capcha_recog

基于CNN的通用型字符验证码识别
综合准确率高于90%

Model graph

Install

pip install tensorflow

　　If you want to run with GPU,you may trypip install tensorflow-gpu

pip install opencv-python

Train

1. Prepare your dataset (png or jpg format) in './train' directory.
   　　We also offer a dataset if you do not have one, run unzip train.zip to decompress it.

2. Modify config.py to your liking, set basic parameters like IMAGE_HEIGHT, IMAGE_WIDTH, CAPTCHA_LENGTH and CHAR_SET_LEN.

3. run python train.py
   　　After training the steps you set (in config.py) the model will be saved automatically.

Test

1. Prepare your test dataset or run unzip test.zip to use our's.

2. run python test.py to load model and predict all capchas in './test'
   you can also modify test.py to test only one image.

Blog

capcha recognition based on CNN