DESCRITION

A simple captcha breaker. It has been tested with only one type of numeric captcha image (found in the "Resources > in" folder. It extracts a 4-digits long string from an image using Pytesseract.

INSTALLATION

1. Download code.
2. Create a virtual environment and activate it.
3. pip install -r requirements.txt

USAGE

1. place captcha image in anyformat accepted by Pillow in "Resources > in".
2. In the root folder, open "captcha_solver.py" and change the "file_name" to the name of the captcha image uploaded in step 1.
3. Run it!
4. The function returns a 4-digit long string.

OTHER SETTINGS

1. captcha_solver.py

   1. avg_deviation: in order to reach absolute contrast, each pixel is compared with the average brightness of the whole image. If it's lower or equal than the sum of average and the user-defined avg_deviation, then the pixel is converted to pure black(pixel value of zero). Otherwise, it's converted to pure white (pixel value of 255). This setting is increase by 1 for each unsuccessful iteration of the captcha breaker, in case the returned value does not match the length ofcaptcha_text_length (see below).The avg_deviation also dictates how many iterations the captcha breaker should attempt before giving up.
   2. u_filter: short for scipy's "uniform filter", which compares each pixel to its direct neighbours and converts the pixel to the brightness of the majority of pixels that surround it. The u_filter value specifies how many pixels it should compare the current pixel with.
   3. captcha_text_length: the number of digits in the captcha.

2. Logger >init.py, line 6:logger.setLevel(logging.WARN)

   1. This sets the minimum log value that is printed onto the terminal when the code is run. To see all logs, change this tologger.setLevel(logging.INFO).

FURTHER DEVELOPMENT

1. Instead of onlyavg_deviation being changed in each iteration, ideally it would cover every possible permutation ofavg_deviation andu_filter until a solution is found.

1. This code rellies on Pytesseract to work. Ideally it would use its own trained model instead.

LICENSE

Apache License, version 2.0