It explains my thought process and setting up the problem
https://hyeonukim.github.io/devblog/categories/captcha-hacking/
- Inside ASM/src/ folder, run 'run.py'
- you should be able to see the problem in localhost:10000
- Inside Test/ folder, run 'run.py'
Detailed version of problem description
https://hyeonukim.github.io/devblog/posts/CaptchaHacking1/
This problem is one of Samsung's Capture the Flag question from 2017.
You are supposed to automate the process of solving 100 math problems within 80 seconds in a given Captcha Image
https://hyeonukim.github.io/devblog/posts/CaptchaHacking2/
We found that we don't need much images to train since all characters were in equal size, and we could distinguish each character using color codes
https://hyeonukim.github.io/devblog/posts/CaptchaHacking3/
Using the information about different color code, we could separate each characters and found ROI to save them into separate folder accordingly
https://hyeonukim.github.io/devblog/posts/CaptchaHacking4/
Using our cleaned data, we could train the data using KNN model and print out the math problem into string from an image
https://hyeonukim.github.io/devblog/posts/CaptchaHacking5/
We automated the process by getting the response of url to download the image then using our KNN model, we submit answer, repeat the process.