• Share
  • Email
  • Embed
  • Like
  • Save
  • Private Content







Total Views
Views on SlideShare
Embed Views



0 Embeds 0

No embeds



Upload Details

Uploaded via as Microsoft Word

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
Post Comment
Edit your comment

    MIMIC-PPT.doc MIMIC-PPT.doc Document Transcript

    • MIMIC-PPT: Mimicking-based Steganography for Microsoft Power- Point Document 1 Yuling Liu, 1Xingming Sun, 1Yongping Liu and 2Chang-Tsun Li 1 School of Computer and Communication, Hunan University, Changsha, China 2 Department of Computer Science, University of Warwick, London, England Corresponding author: Xingming Sun Abstract: Communications via Microsoft PowerPoint (PPT for short) documents are commonplace, so it is crucial to take advantage of PPT documents for information se- curity and digital forensics. In this paper, we propose a new method of text stegano- graphy, called MIMIC-PPT, which combines text mimicking technique with characte- ristics of PPT documents. Firstly, a dictionary and some sentence templates are auto- matically created by parsing the body text of a PPT document. Then, cryptographic information is converted into innocuous sentences by using the dictionary and the sen- tence templates. Finally, the sentences are written into the note pages of the PPT doc- ument. With MIMIC-PPT, there is no need for the communication parties to share the dictionary and sentence templates while the efficiency and security are greatly im- proved. Keywords: Text steganography, linguistic steganography, text mimicking, information security, digital forensics INTRODUCTION these days, and digital text has diverse Communications via digital texts forms, such as webpage, e-mail, various have long been a commonplace for per- types of formatted text documents, in- sonal, business, or academic purposes in cluding PDF, DOC, PPT, and so on. Thus, 1
    • it is convenient to transmit secret mes- Current implementations of text sages by using text documents as the steganography exploit spacing flexibility mediums. in typesetting by making minute changes There are two main techniques to to the layout of different components and protect private communication of text to the kerning in order to encapsulate documents. The first technique is crypto- hidden information. The key limitation of graphy, which encrypts a message to this approach is that it is vulnerable to make it unintelligible to humans. Thus, simple retypesetting attacks. The other those who do not possess the secret key important method of text steganography cannot obtain the original message. Most is linguistic steganography based on the researchers have made a great deal of ef- knowledge of natural language fort on that. However, an encrypted processing. It is much more ambitious, in communication always arouses suspicion that it should survive attempts to remove (Petitcolas, et al., 1999). The second hidden information through file reformat- technique is text steganography, which ting, OCR or retyping (Topkara, et al., refers to the hiding of information within 2005 ). Publicly available methods of text documents (Murphy and Vogel, linguistic steganography can be grouped 2007). Unlike cryptography, the goal of into two categories. The first group of text steganography is to convey secret methods, called text mimicking technique, messages in text documents, by conceal- is based on directly generating a new ing the existence of a covert communica- cover text for a given message. The tion (Bergmair, 2007). second group of methods is based on 2
    • linguistically modifying a given cover We can directly write encrypted informa- text in order to encode a message,while tion or whitespace characters into the preserving the meaning as much as poss- note pages for the purpose of secret ible (Chiang, et al., 2004; Topkara, et al., communication. However, it is difficult 2006). Due to the sensitivity of modify- that the contents of the note pages are in- ing a given cover text, however, the terrelated with the contents of the body amount of hidden information is limited. text and resist attacks by humans or ma- Therefore, this paper falls in the former. chines. PPT is a presentation program de- In this paper, we propose a new ste- veloped by Microsoft for its Microsoft ganographic method for hiding data in the Office system. A PPT document is com- note pages of PPT documents by utilizing posed of one or more sheets of slides. text mimicking technique, called MIM- Each sheet of slide in a PPT document IC-PPT. To provide an opportunity for may contain several text frames and a deniability, we first create a dictionary note page. All the text frames of a PPT table and a sentence template database by document constitute the body text, while parsing the body text of a PPT document. all the note pages are accessorial expla- Then we randomly select a sentence tem- nations, which are often ignored by care- plate and substitute parts-of speech for less readers and not visible to the au- words in accordance with the assigned dience when presenting. Therefore, the binary bits. The experimental results note pages provide a useful vehicle for show that it is feasible to send a secret hiding information in a PPT document. message in the note pages along with a 3
