Pattern-based Data Hiding for Binary Image Authentication by Connectivity-preservingHuijuan Yang, Alex C. Kot, IEEE Fellow

IEEE Transactions on Multimedia, Vol. 9, No. 3, Apr. 2007

Multimedia Security Final Project

R97922062 葉容瑜 R97922003 程瀚平

Introduction Proposed Method The Authentication Mechanism Experimental Results Conclusions

Introduction(1/3)

Digital documentsEx. certificates, digital books, fax, personal documents

How to ensure the authenticity and integrity of digital documents, as well as detection of tampering and forgery, become a serious concern

Powerful image editing software Data hiding for binary images authentication has

been a promising approach to alleviate these concerns

Introduction(2/3)

Data hiding on binary images can be done the lower level: flipping pixels from black to white and

vice versathe higher level: modifying width of strokes and

spacings between characters and words

In this paper, our focus is on data hiding for binary images in lower level for the purpose of image authentication

Introduction(3/3)

Define a “connectivity-preserving” criterion to assess the “flippability” of a pixel

Connectivity among pixels plays an important role to their visual qualities

Wu et al.’s approach Proposed approach

Visual distortionConnectivitySmoothness

4-connectivity8-connectivity

Uneven embeddability of the image

ShufflingEmbeddable blocks/Embeddable pixels(cryptographic signature)

The Main Objectives

1. Assess the “flippability” of a pixel using the connectivity-preserving criterion to achieve good visual quality of the watermarked image

2. Handle the “uneven embeddability” of the image by adaptively embedding the watermark only in those “embeddable” blocks

3. Study the invariant features in flipping pixels in binary images to achieve blind watermark extraction

4. Explore different ways of partitioning the image to achieve larger capacity

5. Investigate on how to locate the “embeddable” pixels in the watermarked image so as to incorporate cryptographic signature to achieve higher security

Introduction Proposed Method

Flippability DecisionBlock PartitionEmbeddabilityCapacitiesWatermark Embedding and Extraction

The Authentication Mechanism Experimental Results Conclusions

Flippability Decision

FlippabilityThe transitions from the pixel to its eight neighbors in a

3 * 3 blockIn particular, the 4- and 8-connectivity among pixels

VH TransitionIR TransitionC Transition

VH Transition

Nvw: the number of uniform white transitions along vertical and horizontal directions

Nvb: the number of uniform black transitions along vertical and horizontal directions

Black: 1White: 0

Nvw = 0, Nvb = 2

=> Nvw = 0, Nvb = 0

Nvw = 0, Nvb = 0

=> Nvw = 0, Nvb = 0

IR Transition

Nir: the number of the interior right angle transitions

Black: 1White: 0

Nir = 0

=> Nir = 1

Nir = 0

=> Nir = 0

C Transition

Nc: the number of transitions from the center pixel to the sharp corners

Black: 1White: 0

Nc = 1

=> Nc = 0

Nc = 0

=> Nc = 0

Flippability/Connectivity-Preserving Criterion Flippable

VH transition, IR transition, and C transitionremain the same before and after flipping the center pixel

Flip the pixel will not Destroy the connectivity b/w pixels in the

neighborhood(VH)Create extra clusters as well(IR)Destroy the 8-connectivity among pixels(C)

By satisfying the “Connectivity-Preserving” criterion, the local connectivity is preserved

Block Partition

Several different types of blocksFixed 3*3 block (FB)Non-interlaced block (NIB)Interlaced block (IB)

Embeddability

Determined pixelsNon-interlaced block scheme:

all pixels except the boundary pixelsInterlaced block scheme:

all pixels except those lie in the sharing rows and columns

The embeddability of a block depends on the “flippability” of the determined pixels in the block

Capacities

Only one pixel is flipped in each block

=> The prob. of a pixel to be “flippable” in a block is independent to other pixels

Assume the probability that a pixel satisfies the “Flippability Criterion” is pFB: The prob. of each block to be “embeddable” is pNIB: The prob. is 1 – (1-p)^(n-2)2

IB: The prob. is 1 – (1-p)^(n-2)2

A larger block size definitely will increase the prob.for a block to be “embeddable”, however, the total number of blocks will be decreased

