A cracker for sliding captcha of GEETEST, NetEase, Tencent and iQiyi based on OpenCV

macOS Mojave Version 10.14.1

• Python 3.6.1
• Selenium 3.11.0
• cv2 4.0.0
• PIL 1.1.7
• Chrome 76.0.3809.100
• ChromeDriver 76.0.3809.126

Download ChromeDriver from https://chromedriver.chromium.org/downloads

brew install selenium-server-standalone
sudo pip install PIL
pip3 install virtualenv virtualenvwrapper
export VIRTUALENVWRAPPER_PYTHON=/usr/local/bin/python3
export WORKON_HOME=$HOME/.virtualenvsexport PROJECT_HOME=$HOME/Develsource /usr/local/bin/virtualenvwrapper.sh
brew install opencv
echo /usr/local/opt/opencv/lib/python3.6/site-packages >> /usr/local/lib/python3.6/site-packages/opencv3.pth

git clone https://github.com/lcyyzy/CaptchaCracker.git

CaptchaCracker
python cracker.py

In order to prevent being cracked, the background in the sources of the front end is often cut into many small pieces and displayed after re-splicing. The first step is to get the orginal background picture and the puzzle piece.

We can get the moving coordinates of each small background pieces from the elements of the front end and perform the inverse operation to restore the original background.

We can get the puzzle piece from the sources directly.

Extract the contour trajectory of the puzzle piece using cv2.findContours in the opencv and construct it into a two-dimensional matrix (operator).

Process the background by cv2.GaussianBlur and the Canny edge detection operator - cv2.Canny to find the edge in the background.

Use cross-correlation operation to process the edge of background and the contour operator of the puzzle piece. The position with the largest value is the result.

We mark the upper left corner of the position with a red point to show its correctness.

The server may check the track of the movement to determine whether it is human behavior or not. We should add some noises into the track and use selenium to simulate the dragging process.