    • PPT document. MIMIC-PPT is not only analysis, but it will not stand up to any dictionary-free, but also can effectively analysis that understands the grammar generate meaningful sentences correlated structure. In order to improve the results, with the body text to be written into the Peter Wayner proposes a method to gen- note pages of the PPT document. erate texts using probabilistic con- In order to disguise cryptographic text-free grammars and to hide informa- information as normal communications tion according to the choices it makes to thwart the censorship of ciphertext, it (Wayner, 2002). These generated texts is necessary to introduce text mimicking are grammatically correct. technique, which converts ciphertext into Another development in text mi- text that looks innocuous natural lan- micking is Stego (Walker, 1994), a mi- guage text. Publicly available imple- micry method proposed by John Walke. mentations of linguistic steganography By using a user-defined dictionary, Stego mainly rely on this technique. converts a binary file (secret message) The primary text mimicking method into a text that resembles natural lan- is proposed by Peter Wayner (Wayner, guage. The text has structure, but does 1992,1995,1997,1999). In his basic mi- not comply with any grammar rule. micry algorithm, the method recodes a A later development in text mi- text so that its statistical properties of micking is Texto (Maher,1995), which characters are more like that of another includes a “structs” file that contains different natural language text. The text some usually-correct English sentence may fool attacks based upon statistical structures, and a “words” file which 4
    • contains 64 verbs, 64 adjectives, 64 ad- selecting words for types in accordance verbs, 64 places, and 64 things. In order with the assigned binary code. The chal- to facilitate exchange of binary strings, lenges are to create large and sophisti- especially encrypted data, Texto can cated dictionaries and to create mea- transform uuencoded or pgp AS- ningful sentence templates (Chapman CII-armoured ASCII data into English and Davida, 1997). Later, Chapman et al. sentences. (2001) describes an “extensible contex- A successful development in text tual template” approach combined with a mimicking is NICETEXT (Chapman and synonymy-based replacement strategy, Davida, 1997), a mimicry method pro- so that more realistic text is generated. posed by Mark Chapman. NICETEXT is Chapman and Davida (2002) extends the an improvement over Texto. The original NICETEXT protocol to enable deniable NICETEXT approach generates a set of cryptography/messaging using the con- meaningful English sentences by large cepts of plausible deniability. In addition, code dictionaries and sentence templates. Essam A. El-Kwae proposes a new tech- In their dictionaries, almost 175,000 nique for hiding multimedia data in text, words are categorized into 25,000 types, which is similar to NICETEXT. It intro- and within each type a word is assigned a duces some marker types, which are spe- unique binary code. Each sentence tem- cial types whose words do not repeat in plate contains a sequence of word-types. any other type. Each generated sentence The encoder generates a text by ran- must include at least one word from the domly choosing a sentence template and marker types (EI-Kwae and Cheng, 5
    • 2002). naries and sentence templates between Different from the above text mi- the communication parties. Furthermore, micking techniques, Sams Big G Play- the generated text not only relates closely Maker (PlayMaker for short) only utiliz- to the body text of the PPT document, es normal sentence templates without a but also simulates certain aspects of the dictionary (GMBH, 2000) . In the system, writing style of the body text. Then the each letter or symbol is corresponding to text is written into the note page, which a normal sentence of a play book. is an intrinsic part of a PPT document, All the above methods are effective, and security is thus achieved. and they can generate cover texts directly. MIMIC-PPT However, the texts produced by these Similar to other text