Digital Image Steganalysis

1
Digital Image Steganalysis

• Kwang-Soo Lee

2
Outline

• Steganography
• LSB Steganography
• LSB Steganalysis

3
Cryptography

• Cryptography scrambles a message to obscure its
  meaning.
• Today secure communication is often identified
  with cryptography.
• However, cryptography reveals the fact that
  communication is happening.

???
_at_2d(7
4
Steganography

• The word steganography comes from Greek,
  steganos and graphein.
• Steganography is the art of hiding information in
  ordinary-looking objects.
• Steganography aims to conceal the existence of
  secret communication.

5
Classical Steganography

• Examples
• Hidden tattoo,
• Covered writing,
• Invisible ink,
• Microdots,
• Character arrangement,
• Paper mask,
• etc.
• Hiding a secret message in physical objects.
• Secrecy depends on keeping the methods secret.

6
Modern Steganography

• Hiding information in digital objects, Invisibly.
• The Invisibility must depend on just the
  stego-key, not the stego system.

7
LSB Steganography

• Replacing least-significant-bits (LSBs) of
  digital data with message bits.
• Using digital multimedia, such as image, audio,
  video, as cover-objects.
• Embedding random message bits in LSBs will not
  cause any discernable difference from the
  cover-signals.
• Easy to implement, High payloads.

Embedding
11001000
Extracting
8
Digital Images for Steganography

• Types of digital images
• binary, gray-scale, RGB color, palette, JPEG,
  etc.
• The LSB plane of image data looks like random
  noise.
• Bit-plane decomposition of the Lena image in
  gray-scale.

lena.bmp
6th Bit Plane
4th Bit Plane
LSB Plane
9
LSB Steganalysis

• Steganalysis is the science of detecting hidden
  messages in digital signals.
• It takes advantage of statistical or perceptual
  distinction of stego-signals from cover-signals.
• LSB steganalysis
• Visual attack,
• histogram analysis (PoV analysis),
• Closed color analysis,
• Regular-singular (RS) analysis,
• Sample pair (SP) analysis,
• LR Cube analysis,
• Etc.

10
PoV analysis

• Proposed by Westfeld and Pfizmann (IH 1999) .
• PoV means a pair of values which differ just in
  the LSBs.
• 0 1 2 3 4 5 6 7 8
  9 10
• LSB embedding tends to equalize those frequencies
  of the values of each PoV.

LSB Embedding
cover-image histogram
stego-image histogram
11
Sample Pair Analysis

• Proposed by Dumitrescu et al. (IH 2003)
• Based on symmetry of quantized noise
  distribution.
• Take advantage of spatial correlation such as
  pixel adjacency.
• Estimate the length of hidden message.
• Outperform PoV analysis.

LSB Embedding
cover-image histogram
stego-image histogram
12
LR Cube Analysis

• Left and Right cube analysis (LRCA), developed by
  us (IH 2005)
• Our method uses high dim. vectors as basic units
  drawn from digital signals.
• Consider the vector noise distribution and its
  distortion of LSB embedding.

13
LR Cube Analysis

• Left cube and Right cube, and the possible cube
  patterns or complexities.
• Cover-signals show similar complex levels between
  the left cubes and the right cubes, but these are
  not the case for stego-signals after the LSB
  embedding
• LRCA works by measuring the similarities between
  these two distributions.

14
Thank youKwang-Soo Leekslee_at_cist.korea.ac.kr