CAPTCHA-OCR

图像验证码自动识别系统

开发准备

• ubuntu 14.04

搭建环境

安装OpenCV

sudo apt-get install libcv-dev

安装PaddlePaddle

• http://www.paddlepaddle.org/doc_cn/build_and_install/install/ubuntu_install.html

安装CUDA

wget https://developer.nvidia.com/compute/cuda/8.0/Prod2/local_installers/cuda-repo-ubuntu1404-8-0-local-ga2_8.0.61-1_amd64-deb
sudo dpkg -i cuda-repo-ubuntu1404-8-0-local-ga2_8.0.61-1_amd64.deb
sudo apt-get update
sudo apt-get install cuda

安装cuDNN

https://developer.nvidia.com/cudnn

安装MxNet

下载mxnet

git clone --recursive https://github.com/dmlc/mxnet

将mxnet/make/config.mk复制到mxnet/，修改以下几行：

USE_CUDA = 1
USE_CUDA_PATH = /usr/local/cuda

在mxnet/目录下编译

make -j4

安装python支持

cd python
python setup.py install

安装tesseract

sudo apt-get install tesseract-ocr

安装pytesseract

sudo pip install pytesseract

安装pyqt4

sudo apt-get install libxext6 libxext-dev libqt4-dev libqt4-gui libqt4-sql qt4-dev-tools qt4-doc qt4-designer qt4-qtconfig "python-qt4-*" python-qt4

安装selenium

pip install selenium

下载项目

git clone https://github.com/wangmengzhi/CAPTCHA-OCR.git

cd CAPTCHA-OCR

把样本转为Paddle格式

python preprocess.py

训练

sh train.sh

配置tesseract

• 将captcha和captcha_digits拷贝到/usr/share/tesseract-ocr/tessdata/configs/

运行

python CAPTCHA-OCR.py