mimicking methods are often implausible to human techniques, such as the NICETEXT sys- readers, and it is unusual to transmit the tem, dictionaries and sentence templates texts between the communication parties. are necessary in MIMIC-PPT. However, Moreover, these methods need a great the dictionaries and sentence templates amount of resources (both the time and need not be transmitted between the effort) to design a sophisticated dictio- communication parties. By utilizing ex- nary or a good predesigned grammar. On isting linguistic tools, the senders and the the other hand, the proposed MIM- receivers can automate the creation of a IC-PPT in this paper provides legitimate dictionary table and some sentence tem- cases in using an existing PPT document. plates according to the following rules: And there is no need to share the dictio- Rule 1 and Rule 2. To make our descrip- 6
    • tion clear, two definitions are presented speech are merged in a line. We record first as follows. their occurrences as an extra attribute. Definition 1. Content words are words The basic form of the dictionary table that have meaning, such as nouns, verbs, consists of Part-of-speech, Word, Occur- adjectives, adverbs. rences. Definition 2. Function words are words Rule 2 (Sentence Template Creation that exist to explain or create grammati- Rule): For each sentence in the body text, cal or structural relationships into which we preserve function words and punctua- the content words may fit,such as pro- tions while replacing content words with nouns, prepositions conjunctions, deter- parts of speech to obtain a sentence tem- miners, interrogatives, and so on. plate. Rule 1 (Dictionary Table Creation Rule): MIMIC-PPT is divided into two First, we extract and segment the body processes, Hiding Process and Retrieving text of a PPT document using the exist- Process. During these two processes, ing morphological analyzer (Toutanova there is no need to share dictionaries and et al., 2003; Zhang et al., 2005), which sentence templates. Details of the hiding can fulfill the task of word segmenting process and the retrieving process will be and part-of-speech tagging. Then, we described later. pick up all the content words to obtain a Hiding Process: In order to hide a secret crude dictionary table, where each word message in a PPT document with the text and its part-of-speech are on a single line. mimicking technique, the hiding process The same words with the same parts of consists of three stages: a preprocessing 7
    • stage, a generating stage, and a writing tence template database is a set of sen- stage. The preprocessing stage is to au- tences with function words, punctuations tomatically create a dictionary table and and parts of speech of content words. some sentence templates according to A secret message M is encrypted to Rule 1 and Rule 2, and to encrypt the se- get an m -bit binary string C c1c2 cm , cret message into a binary string. The where each ci is a bit. Because it is un- generating stage is to convert the binary likely that m equals the number of bits string into a set of innocuous sentences required to terminate the generated sen- by utilizing the dictionary table and the tence at the end of a sentence template, sentence templates. The generated sen- or the end of a word, the length of mes- tences are related to the body text of the sage is added in front of C and strings of PPT document. The writing stage is to random 0’s and 1’s are appended to the write the sentences into the note pages of end of C . That is, we hide into the PPT the PPT document to obtain a ste- document a binary go-document. string C' l1l2 lL c1c2 cm r1r2  , with In the preprocessing stage, we in- | C | l1l2 lL being the length of the secret troduce Rule 1 to create a dictionary ta- message with the value m , and ri being ble D and Rule 2 to automate construct- the appended bits that are selected ran- ing a sentence template database S , domly. The senders and the receivers which is a set of sentence templates. The should agree on the value of L before- dictionary table D includes all the content hand, such that the receivers can fully words in the body text, while the sen- recover C in the retrieving process. 8
