神經網路驗證碼識別
阿新 • • 發佈：2019-01-01
根據（2）博主的規劃，筆者已經建立結構如下(文章結尾附上完整程式碼：主要來源（1）的up主)：
下面我們一個個說明：
（1）datasets
這是資料集資料夾，因為我們要實現驗證碼，所以首先要生成驗證碼圖片，在datasets目錄下有gen_image.py用於生成驗證碼圖片。這裡可以通過下載或者爬蟲獲取各種資料集，筆者採用下面方法
需要安裝captcha(這是一個生成驗證碼圖片的庫)
pip install captcha如果報錯no module named setuptools可以參考
然後執行產生圖片的指令碼（gen_image.bat）
python C:/Users/asus-/Desktop/captcha_demo/datasets/gen_image.py ^ --output_dir C:/Users/asus-/Desktop/captcha_demo/datasets/images/ ^ --Captcha_size 4 ^ --image_num 1000 ^ pause
--output_dir就是輸出圖片的儲存路徑
--Captcha_size就是識別碼圖片上面字元的個數
--image_num就是產生圖片的數量，但是有可能少於這個數，因為有可能產生重複的隨機數，會覆蓋前面的
關於gen_image.py為：
import tensorflow as tf from captcha.image import ImageCaptcha import random import sys FLAGS = tf.app.flags.FLAGS tf.app.flags.DEFINE_string('output_dir', '/ ', 'This is the saved directory of the picture') tf.app.flags.DEFINE_integer('Captcha_size', 3, 'This is the number of characters of captcha') tf.app.flags.DEFINE_integer('image_num', 1000, 'This is the number of pictures generated ,but less than image_num') #驗證碼內容 Captcha_content = ['0','1','2','3','4','5','6','7','8','9'] # 生成字元 def random_captcha_text(): captcha_text = [] for i in range(FLAGS.Captcha_size): ch = random.choice(Captcha_content) captcha_text.append(ch) return captcha_text # 生成字元對應的驗證碼 def gen_captcha_text_and_image(): image = ImageCaptcha() captcha_text = random_captcha_text() captcha_text = ''.join(captcha_text) captcha = image.generate(captcha_text) image.write(captcha_text, FLAGS.output_dir + captcha_text + '.jpg') def main(unuse_args): for i in range(FLAGS.image_num ): gen_captcha_text_and_image() sys.stdout.write('\r>> Creating image %d/%d' % (i+1, FLAGS.image_num)) sys.stdout.flush() sys.stdout.write('\n') sys.stdout.flush() print("Finish!!!!!!!!!!!") if __name__ == '__main__': tf.app.run()
執行後為：
接著轉化圖片為tfrecord格式，tfrecord資料檔案是一種將影象資料和標籤統一儲存的二進位制檔案，能更好的利用記憶體，在tensorflow中快速的複製，移動，讀取，儲存等.
同樣這裡寫了一個簡單的指令碼：
python C:/Users/asus-/Desktop/captcha_demo/datasets/gen_tfrecord.py ^ --dataset_dir C:/Users/asus-/Desktop/captcha_demo/datasets/images/ ^ --output_dir C:/Users/asus-/Desktop/captcha_demo/datasets/ ^ --test_num 10 ^ --random_seed 0 ^ pause
從上到下依次是資料集位置，tfrecord生成位置，測試集個數，隨機種子（用於打亂資料集）
import tensorflow as tf
import os
import random
import math
import sys
from PIL import Image
import numpy as np
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_string('dataset_dir', '/ ', 'This is the source directory of the picture')
tf.app.flags.DEFINE_string('output_dir', '/ ', 'This is the saved directory of the picture')
tf.app.flags.DEFINE_integer('test_num', 20, 'This is the number of test of captcha')
tf.app.flags.DEFINE_integer('random_seed', 0, 'This is the random_seed')
#判斷tfrecord檔案是否存在
def dataset_exists(dataset_dir):
for split_name in ['train', 'test']:
output_filename = os.path.join(dataset_dir,split_name + '.tfrecords')
if not tf.gfile.Exists(output_filename):
return False
return True
#獲取所有驗證碼圖片
def get_filenames_and_classes(dataset_dir):
photo_filenames = []
for filename in os.listdir(dataset_dir):
#獲取檔案路徑
path = os.path.join(dataset_dir, filename)
photo_filenames.append(path)
return photo_filenames
def int64_feature(values):
if not isinstance(values, (tuple, list)):
values = [values]
return tf.train.Feature(int64_list=tf.train.Int64List(value=values))
def bytes_feature(values):
return tf.train.Feature(bytes_list=tf.train.BytesList(value=[values]))
def image_to_tfexample(image_data, label0, label1, label2, label3):
#Abstract base class for protocol messages.
