Digital Image Steganalysis

Center for Information Security Technologies, Korea University

Digital Image Steganalysis

Kwang-Soo Lee

Outline

Steganography

LSB Steganography

LSB Steganalysis

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. ???

@2*$#d(*%7*

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.

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.

Modern Steganography

Hiding information in digital objects, Invisibly.

The Invisibility must depend on just the stego-key, not the stego

system.

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.

11001000Extracting

Embedding

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

LSB Steganalysis

Steganalysis is the science of detecting hidden messages in digital signals.

It takes advantage of statistical or perceptual distinction of stego-signals fro

m cover-signals.

LSB steganalysisVisual attack, histogram analysis (PoV analysis),Closed color analysis,Regular-singular (RS) analysis,Sample pair (SP) analysis,LR Cube analysis,Etc.

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 Po

V.

cover-image histogram stego-image histogram

LSB

Em

bed

din

g

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0 1 2 3 4 5 6 7 8 9 10

Pixel value

Re

lativ

e f

req

ue

ncy

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0 1 2 3 4 5 6 7 8 9 10

Pixel value

Re

lativ

e f

req

ue

ncy

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.

cover-image histogram stego-image histogram

LSB

Em

bed

din

g

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0 1 2 3 4 5 6 7 8 9 10

Pixel value

Re

lativ

e f

req

ue

ncy

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0 1 2 3 4 5 6 7 8 9 10

Pixel value

Re

lativ

e f

req

ue

ncy

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.

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.

Center for Information Security Technologies, Korea University

Thank youKwang-Soo Lee

[email protected]