3 ms·

I recovered the redacted text in this image: https://btcblockchain.files.wordpress.com/2015/02/e2.png https://btcblockchain.files.wordpress.com/2015/02/e2.png

uxcolumbo 8y ago
How did you approach this problem and solve it?
- Cyph0n 8y ago
  Not OP, but it seems like Electrum maps the wallet seed to standard English words. The other thing that helps is that the redaction isn't complete. So what I would do is: 1. Estimate the possible locations of spaces (e.g., using OpenCV). 2. "Guess" certain letters based on what is visible above and below the black bar; these are your constraints. 3. Look up dictionary words that meet the constraints found in (2). 4. Loop through steps 1-3 to build a list of possible seeds. 5. Run these seed candidates through Electrum.
  - btiown442 8y ago
    Yes this is generally the idea. Didn't know about OpenCV though. Was a font-matching and trial and error by hand in MS Notepad + some scripts to help brute force the thing kind of solution... Turns out there are actually at least a couple dozen possible solutions based on what is visible so some small amount of information was lost through the redaction, but not tons.
  - dempseye 8y ago
    Yes, it's a monospace font. The start, end and pattern of visible ascenders behind the redaction, along with the word list in electrum would allow you to reduce the search space by a lot, I suppose.
- btiown442 8y ago
  I'll give you a hint. Download the image. Because it is PNG that means it is lossless... Zoom in very closely on the top row of pixels above the redacted area and you may discover you have all that you need...
  - uxcolumbo 8y ago
    Thanks. I already did that, but there were quite a few letters that were completely covered. Programming isnt' my day job - but I have an interest in it, so was curious to see what tools / approach you used to automate this and whether it was fully automated or whether your solution recovered the partially covered letters and your 'wetware' recovered the rest ;) Cyph0n already provided an approach, which you confirmed - so thanks again.
- MauranKilom 8y ago
  My approach would be to brute-force match the pixels word by word.

uxcolumbo 8y ago

How did you approach this problem and solve it?

Cyph0n 8y ago
Not OP, but it seems like Electrum maps the wallet seed to standard English words. The other thing that helps is that the redaction isn't complete. So what I would do is: 1. Estimate the possible locations of spaces (e.g., using OpenCV). 2. "Guess" certain letters based on what is visible above and below the black bar; these are your constraints. 3. Look up dictionary words that meet the constraints found in (2). 4. Loop through steps 1-3 to build a list of possible seeds. 5. Run these seed candidates through Electrum.
- btiown442 8y ago
  Yes this is generally the idea. Didn't know about OpenCV though. Was a font-matching and trial and error by hand in MS Notepad + some scripts to help brute force the thing kind of solution... Turns out there are actually at least a couple dozen possible solutions based on what is visible so some small amount of information was lost through the redaction, but not tons.
- dempseye 8y ago
  Yes, it's a monospace font. The start, end and pattern of visible ascenders behind the redaction, along with the word list in electrum would allow you to reduce the search space by a lot, I suppose.
btiown442 8y ago
I'll give you a hint. Download the image. Because it is PNG that means it is lossless... Zoom in very closely on the top row of pixels above the redacted area and you may discover you have all that you need...
- uxcolumbo 8y ago
  Thanks. I already did that, but there were quite a few letters that were completely covered. Programming isnt' my day job - but I have an interest in it, so was curious to see what tools / approach you used to automate this and whether it was fully automated or whether your solution recovered the partially covered letters and your 'wetware' recovered the rest ;) Cyph0n already provided an approach, which you confirmed - so thanks again.
MauranKilom 8y ago
My approach would be to brute-force match the pixels word by word.