Information Hiding

• Steganography
• Digital watermarking
  • Fingerprinting
• Methods
• Attacks

Classification

Why is it important?

• Copy protection have never really worked - needs tamper-proof hardware
• Instead we can try to detect copies
  • Tracing original owner
  • Detecting who made the copy
• Sending secret messages without using cryptography

Who is interested?

• Military and intelligence agencies
• Criminals
• Law enforcement and counter intelligence
• Secret communication without encryption

Terminology

• The information to be hidden (watermark, fingerprint, a secret message) is embedded in a cover object (cd, video, text) giving stego object (marked object).
• The embedding is done with a key, a secret variable that is in general known to the object's owner.
• The recovery of the embedded mark may or may not require a key

Steganography

• General problem:
  • Alice and Bob are in prison and need to communicate with each other to discuss their break out plan
  • All messages have to through the warden Wendy
  • If the warden notice any suspicious messages all communication between Alice and Bob will stop
• Lot of historical examples (invisible ink, microfilm, etc.)

Wisdom from cryptography

• Kerckhoffs principle:
  • Assume that the attacker knows the methods
• Secure steganography:
  • An opponent who doesn't know the key cannot obtain any evidence of the communication

Pure steganography

• No key is necessary
• E: C x M -> C
• D: C -> M
• Relies on the secrecy of the method

Secret key steganography

• Uses a shared and secret stego-key
• E: C x M x K -> C
• D: C x K -> M

Public key steganography

• Two keys: one public and one private

Active warden

• It is conjectured that neither public key steganography nor pure steganography is possible with an active (malicious) warden
• A public key needs to be certified - Alice and Bob have no means of certifying the keys when all communication goes through the warden Wendy.

Security of steganography systems

• Three parts: detecting, extracting and disabling embedded information
• Requirements for secure steganographic algorithms:
  • Messages are hidden using a public algorithm and a key; the secret key must identify the sender uniquely
  • Only the holder of the correct key can detect, extract, and prove the existence of the hidden message. Nobody else should be able to find any statistical evidence of a message's existence
  • Even if the enemy knows the contents of one hidden message, he should have no chance of detecting others
  • It is computationally infeasible to detect hidden messages

Digital watermarking

• Additional requirement: robustness
• There should be no way of removing the embedded information without rendering the cover object unusable
• Visible watermarks
• Imperceivable watermarks
• Fingerprinting: a unique watermark in each object

Applications

• Watermarking for copyright protection
  • Embed information about the owner of the object
  • Used resolve rightful ownership
  • Requires a very high level of robustness
• Fingerprinting for traitor tracking
  • Each object have unique watermark identifying for example the buyer of a object
  • Copies can be traced
• Automatic audit of radio transmission
• Data augmentation
• Automatic monitoring of copyrighted material on the Web

Robustness

• Watermarks should resist any kind of distortion introduced by standard or malicious data processing
• No such method has be proposed so far
  • Not clear yet if it is possible
• IFPI (International Federation for the Phonographic Industry) robustness requirements:
  • The marking mechanism should not affect the sonic quality of the sound recording
  • The marking information should be recoverable after a wide range of filtering and processing operations (D/A and A/D conversion, MPEG compression, adding a second mark with the same system, etc.)
  • There should be no other way to remove or alter the embedded mark without sufficient degradation of the sound quality as to render it unusable

Methods

• LSB
• Echo Hiding
• More advanced uses other domains: Discrete Cosine Transform (DCT), Fourier transform, Wavelets

Least Significant Bit tweaking

• Represent the object as vector of integers
• Change the least significant bit in either all or some integers to represent a 1 or 0 in the mark
• For sound: e.g. 16 bit samples
• For images: e.g. red, blue or green components, luminance or chrominance components
• Depending on the amount of embedded information it is quite imperceivable

Echo hiding

• Advanced audio marking systems
• Inserts echos (delay in the 0.5-2 ms range) that cannot be perceived
• Two different echos gives ones and zeros
• Survives compression
• Obvious attack: detect and remove echo

Attacks

• Robustness attacks
• Presentation attacks
• Interpretation attacks

Attacks - Jitter Attack

• Attacks against audio marking systems that tweak low order bits whose locations is specified by a key
• First implementation:
  • Split the signal into chunks of 500 samples
  • Duplicate or remove on sample from each chunk, giving chunks of 499 or 501 samples
  • "Almost imperceptible"
  • Impossible to find the mark - we cannot use the key to find the bits anymore
• Better implementation:
  • Resample the chunks at a lower or higher frequency

StirMark

• Most image marking schemes survive basic manipulations: rotation, compression, resizing
• Combinations are often not handled though
• StirMark: tool developed for simple robustness testing of image marking systems
• Applies minor distortions
• Defeats most commercial image marking systems

StirMark

• Images are slightly stretched, sheared, bent, rotated etc.