    • After the preprocessing stage, the is replaced with the proper word in the binary string C ' is converted into some corresponding small tables, thus a gener- innocuous sentences by utilizing the dic- ated sentence is obtained. tionary table D and the sentence tem- In the writing stage, we firstly plates database S . First, the dictionary compare the number of sentences pro- table is partitioned into 4 small tables duced in the generating stage with the according to the parts of speech of the number of slides of the PPT documents. words, and all the words of each small Then, the sentences are written into the table are mapped into binary codes using note pages of the PPT document evenly. Huffman coding. The previously men- The details of the hiding process are tioned occurrences of words are used to presented in the algorithm below. assign variable-length Huffman codes to Algorithm 1: Hiding Algorithm different words. Short Huffman codes Input: a PPT document T ; and a message are assigned to words with higher occur- to be hidden C ' . rences and longer ones to those with Output: a stego-document T ' . lower occurrences; this results in the Steps: frequently-occurring words having to be 1) Preprocess the body text of the PPT used more often in the generated sen- document T to extract a dictionary tences. Then, a sentence template is se- table D , and a sentence template da- lected randomly. According to current tabase S {s1 , s2 ,, sn } , where s s is a bits of the binary string C ' , each sentence template. part-of-speech of the sentence template 2) Partition the dictionary table D into 4 9
    • small tables D1 , D2 , D3 , D4 according to plate ss S . the set of parts of speech P {n, v, a, d } ; 4) For each tt P , substitute word ww for and construct a Huffman part-of-speech tt to generate a new tree H x for Dx,x 1, 2,3, 4 , as follows. sentence ss ' , where word ww is deter- a) Create a leaf node ni for each mined as follows. word in Dx , and assign the oc- a) Starting from the root of tree H x , currences of each word oi to the traverse H x to its left child if the node ni . current bit of C ' is 0 or to its right b) Initialize a set Q to contain all of child. the leaf nodes. b) Go to the next bit of C ' and con- c) Find in Q node ni and n j with the tinue traversing in a similar way lowest occurrences; and then until ww is reached. remove node ni and n j from Q . 5) Repeat Steps 3) through 4) until the d) Create a new node nij with the end of C 'L m . occurrences oij oi o j , and as- 6) Write the generated sentences sign ni as its left child and n j as its s1 ' s2 ' into the note pages of T to right child. yield a stego-document T ' . e) If Q is empty, then tree H x has Retrieving Process: In the retrieving been constructed and take nij as process, we firstly create a dictionary its root; else, add node nij to Q and table according to Rule 1. And then, we go to Step 2c). utilize the dictionary table to decode the 3) Randomly select a sentence tem- corresponding binary bits of the words 10
    • parsed from the note pages. The sentence part-of-speech w1 / t1 w2 / t2  . templates have nothing to do with the 5) If tt P , decode the corresponding retrieving process. The details are de- bits of word ww in the following way: scribed in the following algorithm. a) Starting from the root of the Algorithm 2: Retrieving Algorithm Huffman tree H x , traverse it to Input: the stego-document T ' . the leaf node ww and record the Output: the message C ' . path traversed. Steps: b) Analyze the path traversed, and 1) Preprocess the body text of ste- set the current bit of C ' to 0 if go-document T ' to create a dictionary the path goes down a left child; table D . or to 1. Go to the next bit 2) Partition the dictionary table D into 4 of C ' for each child traversed. small tables D1 , D2 , D3 , D4 according to 6) Repeat Step 5) until C 'L m has been the set of parts of speech P {n, v, a, d } ; retrieved. and construct a Huffman EXPERIMENTS AND RESULTS tree H x for Dx using Step 2) described MIMIC-PPT is applicable to any in Algorithm 1. language that has a morphological ana- 3) Extract all the note pages from the lyzer or part-of-speech tagger, e.g. Eng- stego-ducument T ' . lish, Chinese, and Japanese. Different 4) Parse sentences of the note pages by from English texts, Chinese texts are ex- using the morphological analyzer to plicit concatenations of characters, and obtain a sequence of words with words are not delimited by spaces. Thus, 11
    • it is more difficult and challengeable to Tagger (Toutanova et al., 2003) to obtain implement MIMIC-PPT for Chinese texts. a sequence of words with parts of speech. According to the algorithms presented in Then, we pick up all the content words, Section 3, we utilize Stanford Log-linear and record the occurrences of each word. Part-Of-Speech Tagger (Toutanova et al., And then we assign each word a Huffman 2003) to implement an English MIM- code according to the occurrences to ob- IC-PPT system and Chinese morphologi- tain a dictionary table. Due to the limit of cal analyzer IRLAS (Zhang et al., 2005) space, Table 2 shows the occurrences and to implement a Chinese MIMIC-PPT the resulting Huffman codes for the