test

#coding:utf-8
from gen_captcha import gen_captcha_text_and_image
from gen_captcha import number
from gen_captcha import alphabet
from gen_captcha import ALPHABET
import tensorflow.contrib.slim as slim
import numpy as np
import tensorflow as tf
from scipy.misc import imread

text, image = gen_captcha_text_and_image() #先生成验证码和文字测试模块是否完全
print("验证码图像channel:", image.shape)  # (60, 160, 3)
# 图像大小
IMAGE_HEIGHT = 60
IMAGE_WIDTH = 160
MAX_CAPTCHA = len(text)
print("验证码文本最长字符数", MAX_CAPTCHA)   # 验证码最长4字符; 我全部固定为4,可以不固定. 如果验证码长度小于4，用'_'补齐

# 把彩色图像转为灰度图像（色彩对识别验证码没有什么用）
def convert2gray(img):
   if len(img.shape) > 2:
      gray = np.mean(img, -1)
      # 上面的转法较快，正规转法如下
      # r, g, b = img[:,:,0], img[:,:,1], img[:,:,2]
      # gray = 0.2989 * r + 0.5870 * g + 0.1140 * b
      return gray
   else:
      return img

# 文本转向量
char_set = number + alphabet + ALPHABET + ['_']  # 如果验证码长度小于4, '_'用来补齐
CHAR_SET_LEN = len(char_set)
def text2vec(text):
   text_len = len(text)
   if text_len > MAX_CAPTCHA:
      raise ValueError('验证码最长4个字符')

   vector = np.zeros(MAX_CAPTCHA*CHAR_SET_LEN)
   def char2pos(c):
      if c =='_':
         k = 62
         return k
      k = ord(c)-48
      if k > 9:
         k = ord(c) - 55
         if k > 35:
            k = ord(c) - 61
            if k > 61:
               raise ValueError('No Map')
      return k
   for i, c in enumerate(text):
      idx = i * CHAR_SET_LEN + char2pos(c)
      vector[idx] = 1
   return vector
# 向量转回文本
def vec2text(vec):
   char_pos = vec.nonzero()[0]
   text=[]
   for i, c in enumerate(char_pos):
      char_at_pos = i #c/63
      char_idx = c % CHAR_SET_LEN
      if char_idx < 10:
         char_code = char_idx + ord('0')
      elif char_idx < 36:
         char_code = char_idx - 10 + ord('A')
      elif char_idx < 62:
         char_code = char_idx-  36 + ord('a')
      elif char_idx == 62:
         char_code = ord('_')
      else:
         raise ValueError('error')
      text.append(chr(char_code))
   return "".join(text)

"""
#向量（大小MAX_CAPTCHA*CHAR_SET_LEN）用0,1编码 每63个编码一个字符，这样顺利有，字符也有
vec = text2vec("F5Sd")
text = vec2text(vec)
print(text)  # F5Sd
vec = text2vec("SFd5")
text = vec2text(vec)
print(text)  # SFd5
"""
# 生成一个训练batch
def get_next_batch(batch_size=128):
   batch_x = np.zeros([batch_size, IMAGE_HEIGHT*IMAGE_WIDTH])
   batch_y = np.zeros([batch_size, MAX_CAPTCHA*CHAR_SET_LEN])

   # 有时生成图像大小不是(60, 160, 3)
   def wrap_gen_captcha_text_and_image():
      ''' 获取一张图，判断其是否符合（60，160，3）的规格'''
      while True:
         text, image = gen_captcha_text_and_image()
         if image.shape == (60, 160, 3):#此部分应该与开头部分图片宽高吻合
            return text, image

   for i in range(batch_size):
      text, image = wrap_gen_captcha_text_and_image()
      image = convert2gray(image)

      # 将图片数组一维化 同时将文本也对应在两个二维组的同一行
      batch_x[i,:] = image.flatten() / 255 # (image.flatten()-128)/128  mean为0
      batch_y[i,:] = text2vec(text)
   # 返回该训练批次
   return batch_x, batch_y

####################################################################
# 申请占位符 按照图片
X = tf.placeholder(tf.float32, [None, IMAGE_HEIGHT*IMAGE_WIDTH])
Y = tf.placeholder(tf.float32, [None, MAX_CAPTCHA*CHAR_SET_LEN])
keep_prob = tf.placeholder(tf.float32) # dropout

# 定义CNN
def crack_captcha_cnn():
   # 将占位符 转换为 按照图片给的新样式
   #slim.conv2d的stride默认为1，slim.max_pool2d的stride默认与f保持一致, padding默认都为SAME
   x = tf.reshape(X, shape=[-1, IMAGE_HEIGHT, IMAGE_WIDTH, 1])
   with slim.arg_scope([slim.conv2d, slim.fully_connected],  activation_fn=tf.nn.relu,
                                weights_initializer=tf.truncated_normal_initializer(stddev=0.1),
                                weights_regularizer=slim.l2_regularizer(0.0005)):
      net = slim.conv2d(x, 32, 3, scope="conv1")#32为通道，3为f即[3, 3]，stride默认为1，padding默认为SAME
      net= slim.max_pool2d(net, 2, scope="pool1")#2即f为[2,2],stride默认与f保持一致
      net = slim.conv2d(net, 64, 1, scope="conv2")
      net = slim.max_pool2d(net, 2, scope="pool2")
      net = slim.conv2d(net, 64, 1, scope="conv3")
      net = slim.max_pool2d(net, 2, scope="pool3")
      net = slim.flatten(net, scope='flat_1')
      net = slim.dropout(net, keep_prob=keep_prob, scope='drop1')
      net = slim.fully_connected(net, 1024, scope="f1")
      net = slim.fully_connected(net, 512, scope="f2")
      z = slim.fully_connected(net, MAX_CAPTCHA*CHAR_SET_LEN, activation_fn=None, scope="f3")
      p = tf.nn.sigmoid(z, name="prediction")
   return z, p

