The concept is easy and creative: get the data from the file, encode into base64, compact and encode it again into hexadecimal, then get each chunks of hexa values (ex: ff 12 d3) to compose the RGB pixels (ex: #ff12d3) that represents the data as an encoded PNG image in such a way that it can be reversed and decoded as the original file.
This was a prove of concept written in Ruby (and parcially in Node.js) so I could upload 6GB of videos and PDFs to Flickr as images, and then get it back as the original files.
[Julian Assange, Cypherpunks 6.3MB PDF as a decodable 4MB PNG]
[
]
Simple text encoding sample:
"extrapolo".to_b64.compact.to_hex
=> "78da8b8ac83048ce752b4f32aa28e302001fbe0464"
final 7 hexa color pixels to represent "extrapolo" data as image
=> #78da8b #8ac830 #48ce75 #2b4f32 #aa28e3 #02001f #be0464
and back to the original data,
"78da8b8ac83048ce752b4f32aa28e302001fbe0464".from_hex.unzip.from_b64
=> "extrapolo"