Hiding Information

1
Hiding Information in the Placement of
Maneuverable Words

• Ibrahim Kamel
• Salma Banawan

2
Presentation Contents

• Hiding information in English text
• Steganography Scenario
• Transforms based on Maneuverable Words
• Transforms Evaluation
• Conclusions and Future Work

3
Using English text for hiding information

• Previously
• Hiding information exploited redundancy in image
  and video data types
• Motive
• English text is the most available media type on
  the Internet
• Major Constraint
• Although redundancy is available in English
  language, word order is crucial for maintaining
  syntax and semantic integrity

4
Steganography Scenario

• The Classic Prisoners Problem
• Alice and Bob (prisoners in different cells) want
  to send secret messages in the presence of Wendy
  (the warden).
• Our scenario
• Alice, Bob and Wendy are average users of the
  English language and their communication entails
  acceptable grammar of somewhat casual English
  statements.
• Alice and Bob send cover texts in which the
  location of maneuverable words (established
  between them) hides part of the secret message.
• Meaning and syntax structure of cover should not
  arise Wendys suspicion
• An encoded cover document to be acceptable if the
  grammar is correct and the meaning of the cover
  document is left undamaged.

5
Transforms Based on Maneuverable Words

• We have indentified transforms that relocate
  specific sentence constituents for hiding part of
  a secret message.
• Criteria for Transforms
• Reversible
• Receiver can identify the hidden information from
  the cover without the original document
• Independent
• More than one transform can be applied to one
  sentence

6
Overview of Transforms

• Tranformations for Hiding Information Based on
  Maneaverable Keywords

• Splitting Separable Phrasal Verbs
• Swapping Operands of a Boolean Operator
• Re-arranging Conditional Statements
• Adverb Placement

7
1) Splitting Separable Phrasal Verbs

• Phrasal Verb a verb combined with an adverb
  particle
• Examples Call off Turn on
• Many phrasal verbs are separable by the object
• The two possible word orders are
• Verb Adverb Particle Object ? represents 0
• Verb Object Adverb Particle ? represents 1

Phrasal Verb Bit Representation Bit Representation Bit Representation
Phrasal Verb Meaning Sample Sentence Bit
call off cancel Well have to call off the party. 0
call off cancel Well have to call the party off. 1
turn on switch on the electricity Turn on the lights. 0
turn on switch on the electricity Turn the lights on. 1
8
2) Swapping Operands of Boolean Operator

• In Boolean logic, certain operators share the
  commutative property
• A operator B B operator A,
• (where A and B are operands of an operator such
  as AND, OR, and NOR.)
• Coordinating conjunctions, such as and, or,
  nor are analogous to Boolean operators found in
  Boolean logic.
• A -- coordinating conjunction -- B ? represents
  0
• B -- coordinating conjunction -- A ? represents
  1

9
2) Example of Encoding Boolean Operands

• The decision of which operand represents A and
  which one is B is part of the convention agreed
  upon between the sender and receiver
• For example, A might be the operand with the
  larger sum of the ASCII values comprising it

AND I would like to go to the park and to the zoo. 0
AND I would like to go to the zoo and to the park. 1
NOR Neither Arwa nor Jena wants to play. 0
NOR Neither Jena nor Arwa wants to play. 1
OR We can eat cake, cookies, or ice cream. 00
OR We can eat cake, ice cream, or cookies. 01
OR We can eat cookies, cake, or ice cream. 10
OR We can eat ice cream, cookies, or cake. 11
10
3) Rearranging Conditional Statements

• If two propositions, p and q, appear in the form
  p?q, the statement is a conditional statement
  interpreted as if p then q

Conditional Statement Formats Conditional Statement Formats
Format Re-arranged Format
if p, q q, if p
for q, p p, for q
when p, q q, when p
Type Bit Representation Bit Representation Bit Representation
Type Sample Sentence Format Bit
p? If you complete your homework early, we can attend the party. If p,q 0
p? We can attend the party, if you complete your homework early. q, if p 1
?q In order for us to attend the party, complete your homework early. for q, p 0
?q Complete your homework early, in order for us to attend the party. p, for q 1
11
4) Adverb Placement

• An adverb is a modifier which can help to
  describe any sentence constituent except the
  subject (i.e. verb, object, or adjective).
• The minimum sentence contains a subject and a
  verb.
• A transitive sentence contains all three parts
  subject, verb and object.

Subject
Verb
.
Object
10
11
00
01
Adverb
Adverb
Adverb
Adverb
12
4) Adverbs Modifying Auxiliary Verbs

• The main verb may be modified by an auxiliary
  verb, or a helping verb
• Adverbs associated with auxiliary verbs, such as
  already, also, and always may actually be
  located at 5 places.

