Reconcile three OCR readings before decoding printed code
At a glance
A Lobsters discussion documents combining Android OCR, Tesseract, and Claude to recover an exact encoded string from a T-shirt. This workflow is listed under OCR, Reverse Engineering, Verification.
- Agents / tools
- Claude, Tesseract, Android Circle to Search
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Obtain a readable image of the printed code block and have Android Circle to Search, Tesseract, and image-capable Claude available. The example was an encoded Bash program printed on a shirt. 2) Extract the text independently with all three tools, retaining each transcription. 3) Compare the outputs for mismatches and ask Claude to produce a table identifying the differing locations for visual inspection. 4) Check those locations against the original image and manually correct the remaining characters. 5) Decode the corrected Base64 string into text to inspect the recovered program.
How it went
The linked author's account shows a recovered script that prints an animated Peace for All message. Manual transcription cleanup remained time-consuming.
Worth knowing
The author notes that Base64 provides no error correction, so transcription must be exact; another shirt in the range contains incomplete code.