small system. In both systems, we assume that table of adverbs. According to punctua- the note pages of PPT documents have no tions, we segment the body text sentence texts. If there are some sentences in the by sentence. Each sentence is to replace note pages, we delete the existing sen- all the content words with the corres- tences and write the generated sentences. ponding parts of speech to obtain a sen- For the ease of description, we firstly take tence template. Some selected sentence the English PPT document “Practical templates are shown in Table 3, where Writing” at the URL <n>, <v>, <a>, and <d> represent parts http://sfl.xjtu.edu.cn/center/writing/up/11 of speech of a noun, a verb, an adjective, 47021618.ppt for example. and an adverb respectively. We take the Firstly, the body text is extracted abstract of this paper as a secret message from the PPT document, and tagged by to be encrypted, and designate the length the Stanford Log-linear Part-Of-Speech of message L 16 . Table 4 shows some 12
    • sentences generated by the English sentences are evenly written into the note MICMIC-PPT system. Finally, these pages of the PPT document. Table 2: A dictionary table of adverbs Part-of-speech Word Occurrences Huffman Code Adverb Never 2 000 Adverb carefully 2 001 Adverb only 1 0100 Adverb verbally 1 0101 Adverb also 2 011 Adverb not 5 10 Adverb far 1 11000 Adverb too 1 11001 Adverb there 1 11010 Adverb again 1 11100 Adverb really 1 11100 Adverb so 1 11101 Adverb precisely 1 11110 Adverb already 1 11111 Table 3: Some sentence templates No. Sentence template 1 <n> or <n> <n>. 2 <v> <a> that your <n> <v> <a>, <a> and <a>. 13
    • 3 <v> of <v> on <n>. 4 <v> the <n> by <v> and <v> the <v> <n> ; 5 <n> of <n> <v>. 6 What <v> the <n> about? 7 <n>the<n><d>,<v> to <v> out <d> what the <n> <v> about ; 8 How <a> <n> <v> <d> in the <n> and what <v> their <n>? 9 To <v> what <v>, <v> <n> in a <a> <n>. 10 <v> two or <a> <n> to <v> a <a> <n>. Table 4: Some generated sentences No. Generated sentence 1 Idea or events step-by-step. 2 Tell straightforward that your charts writing sure, loyal and much. 3 Be of figure on scene. 4 Tell the order by are and tell the following information; 5 Series of events writing. 6 What stretch the order about? 7 Practical the object also, reading to writing out carefully what the scene help about; 8 How useful step-by-step writing never in the expository-composition and what is their order? 9 To writing what be, is Pie in a faithful observe. 10 Illustrated two or coherent perspective to used a orderly details. 14
    • Table 5: Comparison of several systems The systems Words Bytes Expansion Lexical Syntactically Semantically Rate Items Correct Coherent Mimicry 36076 151682 232.64 Valid Yes No applet Stego 3716 15720 24.11 Valid No No Texto 2150 9010 13.82 Valid Yes No Nicetext 9721 41995 64.41 Valid Yes No PlayMaker 4829 21095 32.35 Valid Yes No Spammimic 11513 46172 70.81 Valid Chinese 2243 4592 7.04 Valid Yes No MIMIC-PPT Compared with the existing systems lowing website of linguistic steganography, as mentioned (http://www.pku.edu.cn/cernet2004/pptli before, MICMIC-PPT system can gen- st.htm) as an example. The numbers of erate texts more efficiently and securely. words and bytes of the generated texts In order to evaluate the efficiency, we are listed in Table 5, where words of the take the same message (the abstract of Chinese MIMIC-PPT system are Chinese this paper) as input to generate texts by characters. Because of the inherent dif- using the existing systems and the Chi- ferences between English and Chinese, nese MIMIC-PPT system. Thereinto, we one byte (8 bits) represents an English take the first PPT document on the fol- letter, while two bytes (16 bits) represent 15
    • a Chinese character. We also introduce texts, except for the text produced by the expansion rate to measure the effi- Stego. It is because Stego is only dictio- ciency, which is the ratio of the number nary-based, while not complying with of bytes of the generated text divided by any sentence template. Due to limits of the number of bytes of the secret mes- current automatic semantic analysis, we sage. The results indicate that the expan- manually evaluate semantic coherence. sion rate of the Chinese MIMIC-PPT Every individual sentence of the gener- system is lower than other systems. This ated texts makes sense. However, it is achieved for the reason that we pick all should be noted that the sequences of the content words, which are most fre- sentences of all the generated texts do quently used, and we utilize Huffman not have coherent contexts. Some results coding to avoid discarding any content of several systems are also shown in Ta- word. ble 5. To demonstrate the qualities of the SECURITY ANALYSIS texts produced by these systems, three Different from existing linguistic levels of linguistic correctness are con- steganography methods, to transmit a ducted, namely lexical level, syntactic generated text along with a PPT docu- level and semantic level. Utilizing some ment is more reasonable and secure on existing resources of lexical and syntac- MIMIC-PPT system. It is normal to send tic analysis, it is observed that all the and receive a meaningful PPT document generated texts contain valid lexical via the Internet. And a note page is an items, and they are syntactically correct essential part of each slide in PPT docu- 16
