Computer scientists invent a way to embed hidden info in ordinary text

A newly invented text steganographic (“hidden writing”) method can conceal information in seemingly plain text. The method is called FontCode, and it works by making minute alterations to the shapes of fonts in text to form an encoded message, an article in Science Daily states.

Many earlier text and document methods are limited to certain types of documents. But FontCode works with most documents as well as the majority of fonts. The hidden message will remain unchanged even after the document is printed out or saved as another file type.

A team of computer researchers from Columbia University (CU) came up with the new text steganographic method. They were overseen by Changxi Zheng, who also served as the senior author of the recently published paper.

FontCode can hide texts, metadata, and digital signatures in a text document or an image. It works for both digital documents and physical printouts. It is also compatible with most font families, word processing programs, and image-editing and drawing programs.

Every letter in a text can contain a perturbation. That means the only restriction on the encoded data is the amount of regular text in the carrier document. (Related: Cryptocurrency blockchains being sabotaged in “51 percent attacks” to steal virtual coins.)

New “hidden writing” method embeds your secret message in normal text

FontCode will change the width of a stroke, fiddle with the ascenders and descenders, or adjust the curves in letters. The perturbations are so minor that only an electronic eye would notice them.

The method can be instantly read by machines in the same manner as barcodes and QR codes. But unlike these striking codes, FontCode does not stand out from the printed material.

This anonymity protects messages embedded by FontCode. Potential spies will have to go every document in order to make sure nothing escaped them. Such a massive effort is not feasible, especially given the perturbations they are looking for are minuscule.

FontCode is not just limited to embedding data. It can also serve as an encryption method. Two parties can record the perturbations in a numbered location in a code book. Then they can hash out a private key that sets the locations or order of the perturbations.

FontCode is the first embedding method that does not rely on any particular document type. It is also the first one that can maintain the embedded data after the document gets printed or turned into another file type.

So you could embed a message in a Word file and save it as a PDF file, or save a JPEG image as a PNG file. The encoding will be preserved.

1,700-year-old Chinese math theorem helps decode cutting-edge embedding method

FontCode takes your secret message and turns it into a bit string. This is further changed into a sequence of integers, each of which is hidden in a five-letter block in normal text per the code book.

Recovering the message is a matter of reversing the process. If the file is a digital one or a photo from a smartphone, FontCode uses artificial electronic brains called convolutional neural networks (CNNs) to match each perturbed letter to the original.

Rasterized/pixel fonts like PNGs and IMGs require a different approach because their very nature might distort a part of the letter. To prevent the loss of data, the researchers employ the Chinese Remainder Theorem to recreate missing information.

The combination of ancient Chinese mathematics and modern electronic brain allows FontCode to recover the entire message despite losing 25 percent of the perturbations. Future versions of the method would be usable for other languages and character sets, such as Chinese and Japanese.

You can learn about more inventions like FontCode at Inventions.news.

Sources include:

ScienceDaily.com

CS.Columbia.edu