https://github.com/jboirazian/captchacv2

A simple proof of concept for beating outdated Captcha methods using only opencv and numpy
https://github.com/jboirazian/captchacv2

Last synced: about 2 months ago
JSON representation

A simple proof of concept for beating outdated Captcha methods using only opencv and numpy

• Host: GitHub
• URL: https://github.com/jboirazian/captchacv2
• Owner: jboirazian
• Created: 2024-03-05T21:01:31.000Z (10 months ago)
• Default Branch: main
• Last Pushed: 2024-03-05T21:47:11.000Z (10 months ago)
• Last Synced: 2024-03-05T22:26:03.746Z (10 months ago)
• Language: Python
• Homepage:
• Size: 48.8 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# CaptchaCV2

A simple proof of concept for beating outdated Captcha methods using only opencv and numpy

## Summery

Many outdated websites and mobile apps still use the following captcha method:

![image](https://github.com/jboirazian/CaptchaCV2/assets/21143405/e0f80433-8e73-4cb9-a5b3-6f65eb995c65)

While you might think that it can be solved with your favorite OCR framework , the reality is that it can be solved by using only [Template matching](https://docs.opencv.org/4.x/d4/dc6/tutorial_py_template_matching.html)

### Method

In order to perform to perform Template matching... we are gonna need templates. In the folder **/selected_patterns** you will find the 3 pairs 0 to 9 diggits.

These images were reconstructed from 1000 digits of the same number and position in order to obtain the real number without that sky blue jitter :

![image](https://github.com/jboirazian/CaptchaCV2/assets/21143405/75e70ec7-6f1a-4ce6-93b2-9ebf06ce5361) ![image](https://github.com/jboirazian/CaptchaCV2/assets/21143405/641b76f1-abc0-4802-b0ad-b7952faf7f60) ![image](https://github.com/jboirazian/CaptchaCV2/assets/21143405/47169727-ec31-4349-a487-bf2c0f9bc4b4)

## How to run it

+ Install numpy and cv2

+ git clone https://github.com/jboirazian/CaptchaCV2.git

+ python3 verify.py

### Example

```python3

import modules.digit_recogniton as dr

import cv2

if __name__ == "__main__":

    img=cv2.imread('example.png', cv2.IMREAD_GRAYSCALE)

    print(dr.predict_number(image=img))

```

## Performance

Without the need of a GPU and with the provided image example , the prediction was achived in less than 5 ms on my PC.

I was also able to run in on an really old Android table using [Termux](https://termux.dev/en/) and got around 50 ms