定义最后输出层的神经元

    n_ouput_layer = 40

    kernel_shape1=[5, 5, 1, 32]

    kernel_shape2=[5, 5, 32, 64]

    full_conn_w_shape = [15 * 12 * 64, 1024]

    out_w_shape = [1024, n_ouput_layer]

    bias_shape1=[32]

    bias_shape2=[64]

    full_conn_b_shape = [1024]

    out_b_shape = [n_ouput_layer]

    data = tf.reshape(data, [-1, 57, 47, 1])

    # 经过第一层卷积神经网络后，得到的张量shape为：[batch, 29, 24, 32]

    with tf.variable_scope("conv_layer1") as layer1:

        layer1_output = convolutional_layer(

            data=data,

            kernel_size=kernel_shape1,

            bias_size=bias_shape1,

            pooling_size=[1, 2, 2, 1]

        )

    # 经过第二层卷积神经网络后，得到的张量shape为：[batch, 15, 12, 64]

    with tf.variable_scope("conv_layer2") as layer2:

        layer2_output = convolutional_layer(

            data=layer1_output,

            kernel_size=kernel_shape2,

            bias_size=bias_shape2,

            pooling_size=[1, 2, 2, 1]

        )

    with tf.variable_scope("full_connection") as full_layer3:

        # 讲卷积层张量数据拉成2-D张量只有有一列的列向量

        layer2_output_flatten = tf.contrib.layers.flatten(layer2_output)

        layer3_output = tf.nn.relu(

            linear_layer(

                data=layer2_output_flatten,

                weights_size=full_conn_w_shape,

                biases_size=full_conn_b_shape

            )

        )

        # layer3_output = tf.nn.dropout(layer3_output, 0.8)

    with tf.variable_scope("output") as output_layer4:

        output = linear_layer(

            data=layer3_output,

            weights_size=out_w_shape,

            biases_size=out_b_shape

        )

    return output;

def train_facedata(dataset, model_dir,model_path):

    # train_set_x = data[0][0]

    # train_set_y = data[0][1]

    # valid_set_x = data[1][0]

    # valid_set_y = data[1][1]

    # test_set_x = data[2][0]

    # test_set_y = data[2][1]

    # X = tf.placeholder(tf.float32, shape=(None, None), name="x-input")  # 输入数据

    # Y = tf.placeholder(tf.float32, shape=(None, None), name='y-input')  # 输入标签

    batch_size = 40

    # train_set_x, train_set_y = dataset[0]

    # valid_set_x, valid_set_y = dataset[1]

    # test_set_x, test_set_y = dataset[2]

    train_set_x = dataset[0][0]

    train_set_y = dataset[0][1]

    valid_set_x = dataset[1][0]

    valid_set_y = dataset[1][1]

    test_set_x = dataset[2][0]

    test_set_y = dataset[2][1]

    X = tf.placeholder(tf.float32, [batch_size, 57 * 47])

    Y = tf.placeholder(tf.float32, [batch_size, 40])

    predict = convolutional_neural_network(X)

    cost_func = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=predict, labels=Y))

    optimizer = tf.train.AdamOptimizer(1e-2).minimize(cost_func)

    # 用于保存训练的最佳模型

    saver = tf.train.Saver()

    #model_dir = './model'

    #model_path = model_dir + '/best.ckpt'

    with tf.Session() as session:

        # 若不存在模型数据，需要训练模型参数

        if not os.path.exists(model_path + ".index"):

            session.run(tf.global_variables_initializer())

            best_loss = float('Inf')

            for epoch in range(20):

                epoch_loss = 0

                for i in range((int)(np.shape(train_set_x)[0] / batch_size)):

                    x = train_set_x[i * batch_size: (i + 1) * batch_size]

                    y = train_set_y[i * batch_size: (i + 1) * batch_size]

                    _, cost = session.run([optimizer, cost_func], feed_dict={X: x, Y: y})

                    epoch_loss += cost

                print(epoch, ' : ', epoch_loss)

                if best_loss > epoch_loss:

                    best_loss = epoch_loss

                    if not os.path.exists(model_dir):

                        os.mkdir(model_dir)

                        print("create the directory: %s" % model_dir)

                    save_path = saver.save(session, model_path)

                    print("Model saved in file: %s" % save_path)

        # 恢复数据并校验和测试

        saver.restore(session, model_path)

        correct = tf.equal(tf.argmax(predict,1), tf.argmax(Y,1))

        valid_accuracy = tf.reduce_mean(tf.cast(correct,'float'))

        print('valid set accuracy: ', valid_accuracy.eval({X: valid_set_x, Y: valid_set_y}))

        test_pred = tf.argmax(predict, 1).eval({X: test_set_x})

        test_true = np.argmax(test_set_y, 1)

        test_correct = correct.eval({X: test_set_x, Y: test_set_y})

        incorrect_index = [i for i in range(np.shape(test_correct)[0]) if not test_correct[i]]

        for i in incorrect_index:

            print('picture person is %i, but mis-predicted as person %i'

                %(test_true[i], test_pred[i]))

        plot_errordata(incorrect_index, "olivettifaces.gif")

#画出在测试集中错误的数据

def plot_errordata(error_index, dataset_path):

    img = mpimg.imread(dataset_path)

    plt.imshow(img)

    currentAxis = plt.gca()

    for index in error_index:

        row = index // 2

        column = index % 2

        currentAxis.add_patch(

            patches.Rectangle(

                xy=(

                    47 * 9 if column == 0 else 47 * 19,

                    row * 57

                    ),

                width=47,

                height=57,

                linewidth=1,

                edgecolor='r',

                facecolor='none'

            )

    )

    plt.savefig("result.png")

    plt.show()

def main():

    dataset_path = "olivettifaces.gif"

    data = load_data(dataset_path)

    model_dir = './model'

    model_path = model_dir + '/best.ckpt'

    train_facedata(data, model_dir, model_path)

if __name__ == "__main__" :

    main()

C:\python36\python.exe X:/DeepLearning/code/face/TensorFlow_CNN_face/facerecognition_main.py

valid set accuracy:  0.825

picture person is 0, but mis-predicted as person 23

picture person is 6, but mis-predicted as person 38

picture person is 8, but mis-predicted as person 34

picture person is 15, but mis-predicted as person 11

picture person is 24, but mis-predicted as person 7

picture person is 29, but mis-predicted as person 7

picture person is 33, but mis-predicted as person 39

Author：Maddock转载请注明出处：Maddock 计算机视觉、图像处理、机器学习

FlyAI是为AI开发者提供数据竞赛并支持GPU离线训练的一站式服务平台。每周免费提供项目开源算法样例，支持算法能力变现以及快速的迭代算法模型。更多TensorFlow实战项目都在FlyAI!

获取更多项目样例开源代码 请访问：https://www.flyai.com