    • ments, which provides accessorial de- comprehensible text, as mentioned in scription for the presentation. Through Table 5. However, the sequence of sen- parsing the generated text of the Chinese tences will be later written into the note or English MIMIC-PPT systems, all the pages evenly, so the necessity for seman- words are the content words used in the tically coherence between sentences body text of PPT documents, and most should not be taken as an absolute re- words are high-frequency words. Addi- quirement of the MIMIC-PPT system. tionally, the sentence templates are also Each sentence produced by the MIM- the styles of the body text of PPT docu- IC-PPT system is derived from the sen- ments. Therefore, the notes will simulate tence templates of the body text, thus it the content and the writing style of the is possible to draw attention from a hu- body text so that it can provide the op- man reader. In addition, first encrypting portunity of deniability. Deniability is the secret message before the generating derived from the fact that even if an ad- stage can ensure that an adversary cannot versary finds the notes suspicious, the obtain the real secret message even if he sender may claim that the notes are the or she knows our algorithm. real explication of the representation. To strengthen the security of the Due to the random choosing of the MIMIC-PPT system, the generated text sentence templates derived from existing should be as imperceptible as possible to sentences in the body text, the sequence adversaries. This can be achieved by the of sentences generated by the MIM- following ways. The first one is to dy- IC-PPT system does not add up to an namically select a key-dependent subset 17
    • of the dictionary table. The other one is niability and be applied to other docu- to utilize natural language generation ment formats that offer notes. Addition- techniques for the purpose of creating ally, the generating stage is not done by sophisticated sentence templates. More- creating large and sophisticated type dic- over, not all the sentence templates de- tionaries beforehand; rather, the dictio- rived from the body text are appropriate nary and sentence templates are auto- for the generating stage, and thus it is matically generated by parsing the body necessary to choose some sentence tem- text of PPT documents. Therefore, the plates to obtain a sentence template da- communication parties need not share tabase according to some selection rules. dictionaries and sentence templates. As a In addition, a PPT document can be ex- result of Huffman coding in each small tended to the form of a Microsoft Po- dictionary table, the generated text can werPoint Show document (PPS for short), not only mimic the statistical properties in which the notes are fully invisible for of the body text, but also enhance the the readers. efficiency. CONCLUSION Because most expressions of the This paper presents a new method body text in PPT documents are concise, of text steganography for Microsoft Po- the sentence templates extracted from werPoint documents, MIMIC-PPT. The them are simpler than general sample body text is mimicked to generate some texts. In addition, the content of each sentences to be written into the note note page is not closely related to the pages. Thus, the method can provide de- content of the slide when the generated 18
    • sentences are written into the note page Contents, San Jose, CA, 2007. evenly. We intend to investigate some E. A. EI-Kwae and L. Cheng, 2002. HIT: natural language generation techniques A New Approach for Hiding Mul- in the near future for improving the sen- timedia Information in Text. Pro- tence templates, and adaptively writing ceedings of the SPIE 4675: 132-140, the generated sentences into the note San Jose, CA, 2002. pages. F. A. P. Petitcolas, R. J. Anderson, and M. ACKNOWLEDGEMENTS G. Kuhn, 2007. Information Hid- This paper is supported by Key ing-A Survey. Proceedings of the Program of National Natural Science IEEE, 87(7): 1062-1078, 1999. