How to foil text analysis (for human eyes only)?

Share interesting information or links related to portable apps here.
Post Reply
Message
Author
User avatar
webfork
Posts: 10818
Joined: Wed Apr 11, 2007 8:06 pm
Location: US, Texas
Contact:

How to foil text analysis (for human eyes only)?

#1 Post by webfork »

If you're using a free email service, you probably know the data you send is getting turned upside down to look for userdata. Whether it's words, faces, patterns or even colors, machine learning is only going to get better at finding information.

One way to foil this is to send images as text that's been tweaked or otherwise modified to make it look odd. One old method to do this is the (now defunct) program "For Human Eyes Only" (homepage on archive.org) pictured below:

Image

As I'm currently working on a LibreOffice Draw template to easily add Portablefreeware.com watermarks, I tried to build something with light and dark values, and added some "lorem ipsum" text at the top to hopefully get it thrown out by machine learning tools.

Image

Anyway, this is far from fool proof and I don't expect to ever really solve this problem, just something I thought would be amusing. Any suggestions for a way to quickly and easily modify images in this way?

---

Related topics:

Obfusticated Email Link Creator
viewtopic.php?f=4&t=6568&p=41791

Self-destructive eMail messages
viewtopic.php?f=6&t=16360&p=49681

User avatar
webfork
Posts: 10818
Joined: Wed Apr 11, 2007 8:06 pm
Location: US, Texas
Contact:

Re: How to foil text analysis?

#2 Post by webfork »

Update here ... I put together two PhotoDemon filters that distorts text pretty well and pushed it into a Macro ...

Image
Image

Result:

Image

What follows is a simple macro to run both with the settings I used. Just paste this into a text file and save with a .PDM extension and run it from the Tools - Play Macro menu:

Code: Select all

<?xml version="1.0" encoding="UTF-8"?>
	
	<pdData>
		
		<pdDataType>Macro</pdDataType>
		<pdMacroVersion>8.2014</pdMacroVersion>
		<processCount>2</processCount>
		
		<processEntry index="1">
			<ID>Waves</ID>
			<Parameters><version>1</version><xwavelength>17</xwavelength><ywavelength>1</ywavelength><xamplitude>11</xamplitude><yamplitude>3</yamplitude><edges>0</edges><quality>5</quality></Parameters>
			<MakeUndo>4</MakeUndo>
			<Tool>-1</Tool>
		</processEntry>
		
		<processEntry index="2">
			<ID>Fog</ID>
			<Parameters><version>1</version><scale>36.6</scale><contrast>21</contrast><density>38</density><quality>6</quality><rndseed>3111219447.07667</rndseed></Parameters>
			<MakeUndo>4</MakeUndo>
			<Tool>-1</Tool>
		</processEntry>
		
	</pdData>

User avatar
lintalist
Posts: 434
Joined: Sat Apr 19, 2014 12:52 am
Contact:

Re: How to foil text analysis?

#3 Post by lintalist »

The PhotoDemon method doesn't do much I think, 100% accuracy with tesseract (free OCR appsuite https://github.com/tesseract-ocr - was also used by Google and fairly safe to assume all cloud tool providers have something similar). Just saying.

User avatar
Midas
Posts: 6705
Joined: Mon Dec 07, 2009 7:09 am
Location: Sol3

Re: How to foil text analysis?

#4 Post by Midas »

lintalist wrote: 100% accuracy with tesseract

I wondered as much. As they say, the proof is always in the pudding...

User avatar
lintalist
Posts: 434
Joined: Sat Apr 19, 2014 12:52 am
Contact:

Re: How to foil text analysis?

#5 Post by lintalist »

If you want to make it hard(er) you are probably better off to add noise, add lots of black dots, skew the page/image and blurring the text will help - although there are ways to pre-process the image to remove such noise and sharpen it to produce better results. Basically encrypting would be way to go, but the recipient needs to be able to handle it of course - although it doesn't have to be too complicated https://www.portablefreeware.com/index.php?sc=68 :)

Tesseract result of first image:
Hey dog, I hé@ard you like ppttes. I made you a phone so you could
phone while you phone,
edit: and second image
Agate purist will say, wnerrin doubt, just don't add th oO the database
on't like such a black-arid-white view of the world_Lvvill continue to ad
database that.-consider to be "freeware gh", and you decidé for yourself (by
@ app synopsis) whether its freeware.

User avatar
webfork
Posts: 10818
Joined: Wed Apr 11, 2007 8:06 pm
Location: US, Texas
Contact:

Re: How to foil text analysis?

#6 Post by webfork »

Definitely encryption is the best way to go but if your recipient won't or can't handle encryption, this is one way to make it at least little less visible.
lintalist wrote: Fri Mar 13, 2020 9:10 amyou are probably better off to add noise ...
Agreed. I was trying to do something that was both very readable and easy to reproduce.
lintalist wrote: Fri Mar 13, 2020 9:10 am...ways to pre-process the image to remove such noise and sharpen it to produce better results.
Yeah ... I just spent a bunch of energy earlier this year figuring out ways to clean up old JPEG files. But my thought is that takes time, effort, or at least processor cycles, so running that kind of analysis would be rare. Sort of a security through obscurity angle.
lintalist wrote: Fri Mar 13, 2020 9:10 amTesseract result of first image
Thanks for that. Some of this thread is sort of an excuse to mess with two programs I really enjoy. But it only really makes sense if I test what I come up with on this thread. :)

Post Reply