No captcha !

Ali Nik

About Captcha

"Completely Automated Public Turing test to tell Computers and Humans Apart"

Prevent from :

  • Spam/SpamBot
  • DOS/DDOS
  • etc

Captcha Types

  • Normal Captcha
  • Text Captcha
  • ReCaptcha V2
  • Invisible ReCaptcha V2
  • ReCaptcha V1
  • ClickCaptcha
  • FunCaptcha
  • Audio Captcha
  • NuCaptcha
  • KeyCaptcha

"Which digit is 1st in the number 4581944?"

Breaking Captcha

 

  • Captcha Vuln. -- > Session ID / Cookies

  • OCR -- > Tesseract OCR / OpenCV
    • Removing Noises --> PIL

    • Separating Characters

    • Extracting Features

    • Classifying Features

  • Recognize Direct !
  • OCR -- > Tesseract OCR / OpenCV

Deep Learning Toolchain

  • Theano
  • Torch
  • TensorFlow
  • Caffe/Caffe2
  • PyLearn
  • Singa
  • Deeplearning4J
  • Keras
  • ...

Theano A CPU and GPU Math Compiler in Python

  • Base on Numpy

  • Implemented by Cython

  • Dynamic C code generation

  • GPU & CUDA

Keras is a high-level neural networks API, written in Python and capable of running on top of either TensorFlow or Theano.

  • Allows for easy and fast prototyping (through user friendliness, modularity, and extensibility).
  • Supports both convolutional networks and recurrent networks, as well as combinations of the two.
  • Runs seamlessly on CPU and GPU.

Recognise Direct

  • Dataset : Generate Captcha  --> python Captcha
  • Build Model : CNN --> Keras + Theano Backend
  • Training
  • Predict

 > No Captcha

       > spec.py

       > gen.py

       > load.py

       > model.py

       > train.py

       > pre.py

No Captcha Structure

 > No Captcha

       > spec.py

              > Captcha Format(Num,Alpha,Length,Size)

       > gen.py

              > Generate Random Text Based on Spec

              > Generate ImageCaptcha Based on Text

              > Labeling By Text and Save

       > load.py

              > Text2Vec Func. (Captcha Text To Vector Array)

              > Loading Dataset (Pre Proccess Image)

                    > Split to train/test Data

       > model.py

                    > Building CNN Model

                    > Create Weight File

       > train.py

                    > Input Data parameters

                    > Training with th backend : train Data

                    > Calculate Accuracy : test Data

       > pre.py

                    > Load Trained Model

                    > Lookup in Trained Model for textvec of Captcha

                    > Return Text of Captcha

Result

Dataset : 20K Numeric Chaptcha (0-9)

CNN Model : 3 Conv. Layer , 2 Fully-Connected Layer

Training : 15K Train Data , 5K Test Data

 

Total Prediction : 5000

Predict Correctly : 4845

Accuracy : 96.9%
 

 

No CAPTCHA Pentest !

Web Service For Solving Captcha :

  • Extracting Captcha Image From Web Page (Web Scraper)
  • Sending Captcha to No CAPTCHA prediction .
  • Receiving Result and Answer !

Contribute !

Thanks

No Captcha !

By 0xanik

No Captcha !

Captcha Penetration Test With Deep Learning !

  • 794