Data Hiding in Binary Image for Authentication and Annotation

1
Data Hiding in Binary Image for Authentication
and Annotation

• Min Wu, Member, IEEE
• and Bede Liu, Fellow, IEEE

2
Introduction (1/2)

• Goals - Propose a new method to embed data in
  binary images.- Blind !!- The hidden data can
  also be extracted after high quality printing and
  scanning.
• The technique is possibly as alternative to or in
  conjunction with the cryptographic authentication
  approach.

3
Introduction (2/2)

• Major Challenge- Because the pixels of image
  take value from only two probabilities, hiding
  data without causing visible artifact becomes
  more difficult !
• Unremovable is not our main consideration because
  there is no obvious threat of removing of
  embedded data in many authentication applications.

4
Proposed Method
introduction

• Some issues- How to select pixels? Flipping
  Pixels- How to embed data in order to be
  blind? Enforce a certain relationship- The
  uneven embedding capacity.
  Shuffling

5
Proposed Method
Flipping pixels

• The Strategy- Score each pixel between 1 and 0
  according to neighbor patterns.- block base
  (3X3)
• The parameters of scoring- smoothness and
  connectivity
• Details of the approach are given later

6
Proposed Method Embedding
Mechanism (1/2)

• If we want to let the method be blind, directly
  encoding hiding data is not appropriate.-
  because the flippable of pixel will be changed
  after embedding.(i.e the flipping score is
  changed.)

7
Proposed Method Embedding
Mechanism (2/2)

• Strategy- We embed data to enforce a certain
  relationship on low-level features of a group of
  pixels.- Ex. To embed a 0 in a block, some
  pixels are changed so that the total number of
  black pixels in the block is an even number.

8
Proposed Method Uneven Embedding Capacity and
Shuffling (1/3)

• Problem The distribution of flippable pixels
  may vary dramatically from block to block.
• Solution1 variable embedding rate from block to
  block .. NOT appropriate!!The Reasons 1)
  A detector has to know exactly how many bits are
  embedded in each block.2) The overhead for
  conveying this side information is significant!

9
Proposed Method Uneven Embedding Capacity and
Shuffling (2/3)

• Solution2 Constant embedding rate and use
  shuffling to equalize the uneven embedding
  capacity.

10
Proposed Method Uneven Embedding Capacity and
Shuffling (3/3)

• Error correcting coding can be used to handle a
  very small number of blocks that have no
  flippable pixels.
• Shuffling also can enhances security.

11
Applications(1/3)

• Signature in Signature- annotate a signature in
  such applications as faxing signed documents and
  storing digitized signatures as transaction
  records.- advantage 1) user-friendly 2)
  visualize 3) integrate the
  authentication data with the
  signature image

12
Applications(2/3)

• Invisible Annotation for Line Drawing- Artist
  may wish to annotate their drawing with
  information.

13
Applications(3/3)

• Tamper Detection for Binary Drawings- The hidden
  data can be an easily recognized pattern to the
  content of the host image.

14
Robustness(1/3)

• Bit Error- We consider a simple odd-even case. A
  single pixel is changed, the bit embedded in this
  block will be decoded in error. However, more
  pixels are changed, it may be de decoded
  correctly.- The probability of getting a
  wrongly decoded bit is (under the assumption
  that the change is independent from pixel to
  pixel)

15
Robustness(2/3)

• - If the total number of changed pixels in the
  whole image is small, it is likely that most of
  those pixels are involved in different embedding
  blocks hence extracted bits from those blocks
  will be wrong.
• Using Shuffling has the disadvantage of
  increasing the sensitivity against geometric
  distortion such as translation Cropping!!

16
Robustness(3/3)

• In order to extract the hidden data from printing
  and scanning, we must recover the origin image.-
  oversampling- add registration marks

17
Security

• The major objective of an adversary is to forge
  authentication data so that an altered document
  can still pass authentication tests.
• Two issues- 1) The probability of making content
  alternations while preserving m-bit embedded
  authentication data - have been
  discussed!- 2) The possibility for an adversary
  to hide specific data in an image - arise
  uncertainty!

18
Future Work

• Further refinement of flippability model for
  different type of binary images.
• The recovery of binary image from high quality
  printing and scanning using fewer or no visible
  registration marks