=> Decrease the capacities

Watermark Embedding

1. Partition the image into equal size square blocks, note that the block size does not need to be square

2. Determine the flippability of the determined pixels based on the “Flippability Criterion”

3. Once a pixel is identified as “flippable”, the block is marked as “embeddable”. The current “flippable” pixel is identified as the “embeddable” pixel, i.e., “embeddable” location of the block

4. Proceed to the next block5. Repeat steps 2 to 4 until all the blocks are processed6. Embed the watermark in the “embeddable” blocks by

flipping the “embeddable” pixels (if needed) to enforce the odd-even feature of the number of black or white pixels in the block

Embeddable pixels = flippable pixels Flipping a pixel in a block may affect

the “flippability” of the pixels in the same block but not the pixels in its neighboring blocks

The “embeddability” of a block is invariant in the watermark embedding processThe “flippability” of a pixel is invariant in the watermark

embedding processA “flippable” pixel which is identified as “embeddable”

is still “flippable,” hence an “embeddable” block remains “embeddable”

The watermark can be extracted blindly from the “embeddable” blocks by computing the odd-even feature of the number of black or white pixels

Introduction Proposed Method The Authentication Mechanism

Locate “Embeddable” Pixels CriterionAuthentication ProcessThe Verification Process

Experimental Results Conclusions

Locate “Embeddable” Pixels Criterion

The odd-even enforcement is employed for the watermark embeddingVulnerable to the “parity attack”

Ex:

an adversary can carefully flip two pixels in the

same block while keeping the odd-even feature of

the block unchanged.

Locate “Embeddable” Pixels Criterion

p-4 conditionFlipping the pixel that does not change the

“flippability” of its previous four (p-4) neighbors that lie in the same 3 x 3 block

d-2 conditionFlipping the pixel that does not affect the

“embeddability” of those d-2 pixels (determined pixel) that have already been processed in the same block

Locate “Embeddable” Pixels Criterion

Authentication Process

The Verification Process

Introduction Proposed Method The Authentication Mechanism Experimental Results

Capacity and VisibilityTest Locating Embeddable Pixels Criterion and

Authentication MechanismComparisons

Conclusions

Capacity and Visibility

Capacity and Visibility

Capacity and Visibility

(a)The original text image of size 336 x 336 (Chinese)

(d) Hide 482 bits by FB 3 x 3

(e) Hide 733 bits by NIB 4 x 4

(f) Hide 1261 bits by IB 4 x 4

Capacity and Visibility

(b) The original text image of size 336 x 336 (English)

(g) Hide 447 bits by FB 3 x 3

(h) Hide 672 bits by NIB 4 x 4

(i) Hide 1237 bits by IB 4 x 4

Capacity and Visibility

(b) The original text image of size 336 x 336 (Handwritten)

(g) Hide 313 bits by FB 3 x 3

(h) Hide 554 bits by NIB 4 x 4

(i) Hide 972 bits by IB 4 x 4

Capacity and Visibility

Evaluate the visual distortion caused by flipping pixelsThe visual distortion table proposed by Wu et al. is

employed.

M. Wu and B. Liu, “Data hiding In binary images for authentication and annotation,” IEEE Trans. Multimedia, vol. 6, no. 4, pp. 528–538, Aug. 2004.

)(1)( iFCiDS

Capacity and Visibility

Distortion score (DS)

Total distortion (TD)

Average per pixel distortion (APPD)

)(1)( iFCiDS )(1)( iFCiDS

n

i

iDSTD1

)(

nTDAPPD /

Capacity and Visibility

)(1)( iFCiDS

Test Locating Embeddable Pixels Criterion and Authentication Mechanism

(a) The original image of size 920 x 230

(b) Hide 1056 bits by proposed algorithm with FB 3 x 3

(c) The watermarked image that is tampered

(d) The original logo image

(e) The reconstructed logo image when no tampering occurs

(f) The reconstructed logo image when the watermarked image has been tampered

Comparisons

(a) Original image of size 173 x 115

(b) The proposed method (c) Wu et al. method

(d) Tseng et al.

(e) Lu et al.

(f) Yang & Kot

111 bits

180 bits

260 bits

Introduction Proposed Method The Authentication Mechanism Experimental Results Conclusions

Conclusions

A novel blind data hiding scheme for binary images authentication based on connectivity-preservingA window of 3 x 3 is employed to access the

“flippablility” of a pixel in a block

Different types and sizes of block can be chosen cater for different applications

The proposed scheme can be applied to a wide variety of binary image authentication