return tf.train.Example(features=tf.train.Features(feature={
'image': bytes_feature(image_data),
'label0': int64_feature(label0),
'label1': int64_feature(label1),
'label2': int64_feature(label2),
'label3': int64_feature(label3),
}))
#把資料轉為TFRecord格式
def convert_dataset(split_name, filenames, dataset_dir):
assert split_name in ['train', 'test']
with tf.Session() as sess:
#定義tfrecord檔案的路徑+名字
output_filename = os.path.join(FLAGS.output_dir,split_name + '.tfrecords')
with tf.python_io.TFRecordWriter(output_filename) as tfrecord_writer:
for i,filename in enumerate(filenames):
try:
sys.stdout.write('\r>> Converting image %d/%d' % (i+1, len(filenames)))
sys.stdout.flush()
#讀取圖片
image_data = Image.open(filename)
#根據模型的結構resize
image_data = image_data.resize((224, 224))
#灰度化
image_data = np.array(image_data.convert('L'))
#將圖片轉化為bytes
image_data = image_data.tobytes()
#獲取label
labels = filename.split('/')[-1][0:4]
num_labels = []
for j in range(4):
num_labels.append(int(labels[j]))
#生成protocol資料型別
example = image_to_tfexample(image_data, num_labels[0], num_labels[1], num_labels[2], num_labels[3])
tfrecord_writer.write(example.SerializeToString())
except IOError as e:
print('Could not read:',filename)
print('Error:',e)
sys.stdout.write('\n')
sys.stdout.flush()
def main(unuse_args):
if dataset_exists(FLAGS.output_dir):
print('tfcecord file has been existed!!')
else:
#獲得所有圖片
photo_filenames = get_filenames_and_classes(FLAGS.dataset_dir)
#把資料切分為訓練集和測試集,並打亂
random.seed(FLAGS.random_seed)
random.shuffle(photo_filenames)
training_filenames = photo_filenames[FLAGS.test_num:]
testing_filenames = photo_filenames[:FLAGS.test_num]
#資料轉換
convert_dataset('train', training_filenames,FLAGS.dataset_dir)
convert_dataset('test', testing_filenames, FLAGS.dataset_dir)
print('Finish!!!!!!!!!!!!!!!!!')
if __name__ == '__main__':
tf.app.run()
執行後：
(2)訓練
執行指令碼（train.bat）:
python C:/Users/asus-/Desktop/captcha_demo/train.py ^
--tfrecord_dir C:/Users/asus-/Desktop/captcha_demo/datasets/train.tfrecords ^
--model_dir C:/Users/asus-/Desktop/captcha_demo/model/Alexnet ^
--batch_size 15 ^
--train_num 10 ^
--print_loss_accuracy_interval 5 ^
--learning_rate 0.005 ^
pause這裡說一個數據集的讀入過程：
def read_and_decode(filename):
# 根據檔名生成一個佇列
filename_queue = tf.train.string_input_producer([filename])
reader = tf.TFRecordReader()
# 返回檔名和檔案
_, serialized_example = reader.read(filename_queue)
features = tf.parse_single_example(serialized_example,
features={
'image' : tf.FixedLenFeature([], tf.string),
'label0': tf.FixedLenFeature([], tf.int64),
'label1': tf.FixedLenFeature([], tf.int64),
'label2': tf.FixedLenFeature([], tf.int64),
'label3': tf.FixedLenFeature([], tf.int64),
})
# 獲取圖片資料
image = tf.decode_raw(features['image'], tf.uint8)
# tf.train.shuffle_batch必須確定shape
image = tf.reshape(image, [224, 224])
# 圖片預處理
image = tf.cast(image, tf.float32) / 255.0
image = tf.subtract(image, 0.5)
image = tf.multiply(image, 2.0)
# 獲取label
label0 = tf.cast(features['label0'], tf.int32)
label1 = tf.cast(features['label1'], tf.int32)
