1. 验证码的识别案例分析
1.1. 识别原理分析（图）
1.2. 处理数据 图片 标签文件
1.3. 识别验证码
1.4. 代码实现
1.4.1. 把验证码图片和标签值写入到tfrecords文件中(测试中发现了一个问题 dense tensor？？)
import tensorflow as tf
import os
os.environ[‘TF_CPP_MIN_LOG_LEVEL’] = ‘2’
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_string(“tfrecords_dir”, “./tfrecords/captcha.tfrecords”, “验证码tfrecords文件”)
tf.app.flags.DEFINE_string(“captcha_dir”, “…/data/Genpics/”, “验证码图片路径”)
tf.app.flags.DEFINE_string(“letter”, “ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789”, “验证码字符的种类”)
tf.app.flags.DEFINE_integer(“sample_num”, 1263, “图片的样本数量”)
tf.app.flags.DEFINE_integer(“captcha_img_height”, 40, “验证码图片的高度”)
tf.app.flags.DEFINE_integer(“captcha_img_width”, 135, “验证码图片的宽度”)
tf.app.flags.DEFINE_integer(“captcha_img_channel”, 3, “验证码图片的通道（彩色图片还是黑白图片）”)
def dealwithlabel(label_str):
# 构建字符索引 {0：'A', 1:'B'......}
num_letter = dict(enumerate(list(FLAGS.letter)))
# 键值对反转 {'A':0, 'B':1......}
letter_num = dict(zip(num_letter.values(), num_letter.keys()))
print(letter_num)
# 构建标签的列表
array = []
# 给标签数据进行处理[[b"NZPP"]......]
for string in label_str:
letter_list = [] # [1,2,3,4]
# 修改编码，b'FVQJ'到字符串，并且循环找到每张验证码的字符对应的数字标记
for letter in string.decode('utf-8'):
letter_list.append(letter_num[letter])
array.append(letter_list)
# [[13, 25, 15, 15], [22, 10, 7, 10], [22, 15, 18, 9], [16, 6, 13, 10], [1, 0, 8, 17], [0, 9, 24, 14].....]
print(array)
# 将array转换成tensor类型
label = tf.constant(array)
return label
def get_captcha_image():
“”"
获取验证码图片数据
file_list: 路径 +文件名列表
return: image
“”"
# 构造文件名
filename = []
for i in range(FLAGS.sample_num):
string = str(i) + ".jpg"
filename.append(string)
# 构造路径+文件
file_list = [os.path.join(FLAGS.captcha_dir, file) for file in filename]
# 构造文件队列
file_queue = tf.train.string_input_producer(file_list, shuffle=False)
# 构造阅读器
reader = tf.WholeFileReader()
# 读取图片数据内容
key, value = reader.read(file_queue)
# 解码图片数据
image = tf.image.decode_jpeg(value)
image.set_shape([FLAGS.captcha_img_height, FLAGS.captcha_img_width, FLAGS.captcha_img_channel])
# 批处理数据 [1263, 40, 135, 3]
image_batch = tf.train.batch([image], batch_size=FLAGS.sample_num, num_threads=1, capacity=6000)
return image_batch
def get_captcha_label():
“”"
读取验证码图片标签数据
:return: label
“”"
file_queue = tf.train.string_input_producer(["…/data/Genpics/labels.csv"], shuffle=False)
reader = tf.TextLineReader()
key, value = reader.read(file_queue)
records = [[1], ["None"]]
number, label = tf.decode_csv(value, record_defaults=records)
# [["NZPP"], ["WKHK"], ["ASDY"]]
label_batch = tf.train.batch([label], batch_size=6000, num_threads=1, capacity=6000)
return label_batch
def write_to_tfrecords(image_batch, label_batch):
“”"
将图片内容和标签写入到tfrecords文件当中
:param image_batch: 特征值
:param label_batch: 标签纸
:return: None
“”"
# 转换类型
label_batch = tf.cast(label_batch, tf.uint8)
print(label_batch)
# 建立TFRecords 存储器
writer = tf.python_io.TFRecordWriter(FLAGS.tfrecords_dir)
# 循环将每一个图片上的数据构造example协议块，序列化后写入
for i in range(1263):
# 取出第i个图片数据，转换相应类型,图片的特征值要转换成字符串形式
image_string = image_batch[i].eval().tostring()
# 标签值，转换成相应的类型
label_string = label_batch[i].eval().tostring()
# 构造协议块
example = tf.train.Example(features=tf.train.Features(feature={
"image": tf.train.Feature(bytes_list=tf.train.BytesList(value=[image_string])),
"label": tf.train.Feature(bytes_list=tf.train.BytesList(value=[label_string]))
}))
writer.write(example.SerializeToString())
# 关闭文件
writer.close()
return None
if name == “main”:
# 获取验证码文件当中的图片
image_batch = get_captcha_image()
# 获取验证码文件当中的标签数据
label = get_captcha_label()
print(image_batch, label)
with tf.Session() as sess:
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
# [b'NZPP' b'WKHK' b'WPSJ' ..., b'FVQJ' b'BQYA' b'BCHR']
label_str = sess.run(label)
print(label_str)
# 处理字符串标签到数字张量
label_batch = dealwithlabel(label_str)
print(label_batch)
# 将图片数据和内容写入到tfrecords文件当中
write_to_tfrecords(image_batch, label_batch)
