symant233 / zfsoft-captcha2 Goto Github PK

Inspired by AHU-JiaoWu, Modified from scikit-learn into tensorflow. Add more comments for tutorial use.

pip install -r requirements.txt -U
python trainer.py # train your own model
python predictor.py  # 预测'./predict/'下的图片
# new
python app.py # flask sever to predict incoming data

Target image examples:

# train
Epoch 36/36
1200/1200 [==============================] - 0s 47us/sample - loss: 0.0321 - acc: 0.9967
# test
196/196 [==============================] - 0s 476us/sample - loss: 0.1501 - acc: 0.9643

• Add some test sets. (Done)
• More comments (for tutorial).
• run a flask sever to predict incoming pic data (Done)
• image_splite improve.
• unitest & travis-ci (Abort)
• command line argv support

1. retrieve some picture
2. tag the right code of each captcha pics
3. write splite codes
4. splite code (let output layer 36 [0-9, a-z])
5. using tensorflow train the tagged data[*]
6. save the model for future uses
7. predict given pic with that model

Step 5 needs some basic tensorflow skills, you can learn these in the link below:

• Google Machine Learnig Crash Course /
• Tensorflow official tutorial / 均支持简体中文

This is optional, you can use predictor.py alone by adding pictures into the predict folder. Or transmit picture data stream to the flask server.

python app.py # run this first

You can go to http://127.0.0.1:5000/upload see the demo.

Or using some python code below with the api (this is POST only)

import requests

with open('Check.gif', 'rb') as f:
    stream = f.read()

r = requests.post('http://127.0.0.1:5000/api', data=stream)
if r.status_code == 200:
    print(r.text)

Due to the licence lacunae of original project, I added MIT LICENCE to my codes. The split picture codes' copyright are belong to @AHU-HUI.

bash@zfsoft-captcha2$ tree
├── app.py                # flask sever (predictor)
├── model
│   └── Model_tf.net      # tensorflow model created by trainer
├── predict/              # predict folder
├── process   
│   └── split_code_imgs.py
├── requirements.txt      
├── trainer.py            # train and test
├── predictor.py          # predictor
├── data
│   ├── train/            # train images
│   └── test_sets/        # test images
├── README.MD
└── LICENSE