label2 = tf.cast(features['label2'], tf.int32)
label3 = tf.cast(features['label3'], tf.int32)
return image, label0, label1, label2, label3
正如上面所說我們這裡的reader對應的是即tfrecord格式的reader
reader = tf.TFRecordReader()(1)資料中up主運行了大概6000次，因為筆者電腦配置較低又是cpu，但是為了後續視覺化設計的神經網路，這裡我就暫且先執行10次，使之產生model
接下來我們通過tensorboard寫了個小demo來直觀的看一下設計的多工alexnet網路
import tensorflow as tf
with tf.Session() as sess:
my_saver = tf.train.import_meta_graph('C:/Users/asus-/Desktop/captcha_demo/model/Alexnet.meta')
my_saver.restore(sess,tf.train.latest_checkpoint('C:/Users/asus-/Desktop/captcha_demo/model/'))
graph = tf.get_default_graph()
writer_test=tf.summary.FileWriter('C:/Users/asus-/Desktop/logs/',sess.graph)
可以看到前5層convolutional，後邊3層full-connected，最後一層的full-connected採取的是全連線層，對應這個採用多工的類子中最後一層對應四個連線層
關於更多的alexent網路，可以查文件
最後貼一下train.py:
import os
import tensorflow as tf
from nets import nets_factory
import numpy as np
import image_reader as ir
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_string('tfrecord_dir', '/ ', 'This is the tfrecord directory of the picture')
tf.app.flags.DEFINE_string('model_dir', '/ ', 'This is the saved model directory of the net')
tf.app.flags.DEFINE_integer('batch_size', 10, 'This is the batch size')
tf.app.flags.DEFINE_integer('train_num', 1000, 'This is the number of train')
tf.app.flags.DEFINE_integer('print_loss_accuracy_interval', 10, 'This is the interval of printing')
tf.app.flags.DEFINE_float('learning_rate', 0.001, 'This is the rate of learning')
# 不同字元數量
CHAR_SET_LEN = 10
# placeholder
x = tf.placeholder(tf.float32, [None, 224, 224])
y0 = tf.placeholder(tf.float32, [None])
y1 = tf.placeholder(tf.float32, [None])
y2 = tf.placeholder(tf.float32, [None])
y3 = tf.placeholder(tf.float32, [None])
image, label0, label1, label2, label3 = ir.read_and_decode(FLAGS.tfrecord_dir)
#使用shuffle_batch可以隨機打亂
image_batch, label_batch0, label_batch1, label_batch2, label_batch3 = tf.train.shuffle_batch(
[image, label0, label1, label2, label3], batch_size =FLAGS.batch_size,
capacity = 50000, min_after_dequeue=10000, num_threads=1)
#定義網路結構
train_network_fn = nets_factory.get_network_fn(
'alexnet_v2',
num_classes=CHAR_SET_LEN,
weight_decay=0.0005,
is_training=True)
# inputs: a tensor of size [batch_size, height, width, channels]
X = tf.reshape(x, [FLAGS.batch_size, 224, 224, 1])
# 資料輸入網路得到輸出值
logits0,logits1,logits2,logits3,end_points = train_network_fn(X)
# 把標籤轉成one_hot的形式
one_hot_labels0 = tf.one_hot(indices=tf.cast(y0, tf.int32), depth=CHAR_SET_LEN)
one_hot_labels1 = tf.one_hot(indices=tf.cast(y1, tf.int32), depth=CHAR_SET_LEN)
one_hot_labels2 = tf.one_hot(indices=tf.cast(y2, tf.int32), depth=CHAR_SET_LEN)
one_hot_labels3 = tf.one_hot(indices=tf.cast(y3, tf.int32), depth=CHAR_SET_LEN)
loss0 = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits0,labels=one_hot_labels0))
loss1 = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits1,labels=one_hot_labels1))
loss2 = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits2,labels=one_hot_labels2))