coord.request_stop()
coord.join(threads)
1.4.2. 训练模型
import tensorflow as tf
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_string("captcha_dir", "./tfrecords_data/captcha.tfrecords", "验证码数据的路径")
tf.app.flags.DEFINE_integer("batch_size", 100, "每批次训练的样本数")
tf.app.flags.DEFINE_integer("letter_num", 26, "每个目标值取的字母的可能性")
tf.app.flags.DEFINE_integer("label_num", 4, "每个样本的目标值数量")
def weight_variables(shape):
"""定义一个初始化权重的函数
"""
w = tf.Variable(tf.random_normal(shape=shape, mean=0.0, stddev=1.0))
return w
def bias_variables(shape):
"""定义一个初始化偏置的函数
"""
b = tf.Variable(tf.constant(0.0, shape=shape))
return b
def read_and_decode():
"""读取验证码数据API, 返回image_batch, label_batch
"""
# 1. 构建文件队列
file_queue = tf.train.string_input_producer([FLAGS.captcha_dir])
# 2. 构建读取阅读器读取文件内容，默认一个样本
reader = tf.TFRecordReader()
key, value = reader.read(file_queue)
# 3. 解析example
features = tf.parse_single_example(value, features={
"image":tf.FixedLenFeature([], tf.string),
"label":tf.FixedLenFeature([], tf.string),
} )
# 4. 解码内容, 如果存取的格式是string类型的就需要解码，如果是int64，float32不需要解码
image = tf.decode_raw(features['image'], tf.uint8)
label = tf.decode_raw(features['label'], tf.uint8)
print(image, label)
# 改变形状
image_reshape = tf.reshape(image, [20, 80, 3])
label_reshape = tf.reshape(label, [FLAGS.label_num])
print(image_reshape, label_reshape)
# 进行批处理, 每批次
image_batch, label_batch = tf.train.batch([image_reshape, label_reshape], batch_size=FLAGS.batch_size, num_threads=1, capacity=FLAGS.batch_size)
print(image_batch, label_batch)
return image_batch, label_batch
def fc_model(image):
"""进行预测结果
参数：image: 100张图片的特征值[100, 20, 80, 3]
返回值：y_predict[100, 4*26]预测值
"""
with tf.variable_scope("model"):
# 将图片数据形状转换成二维的形状
image_reshape = tf.reshape(image, [-1, 20*80*3])
# 随机初始化权重
# matrix [100, 20*80*3] * [20*80*3, 4*26] +[4*26] = [100, 4*26]
weights = weight_variables([20*80*3, FLAGS.label_num * FLAGS.letter_num])
bias = bias_variables([FLAGS.label_num * FLAGS.letter_num])
# 进行全连接层计算
y_predict = tf.matmul(tf.cast(image_reshape,tf.float32), weights) + bias
return y_predict
def predict_to_onehot(label):
"""将读取文件中的目标值转换成one_hot编码
参数：[100, 4] -----> onehot编码
返回：onehot编码
"""
# 进行onehot编码转换，提供给交叉熵损失和准确率的计算
label_onehot = tf.one_hot(label, depth=FLAGS.letter_num, on_value=1.0, axis=2)
print(label_onehot)
return label_onehot
def captcharec():
"""验证码识别
"""
# 1. 读取验证码文件
image_batch, label_batch = read_and_decode()
# 2. 通过输入图片特征数据， 建立模型， 得出预测结果
# 一层全连接神经网络进行预测
# matrix [100, 20*80*3] * [20*80*3, 4*26] +[4*26] = [100, 4*26]
y_predict = fc_model(image_batch)
print(y_predict)
# 3. 先把目标值转换成one_hot编码
y_true = predict_to_onehot(label_batch)
# 4. softmax计算，交叉熵损失计算
with tf.variable_scope("soft_cross"):
# 求平均交叉熵损失，y_true[100, 4,26] -----> [100, 4*26]
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=tf.reshape(y_true,[-1, FLAGS.label_num * FLAGS.letter_num]), logits=y_predict))
# 5. 梯度下降，优化损失
with tf.variable_scope("optimizer"):
train_op = tf.train.GradientDescentOptimizer(0.01).minimize(loss)
# 求出样本每批次预测的准确率
with tf.variable_scope("acc"):
# 比较预测值和目标值位置是否一样（每个样本4个位置） y_predict[100, 4*26] -----> [100, 4, 26]
equal_list = tf.equal(tf.argmax(y_true, 2), tf.argmax(tf.reshape(y_predict,[FLAGS.batch_size, FLAGS.label_num , FLAGS.letter_num]), 2))
# equal_list None个样本[1,0,1,0,1,1........]
accuracy = tf.reduce_mean(tf.cast(equal_list, tf.float32))
init_op = tf.global_variables_initializer()
# 开启会话变量
with tf.Session() as sess:
sess.run(init_op)
# 定义线程协调器和开启线程
coord = tf.train.Coordinator()
# 开启线程去运行读取文件操作
threads = tf.train.start_queue_runners(sess, coord=coord)
# 训练识别程序
for i in range(5000):
sess.run(train_op)
print("第{}批次的准确率为：{}".format(i , accuracy.eval()))
# 回收线程
coord.request_stop()
coord.join(threads)
if __name__ == "__main__":
captcharec()
562
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
