Open Source Radar
A Python proof-of-concept decodes a motion-noise CAPTCHA using a fixed pipeline (no ML). It demonstrates a reproducible filter approach rather than per-challenge learning.
16stars
0forks
0issues
2026since
Star historydaily snapshots by VibeCrowd
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
Reviewgenerated from repository data · Jul 18, 2026
What it is
A fixed, ~150-line Python script (no machine learning, no training data) that turns a motion noise CAPTCHA into an answer by applying a sequence of image processing steps and OCR. The repo provides a concrete pipeline and a sample video to reproduce results.
How it works
The pipeline (solve.py) performs:
- Decode frames from the input video.
- Auto-detect the animated region with high temporal variance.
- Apply a spatial high-pass filter to each frame.
- Do dense block-matching between consecutive frames to estimate local motion (magnitude of shift).
- Short-window average and median de-band to produce filled glyphs.
- Run tesseract OCR with majority voting to produce the final answer. All steps 3–5 are described as identical across challenges, implying a fixed filter rather than per-challenge learning.
Getting started
Reproduce instructions from the README:
# needs: python3 (numpy, pillow), ffmpeg, tesseract
pip install numpy pillow
python3 solve.py captcha.mp4 out
# -> prints SOLUTION: AT ; writes out/solved_montage.png and out/solved.gif
Recent releases
- none
Traction
Stars: 16
Behind the repo
No linked startup or company information provided in the repository data.
Caveats
- License: none listed
- Language: Python
- Created: 2026-07-15
- Last push: 2026-07-15
- Reproduces on a lossy sample video; the note clarifies this is a lower bound on exposure of the real CAPTCHA against uncompressed streams.