loss3 = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits3,labels=one_hot_labels3))
total_loss = (loss0+loss1+loss2+loss3)/4.0
optimizer = tf.train.AdamOptimizer(learning_rate=FLAGS.learning_rate).minimize(total_loss)
# 計算準確率
correct_prediction0 = tf.equal(tf.argmax(one_hot_labels0,1),tf.argmax(logits0,1))
accuracy0 = tf.reduce_mean(tf.cast(correct_prediction0,tf.float32))
correct_prediction1 = tf.equal(tf.argmax(one_hot_labels1,1),tf.argmax(logits1,1))
accuracy1 = tf.reduce_mean(tf.cast(correct_prediction1,tf.float32))
correct_prediction2 = tf.equal(tf.argmax(one_hot_labels2,1),tf.argmax(logits2,1))
accuracy2 = tf.reduce_mean(tf.cast(correct_prediction2,tf.float32))
correct_prediction3 = tf.equal(tf.argmax(one_hot_labels3,1),tf.argmax(logits3,1))
accuracy3 = tf.reduce_mean(tf.cast(correct_prediction3,tf.float32))
# 用於儲存模型
saver = tf.train.Saver()
def main(unuse_args):
with tf.Session() as sess:
# 初始化
sess.run(tf.global_variables_initializer())
# 建立一個協調器，管理執行緒
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
for i in range(FLAGS.train_num):
b_image, b_label0, b_label1 ,b_label2 ,b_label3 = sess.run([image_batch, label_batch0, label_batch1, label_batch2, label_batch3])
sess.run(optimizer, feed_dict={x: b_image, y0:b_label0, y1: b_label1, y2: b_label2, y3: b_label3})
if i % FLAGS.print_loss_accuracy_interval == 0:
acc0,acc1,acc2,acc3,TotalLoss = sess.run([accuracy0,accuracy1,accuracy2,accuracy3,total_loss],feed_dict={x: b_image,
y0: b_label0,
y1: b_label1,
y2: b_label2,
y3: b_label3})
print ("times:%d Loss:%.3f Accuracy:%.2f,%.2f,%.2f,%.2f" % (i,TotalLoss,acc0,acc1,acc2,acc3))
saver.save(sess, FLAGS.model_dir)
# 通知其他執行緒關閉
coord.request_stop()
coord.join(threads)
if __name__ == '__main__':
tf.app.run()
(3)測試
同樣執行一個指令碼：
python C:/Users/asus-/Desktop/captcha_demo/evaluate.py ^
--tfrecord_dir C:\Users\asus-\Desktop\captcha_demo\datasets\test.tfrecords ^
--test_size 10 ^
pause關於evaluate.py 則為：
import tensorflow as tf
import image_reader as ir
from nets import nets_factory
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_string('tfrecord_dir', '/ ', 'This is the tfrecord directory of the picture')
tf.app.flags.DEFINE_integer('test_size', 10, 'This is the batch size')
# 不同字元數量
CHAR_SET_LEN = 10
BATCH_SIZE=1
# placeholder
x = tf.placeholder(tf.float32, [None, 224, 224])
y0 = tf.placeholder(tf.float32, [None])
y1 = tf.placeholder(tf.float32, [None])
y2 = tf.placeholder(tf.float32, [None])
y3 = tf.placeholder(tf.float32, [None])
image, label0, label1, label2, label3 = ir.read_and_decode(FLAGS.tfrecord_dir)
#使用shuffle_batch可以隨機打亂
image_batch, label_batch0, label_batch1, label_batch2, label_batch3 = tf.train.shuffle_batch(
[image, label0, label1, label2, label3], batch_size =BATCH_SIZE,
capacity = 50000, min_after_dequeue=10000, num_threads=1)
#定義網路結構
train_network_fn = nets_factory.get_network_fn(
'alexnet_v2',
num_classes=CHAR_SET_LEN,
weight_decay=0.0005,
is_training=False)
# inputs: a tensor of size [batch_size, height, width, channels]
X = tf.reshape(x, [BATCH_SIZE, 224, 224, 1])
