czczup / captcha-recognition

基于深度学习和TensorFlow的英文和汉字验证码识别

深度学习识别各类验证码

背景介绍

• 来源：第九届**大学生服务外包创新创业大赛赛题A16-验证码识别
• 整体背景：验证码识别是计算机与大数据领域一个非常重要的应用领域，其中包含了图像处理、机器视觉、模式识别以及人工智能等诸多前沿技术，这些同样也是大数据未来发展所必须的核心技术。本赛题以互联网中常见的验证码图片为样本，旨在通过此次竞赛，检验参赛者的图像识别能力，提高对计算机技术、算法模型的认识和应用能力，激发在图像处理、机器学习等领域的创新能力。
• 公司背景：浪潮卓数大数据产业发展有限公司作为浪潮集团旗下的大数据板块，致力于成为数据资源提供商、数据资产运营商和数据交易服务商，以大数据时代的“数商”为发展目标，促进政府、企业和个人实现从互联网化向社会化，乃至数据社会化的转变。

项目说明

问题说明

提供 5 类验证码图片，难度依次递增，每一类验证码提供 1 万个训练样本，最后通过另外 5000 个测试样本的识别率评分。

第 1 类

数字四则运算，有噪点干扰，输出计算结果。

第 2 类

英文字母+数字验证码，包含 5 个字符，有噪点干扰，文字无旋转形变，验证方法为要求用户输出验证码中的字符，大小写不限（ 为验证方便可统一转为大写 ）。

第 3 类

英文字母+数字验证码，包含 4 个字符，有噪点干扰，文字有旋转形变，验证方法为要求用户输出验证码中的字符。

第 4 类

中文验证码，包含 4 个中文汉字，有噪点干扰，验证方法为要求用户选出 4 个汉字中被旋转 90 度的那一个 （四个汉字从左到右序号为 0,1,2,3，输出被旋转的汉字序号即可） 。

第 5 类

中文验证码，包含 4 个中文汉字和 9 个中文单字，有噪点干扰，文字有旋转形变，验证方法为要求用户从 9 个单字中从左到右按顺序选出验证码中的汉字，输出汉字编号。

快速开始

• 步骤一：环境准备

  Python 3.6、TensorFlow、Pillow、OpenCV

  pip install tensorflow-gpu==1.8
  pip install opencv-python
  pip install pillow
  

• 步骤二：准备数据集

  下载地址：百度网盘

  提取码：e6zy

  放置位置：

  • 训练集：level_/image/train
  • 测试集：level_/image/test

• 步骤三：选择验证码类型

  • 第一类：数字四则运算，有噪点干扰，输出计算结果

  • 第二类：数字、英文，有噪点干扰，文字无旋转形变，输出数字、英文

  • 第三类：数字、英文，有噪点干扰，文字有旋转形变，输出数字、英文

  • 第四类：汉字，有噪点干扰，文字无旋转形变，输出汉字

  • 第五类：汉字，有噪点干扰，文字有旋转形变，输出汉字

最后

当时做这个项目的时候是第一次接触深度学习，看着吴恩达和炼数成金的视频课边学边做，真是特别有意思的一段时光。现在回头来看，当时设计的每类验证码识别模型都不同，准确率主要靠调参，确实挺ugly的。还记得答辩的时候评委问我，能不能做一个万能验证码识别模型，可惜当时学识浅陋，我答的“不行”，哈哈。

通过这次竞赛，成功让我入门了深度学习，特别感谢浪潮集团出的题和提供的奖金。当年这题获奖的名额超多的，浪潮真的财大气粗，若要参加服务外包竞赛，浪潮的题目超级推荐。验证码识别是特别好的深度学习入门案例，希望本仓库能给刚入门的你提供一些帮助。

captcha-recognition's Issues

训练多少步

请问楼主训练多少步

生成pb模型

level2的例子中，我想用ckpt文件生成pb模型，但是需要指定输出名字。我是用ArgMax显示不存在。 使用FC2/wx_plus_b结果好像对不上。

汉字识别

像这样验证码,需要在4组选项,每组4个汉字中,选择一组匹配的汉字.
应该经过哪些步骤的处理?

