6
$\begingroup$

I don't know where to ask this question, I hope I am in the right place to ask about this. So, what exactly is the fundamental difference between image and text encryption techniques?

link 1 link 2 link 3

I mean there are a lot of researchers (for example on the above links) that said AES, DES, RSA, etc. are not suitable for encrypting images because of image characteristics (high redundancy, high correlation, and large volume). Some of them imply that it will have a long encryption time because of the size and the low security it provides since the data originally has a high correlation. Hence they develop image encryption schemes using chaos theory, optical transform, random grids, DNA coding, compressive sensing, and so on.

Is the diffusion or confusion technique fundamentally different? If it is, how exactly does it differ from text encryption's confusion and diffusion? If it is not, why not just use text encryption by modifying AES or DES so that it takes an input of an image and output it as an image? I haven't found any articles that literally describe what differs between image encryption and text encryption techniques/methods.

$\endgroup$
6
  • 4
    $\begingroup$ Data is Data, as long as there is an additional requirement like Order Preserving Encryption the rest is "academia.stackexchange.com/questions/160509/…" $\endgroup$ – kelalaka Jan 1 at 10:40
  • $\begingroup$ "aes, des, rsa" Probably it's worth noting that: AES is suitable for encrypting all kinds of data. RSA is not suitable for most practical uses to encrypt data directly (see Hybdrid Encryption). And DES is not suitable to encrypt anything today - it's completely broken, (and 3DES is deprecated). $\endgroup$ – tylo Jan 1 at 16:33
  • $\begingroup$ is it true if i said that conventional encryption scheme like AES tend to have longer execution time than majority of image encryption because AES has a lot of encryption rounds? $\endgroup$ – Muhammad Fauzi Jan 1 at 17:20
  • 3
    $\begingroup$ @Muhammad Fauzi: not even so. Most papers on image encryption report encryption speeds orders of magnitude slower than modern academic crypto. For example your link 1 reports 29.21s and 543.28s for encryption and decryption of a 1024x1024 image (I guess 1 MiB), which is just interminable compared to a decent AES-CTR, even in software, not to mention hardware as used in practice. $\endgroup$ – fgrieu Jan 1 at 17:33
  • $\begingroup$ "AES, DES, RSA, etc. are not suitable for encrypting images because of image characteristics" I love this picture on Wikipedia of an encrypted image. That was with ECB mode; AES is fine if you use a mode like CBC or CTR. $\endgroup$ – David Ehrmann Jan 1 at 21:49
16
$\begingroup$

TLDR: None worth academic interest.


All modern encryption techniques intended for digital computers are applicable, and secure, for all kinds of digitized data, including text and image. There is no need for such encryption specialized to image or text. Argument: the modern baseline for encryption security is resisting Chosen Plaintext Attack, and that model assumes the attacker can choose the plaintext, including image in whatever format.

We can encrypt image, or pretty must anything digital, with e.g. AES-CTR or Chacha. These are widely available on most computer platforms, there is no known attack against them (baring side-channel attacks and IV reuse, which really are attacks on the implementation), and with a competent implementation, speed is faster than most processing peripheral to image processing, like compression, transmission, storage.

From a sampling of papers on image encryption (those at MDPI are free access), most have precisely no advantage over standard techniques: these paper's technique typically is slower and comes with weak security claims (often, relying on some standardized algorithm or metric; that's even though, assuming the existence of secure encryption, we demonstrably can't make a polynomial-time algorithm that recognizes good from bad encryption by examining ciphertext and plaintext). Reasons for why these papers are published (often, for a fee) are discussed there. Whatever these reasons are, peer-reviewed IACR publications seldom if ever buy them: I've only found preprints with this query.

Still, there are a few legitimate reasons to special-case image:

  • It calls for special techniques for the most efficient data compression (before encryption), in order to exploit the correlation between nearby pixels and characteristics of the human eye. There is an interaction with encryption, in that we must be careful that the compressed size does not leak exploitable information.
  • When we want the ciphertext to be a meaningful image: steganography, and Thumbnail-Preserving Encryption (where plaintext and ciphertext are images, and the ciphertext a thumbnail of the plaintext).
  • When we want decryption by analog means, such a superposition of transparencies. That's visual cryptography, a legitimate (if limited) field of research.
  • If we also wanted encryption by analog means. That's the goal of optical image encryption. It has been theorized in this (extremely cited) article (1995), and demonstrated experimentally (1996). I find several reports that it is practically broken in a ciphertext-only setup believable. There are countless variations, many simulations only with bogus security argument.
$\endgroup$
5
  • 1
    $\begingroup$ so basically an argument that said "image encryption is made to scramble high correlation and high redundancy in image type data" is bs because conventional encryption can also do that as well? $\endgroup$ – Muhammad Fauzi Jan 1 at 15:50
  • 2
    $\begingroup$ @Muhammad Fauzi: I let you decide the exact qualifier, but yes: modern encryption (anything academically recognized since the mid 1980s at least) is suitable for any type of digital data. $\endgroup$ – fgrieu Jan 1 at 16:00
  • $\begingroup$ I suppose honey encryption (Juels, Ristenpart, Eurocrypt 2014) for images could also be a legitimate use case for image-specific encryption schemes. However, most likely this would involve use of a standard encryption scheme together with a domain-specific encoding/compression strategy. The same would likely hold for some other use cases one might imagine, such as selective encryption/partial scrambling of content. $\endgroup$ – Polytropos Jan 1 at 21:35
  • 2
    $\begingroup$ As I mentioned in a chat, your IACR query does return an apparently reputable conference paper that doesn't seem to fit one of your "legitimate reasons". $\endgroup$ – nanoman Jan 2 at 7:32
  • 1
    $\begingroup$ @nanoman: thanks for pointing that definitely legitimate article! I've adjusted this answer, and now my question on academia. $\endgroup$ – fgrieu Jan 2 at 10:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.