# 資料輸入網路得到輸出值
logits0,logits1,logits2,logits3,end_points = train_network_fn(X)
# 預測值
predict0 = tf.reshape(logits0, [-1, CHAR_SET_LEN])
predict0 = tf.argmax(predict0, 1)
predict1 = tf.reshape(logits1, [-1, CHAR_SET_LEN])
predict1 = tf.argmax(predict1, 1)
predict2 = tf.reshape(logits2, [-1, CHAR_SET_LEN])
predict2 = tf.argmax(predict2, 1)
predict3 = tf.reshape(logits3, [-1, CHAR_SET_LEN])
predict3 = tf.argmax(predict3, 1)
# 把標籤轉成one_hot的形式
one_hot_labels0 = tf.one_hot(indices=tf.cast(y0, tf.int32), depth=CHAR_SET_LEN)
one_hot_labels1 = tf.one_hot(indices=tf.cast(y1, tf.int32), depth=CHAR_SET_LEN)
one_hot_labels2 = tf.one_hot(indices=tf.cast(y2, tf.int32), depth=CHAR_SET_LEN)
one_hot_labels3 = tf.one_hot(indices=tf.cast(y3, tf.int32), depth=CHAR_SET_LEN)
loss0 = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits0,labels=one_hot_labels0))
loss1 = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits1,labels=one_hot_labels1))
loss2 = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits2,labels=one_hot_labels2))
loss3 = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits3,labels=one_hot_labels3))
total_loss = (loss0+loss1+loss2+loss3)/4.0
train_step = tf.train.AdamOptimizer(learning_rate=0.0001).minimize(total_loss)
# 計算準確率
correct_prediction0 = tf.equal(tf.argmax(one_hot_labels0,1),tf.argmax(logits0,1))
accuracy0 = tf.reduce_mean(tf.cast(correct_prediction0,tf.float32))
correct_prediction1 = tf.equal(tf.argmax(one_hot_labels1,1),tf.argmax(logits1,1))
accuracy1 = tf.reduce_mean(tf.cast(correct_prediction1,tf.float32))
correct_prediction2 = tf.equal(tf.argmax(one_hot_labels2,1),tf.argmax(logits2,1))
accuracy2 = tf.reduce_mean(tf.cast(correct_prediction2,tf.float32))
correct_prediction3 = tf.equal(tf.argmax(one_hot_labels3,1),tf.argmax(logits3,1))
accuracy3 = tf.reduce_mean(tf.cast(correct_prediction3,tf.float32))
def main(unuse_args):
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
#載入模型
my_saver = tf.train.import_meta_graph('C:/Users/asus-/Desktop/captcha_demo/model/Alexnet.meta')
my_saver.restore(sess,tf.train.latest_checkpoint('C:/Users/asus-/Desktop/captcha_demo/model/'))
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
for i in range(FLAGS.test_size):
b_image, b_label0, b_label1 ,b_label2 ,b_label3 = sess.run([image_batch, label_batch0, label_batch1, label_batch2, label_batch3])
print('%d times label:%d,%d,%d,%d' % ((i+1) ,b_label0, b_label1 ,b_label2 ,b_label3))
sess.run(train_step, feed_dict={x: b_image, y0:b_label0, y1: b_label1, y2: b_label2, y3: b_label3})
label0,label1,label2,label3 = sess.run([predict0,predict1,predict2,predict3], feed_dict={x: b_image})
print('predict:',label0,label1,label2,label3)
coord.request_stop()
coord.join(threads)
if __name__ == '__main__':
tf.app.run()
注意筆者在測試的時候
--tfrecord_dir C:\Users\asus-\Desktop\captcha_demo\datasets\test.tfrecords ^資料夾路徑如果由\改為/即為：
--tfrecord_dir C:/Users/asus-/Desktop/captcha_demo/datasets/test.tfrecords ^那麼結果會報錯，意思就是說沒有讀取到資料，呵呵目前還不明覺厲！！
最後附上全部程式碼：