Mosaic Attack

• 'Presentation Attack'
• Designed to defeat systems combining watermarks and web crawlers
• The image is chopped into smaller pieces and put together in the web page in e.g. a table
• Each sub-picture must be small enough such that the (partial) watermark can no longer be detected (for that sub-picture)

Summary

• New field computer security - general theories do not exist yet
• It is still not clear if perfect (with regard to robustness) watermarking systems is possible
• Most commercial system is easily defeated

More information

Example questions from exams

• 90-100 points
• Allowed aids:
  • Writing and drawing tools
  • English/Native language dictionary

11 Points

Explain the following cryptanalytic attacks:

a) Ciphertext-only attack. (2p)

b) Known-plaintext attack. (3p)

c) Chosen-plaintext attack. (3p)

d) Chosen-ciphertext attack. (3p)

Solution

• Ciphertext-only attack.
• The cryptoanalyst has the cipher-text of one or several messages. We want to recover the plain-text or (better) the key.
• Known-plaintext attack.
• We have not only the cipher-text, but also the plain-text of several messages. We want to recover the key.
• Chosen-plaintext attack.
• We get to select the plain-text that gets encrypted. We want to recover the key.
• Chosen-ciphertext attack.
• We can chose what cipher-texts get decrypted.

12 Points

Suppose that the substitution and permutation used in each round of a DES system when taken together map every 32-bit value to zero regardless of the value of its input. What output would this DES then produce? Justify your answer. Otherwise, even if you get the correct answer, you will get ZERO points.

DES

• Input = Output

5 Points

Write a short explanation (one or two sentences is enough) of the following terms:

1. Malicious code
2. Covert channel
3. Control (as used in computer security)
4. Reference monitor
5. Inference attack

Solution

1. Malicious code

Code that behaves in a way unexpected by its designer or user.
2. Covert channel

A way to leak information to parties that should not receive that information. The communication travels unnoticed with other, proper communications.
3. Control (as used in computer security)

A countermeasure, something we do to improve security.
4. Reference monitor

A portion of an operating system kernel (or a database system) that controls and checks all accesses to objects. Needs to be trusted at all times.
5. Inference attack

A statistical attack against a database. Secret data is inferred from non-secret data.

15 Points

There are many reasons why personal computer security management is a problem. Describe at least 5 controls for personal computer security, and motivate why they enhance personal computer security.

Solution

• Perform regular backups
• Practice separation of duty
• Use strong operating systems (with access control, memory protection, file protection etc.)
• Educate users
• Secure the equipment
• Use virus software
• Don't leave computers unattended

5 Points

What are the security risks with the Unix set userid (setuid, suid) feature? (The feature were the privileges of a program running is not the privileges of the user that runs it, but the privileges of the owner of the program executable file.)

Solution

• If there is an exploitable flaw (intentional or unintentional) in the program, an attacker might gain access to the whole system (e.g. if the owner of the setuid program has system administration rights, such as root in Unix).

10 Points

Discuss two advantages of paging as a memory protection scheme and two advantages of segmentation as a memory protection scheme. What are the disadvantages of the different memory protection schemes?

Solution

• Segmentation:
  • + Users can share access to segments
  • + Users cannot access unpermitted segments
  • - Programmers need to be aware of segments
  • - Unefficient
• Paging:
  • + Same as segmentation
  • - (Limited) fragmentation

5 Points

Passwords are often stored in encrypted form to enhance security. Why is it a security risk to allow access to the encrypted passwords, even though the encryption might be one-way (that is, there is no way to decrypt a password)?

Solutions

1. Brute force: use one or several fast computer to try all combinations of passwords by encrypting them and comparing them to the encrypted passwords.
2. Dictionary attack: instead of trying all the different passwords it is often enough to try only likely passwords, e.g. the names of the users, words in a dictionary.

10 Points

Describe at least two different ways to do secure authentication of users over a network. Describe your assumptions of the security of the network and the hosts, and eventual weaknesses with the different schemes.

Solution

• Examples: Public key cryptography (e.g. as used in PGP, SSL, ssh, etc.), Kerberos, Automatic call back, etc. See the book for details.