运行test文件正确率是0

模型还没训练完，在五万step左右，运行test后正确率是0，为什么啊，正常吗

请问一下运行denoise时出现这种错误怎么解决

[ WARN:[email protected]] global loadsave.cpp:248 cv::findDecoder imread_('./image/train/0000.jpg'): can't open/read file: check file path/integrity
Process Process-1:
Traceback (most recent call last):
File "C:\Program Files\Python312\Lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Program Files\Python312\Lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self.kwargs)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 134, in remove_batch
remove_noise(i)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 116, in remove_noise
image = RGB_clean(image)
^^^^^^^^^^^^^^^^
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 31, in RGB_clean
B, G, R = [int(i) for i in get_threshold(image)]
^^^^^^^^^^^^^^^^^^^^
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 17, in get_threshold
pixels = np.array([image[y-1, x-1], image[y-1, x], image[y-1, x+1],
~~~~~^^^^^^^^^^
TypeError: 'NoneType' object is not subscriptable
[ WARN:[email protected]] global loadsave.cpp:248 cv::findDecoder imread('./image/train/2000.jpg'): can't open/read file: check file path/integrity
Process Process-2:
Traceback (most recent call last):
File "C:\Program Files\Python312\Lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Program Files\Python312\Lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self.kwargs)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 134, in remove_batch
remove_noise(i)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 116, in remove_noise
image = RGB_clean(image)
^^^^^^^^^^^^^^^^
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 31, in RGB_clean
B, G, R = [int(i) for i in get_threshold(image)]
^^^^^^^^^^^^^^^^^^^^
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 17, in get_threshold
pixels = np.array([image[y-1, x-1], image[y-1, x], image[y-1, x+1],
~~~~~^^^^^^^^^^
TypeError: 'NoneType' object is not subscriptable
[ WARN:[email protected]] global loadsave.cpp:248 cv::findDecoder imread('./image/train/4000.jpg'): can't open/read file: check file path/integrity
Process Process-3:
Traceback (most recent call last):
File "C:\Program Files\Python312\Lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Program Files\Python312\Lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self.kwargs)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 134, in remove_batch
remove_noise(i)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 116, in remove_noise
image = RGB_clean(image)
^^^^^^^^^^^^^^^^
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 31, in RGB_clean
B, G, R = [int(i) for i in get_threshold(image)]
^^^^^^^^^^^^^^^^^^^^
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 17, in get_threshold
pixels = np.array([image[y-1, x-1], image[y-1, x], image[y-1, x+1],
~~~~~^^^^^^^^^^
TypeError: 'NoneType' object is not subscriptable
[ WARN:[email protected]] global loadsave.cpp:248 cv::findDecoder imread('./image/train/6000.jpg'): can't open/read file: check file path/integrity
Process Process-4:
Traceback (most recent call last):
File "C:\Program Files\Python312\Lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Program Files\Python312\Lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self.kwargs)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 134, in remove_batch
remove_noise(i)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 116, in remove_noise
image = RGB_clean(image)
^^^^^^^^^^^^^^^^
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 31, in RGB_clean
B, G, R = [int(i) for i in get_threshold(image)]
^^^^^^^^^^^^^^^^^^^^
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 17, in get_threshold
pixels = np.array([image[y-1, x-1], image[y-1, x], image[y-1, x+1],
~~~~~^^^^^^^^^^
TypeError: 'NoneType' object is not subscriptable
[ WARN:[email protected]] global loadsave.cpp:248 cv::findDecoder imread('./image/train/8000.jpg'): can't open/read file: check file path/integrity
Process Process-5:
Traceback (most recent call last):
File "C:\Program Files\Python312\Lib\multiprocessing\process.py", line 314, in _bootstrap
self.run()
File "C:\Program Files\Python312\Lib\multiprocessing\process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 134, in remove_batch
remove_noise(i)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 116, in remove_noise
image = RGB_clean(image)
^^^^^^^^^^^^^^^^
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 31, in RGB_clean
B, G, R = [int(i) for i in get_threshold(image)]
^^^^^^^^^^^^^^^^^^^^
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\denoise_opencv.py", line 17, in get_threshold
pixels = np.array([image[y-1, x-1], image[y-1, x], image[y-1, x+1],
~~~~~^^^^^^^^^^
TypeError: 'NoneType' object is not subscriptable

test报错

The passed save_path is not a valid checkpoint: ./model/model_level3.ckpt-800
运行test.py报错

请问下运行generate_tfrecord时出现这个错误怎么解决

C:\Users\86182\Desktop\Captcha-Recognition-master\venv\Scripts\python.exe C:\Users\86182\Desktop\Captcha-Recognition-master\level1\generate_tfrecord.py
2024-04-18 16:40:44.886591: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable TF_ENABLE_ONEDNN_OPTS=0.
2024-04-18 16:40:46.200939: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable TF_ENABLE_ONEDNN_OPTS=0.
Traceback (most recent call last):
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\generate_tfrecord.py", line 89, in
_convert_dataset('train', training_filenames, class_names_to_ids, conf.TFRECORD_PATH)
File "C:\Users\86182\Desktop\Captcha-Recognition-master\level1\generate_tfrecord.py", line 62, in _convert_dataset
tfrecord_writer = tf.python_io.TFRecordWriter(output_filename)
^^^^^^^^^^^^
AttributeError: module 'tensorflow' has no attribute 'python_io'

模型训练问题

请问下我的第二类第三类模型为啥训练了两三天还没训练完

请问识别出来的验证码结果是放在mapping里吗