General Transitive Sentence with auxiliary verb Also, she can play baseball. 00
General Transitive Sentence with auxiliary verb She, also, can play baseball 01
General Transitive Sentence with auxiliary verb She can also play baseball. 10
General Transitive Sentence with auxiliary verb She can play, also, baseball. 10
General Transitive Sentence with auxiliary verb She can play baseball also. 11
13
Preserving SemanticsSwapping Operands of
Boolean Operator

• We concentrate on encoding correlative
  conjunctions such as both A and B, either A
  or B and neither A nor B.
• We will avoid encoding operands of and that are
  in sentences displaying chronological order.
• For example, the meaning of this sentence would
  be unacceptable if we switch the operands
• First, we ate lunch at home, and then we ate
  dinner at the restaurant.
• We will also avoid encoding operands of or and
  nor that are in sentences displaying priority.

14
Preserving SemanticsConditional Statements

• Nested Conditionals
• Established protocol for encoding nested
  conditionals.
• Identifiers and Determiners/Pronouns

Nested Conditional Bit Representation Bit Representation Bit Representation Bit Representation
Nested Conditional Sample Sentence Format Rating Bit
If p1 If you find bananas when you go to the supermarket, please buy some. If p1, when p2, q accept-able 00
If p1 If you find bananas, please buy some, when you go to the supermarket. If p1, q, when p2 accept-able 01
When p2 When you go to the supermarket, if you find bananas, please buy some. when p2, if p1, q accept-able 10
When p2 When you go to the supermarket, please buy some, if you find bananas. When p2, q, if p1 unacceptable 11
15
Identify Determiners

• How can we make this sentence acceptable?
• When you go to the supermarket, please buy some,
  if you find bananas.
• Be sure the identifier of the determiner appears
  in a clause before the determiner.
• A speech tagger can be used to swap the
  identifier with its corresponding determiner.

16
Capacity for Hiding Information

• The transforms are independent of each other
• For example, the sentence
• Typically, she eats both cereal and toast for
  breakfast, can handle 3 transformations and
  read
• For breakfast, she typically eats both toast
  and cereal.
• 4 bits can be hidden in the above sentence
• Therefore, the capacity for encoding a random
  cover document will be considered as the total
  percentages of appearance of all the keywords per
  a Google Book statement.

17
Transforms Evaluation
Statistics from Google Books Corpus
Transform Keywords
1) Phrasal Verbs Verb adverb particle 0.1
2) Operands of Boolean Operator Correlative conjunctions 0.3
3) Conditional Statement Subordinating conjunctions 3.0
4) Adverb Placement Capital and lowercase Adverbs 1.5
TOTAL All Keyword Appearance 4.9
23.85
18
Embedding Capacity in a Random Page of Cover Text

• If we consider each transform embeds 1 bit, then
  we can embed approximately 1.2bits/sentence
• In a random page of text, which may contain
  approximately 20 sentences, we should be able to
  encode 24 bits of information.
• However, adverbs generally encode 2 bits.
• the percentage of appearance of keywords for
  adverbs should be multiplied by 2
• Therefore, the total appearance of keywords for
  transforms would be 6.4, meaning we can embed up
  to 1.5 bits/sentences.
• Approximately 30 bits in a random page of 20
  sentences.

19
Conclusions

• Proposed is a new scheme for hiding secret
  messages in regular English text
• exploits the redundancy that exists in the
  different locations of maneuverable words
• Comprised of 4 feasibly automatable transforms
  splitting separable phrasal verbs, swapping
  operands of Boolean operators, re-arranging
  conditional statements, and adverb placement
• Using Google Books Corpus, we have shown that the
  keywords which signal these transforms are widely
  found in English text
• Previous work on syntactical transforms attained
  embedding capacity of 0.5bits/sentence.
• Our statistic analysis shows that we should be
  able to embed up to 1.5bits/sentence in a random
  English document.

20
Future Work

• Apply implementation of transforms on a corpus to
  verify statistical analysis results
• For increasing capacity
• Identifying the most suitable type of cover text
  for each transform
• Proposals for increasing embedding capacity in
  conditional statements
• Take advantage of the fifth location that can be
  automated for the placement of adverbs

21

• THANK U ?
• Thank you.
• Thank You.
• Thanks for being a super audience.
• For being a super audience, thanks.
• SHUKRAN.
• SHUKRAN jazeelan.
• (after this presentation all the above should all
  be quite suspicious!)