# 训练
def train_crack_captcha_cnn():
   z, p = crack_captcha_cnn()
   #sigmoid多任务模式，一个标签中有多个目标
   loss = tf.reduce_mean(slim.losses.sigmoid_cross_entropy(z, Y))
   #loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=z, labels=Y))
   optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)

   predict = tf.reshape(p, [-1, MAX_CAPTCHA, CHAR_SET_LEN])
   max_idx_p = tf.argmax(predict, 2)
   max_idx_l = tf.argmax(tf.reshape(Y, [-1, MAX_CAPTCHA, CHAR_SET_LEN]), 2)
   correct_pred = tf.equal(max_idx_p, max_idx_l)
   accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
   saver = tf.train.Saver()
   saver_r = tf.train.import_meta_graph('E:/data/tensorflow-master/1.Cnn_Captcha/model-8800.meta')
   # 指定监测结果输出目录
   with tf.Session() as sess:
      #sess.run(tf.global_variables_initializer())
      summary_writer =  tf.summary.FileWriter('E:/data/tensorflow-master/1.Cnn_Captcha/result/', sess.graph)
      saver_r.restore(sess, tf.train.latest_checkpoint('E:/data/tensorflow-master/1.Cnn_Captcha/'))
      step = 0
      while True:
         batch_x, batch_y = get_next_batch(64)
         _, loss_ = sess.run([optimizer, loss], feed_dict={X: batch_x, Y: batch_y, keep_prob: 0.75})
         # 把Tensor添加到观测中
         tf.summary.scalar('loss', loss)
         print(step, loss_)

         # 每100 step计算一次准确率
         if step % 100 == 0:
            batch_x_test, batch_y_test = get_next_batch(100)
            #acc = sess.run(accuracy, feed_dict={X: batch_x_test, Y: batch_y_test, keep_prob: 1.})
            tf.summary.scalar('accuracy', accuracy)
            #sum_ops = tf.merge_all_summaries()
            sum_ops = tf.summary.merge_all()
            metall = sess.run(sum_ops, feed_dict={X: batch_x_test, Y: batch_y_test, keep_prob: 1.})
            summary_writer.add_summary(metall, global_step=step)  # 写入文件
            #print(step, acc)
            # 如果准确率大于50%,保存模型,完成训练
            if step % 1000 == 0 and step != 0:
               saver.save(sess, "E:/data/tensorflow-master/1.Cnn_Captcha/model/", global_step=step)
               break
         step += 1
def train():
   train_log_dir = "E:/data/tensorflow-master/1.Cnn_Captcha/result/"
   if not tf.gfile.Exists(train_log_dir):
      tf.gfile.MakeDirs(train_log_dir)
   with tf.Graph().as_default():
      batch_x, batch_y = get_next_batch(64)
      x = tf.reshape(batch_x, shape=[-1, IMAGE_HEIGHT, IMAGE_WIDTH, 1])
      y = tf.cast(batch_y, tf.float32)
      z, p = crack_captcha_cnn(tf.cast(x, tf.float32), 0.75)
      loss = tf.reduce_mean(slim.losses.sigmoid_cross_entropy(z, y))
      optimizer = tf.train.AdamOptimizer(learning_rate=0.001)
      tf.summary.scalar('losses/total_loss', loss)
      predict = tf.reshape(p, [-1, MAX_CAPTCHA, CHAR_SET_LEN])
      max_idx_p = tf.argmax(predict, 2)
      max_idx_l = tf.argmax(tf.reshape(y, [-1, MAX_CAPTCHA, CHAR_SET_LEN]), 2)
      correct_pred = tf.equal(max_idx_p, max_idx_l)
      accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
      tf.summary.scalar('losses/accuracy', accuracy)
      train_tensor = slim.learning.create_train_op(loss, optimizer,global_step=10)
      slim.learning.train(train_tensor, train_log_dir)

#train()
train_crack_captcha_cnn()
saver = tf.train.import_meta_graph('E:/data/tensorflow-master/1.Cnn_Captcha/model-8800.meta')
with tf.Session() as sess:
   saver.restore(sess, tf.train.latest_checkpoint('E:/data/tensorflow-master/1.Cnn_Captcha/'))
   graph = tf.get_default_graph()
   image = imread("E:/data/tensorflow-master/1.Cnn_Captcha/create_image/image/0h6Y.jpg")
   batch_x = np.zeros([1, IMAGE_HEIGHT * IMAGE_WIDTH])
   batch_y = np.zeros([1, MAX_CAPTCHA * CHAR_SET_LEN])
   batch_x[0, :] = image.flatten() / 255  # (image.flatten()-128)/128  mean为0
   batch_y[0, :] = text2vec("0h6Y")
   p = graph.get_tensor_by_name("prediction:0")
   predict = tf.reshape(p, [-1, MAX_CAPTCHA, CHAR_SET_LEN])
   X = graph.get_tensor_by_name("Placeholder_3:0")
   Y = graph.get_tensor_by_name("Placeholder_1:0")
   keep_prob = graph.get_tensor_by_name("Placeholder_2_1:0")
   max_idx_p = tf.argmax(predict, 2)
   p = sess.run(max_idx_p, feed_dict={X: batch_x, Y: batch_y, keep_prob: 1})
   print(p)
   print(np.argmax(np.array(batch_y).reshape([MAX_CAPTCHA, CHAR_SET_LEN]), axis=1))
   # Do some work with the model
   ...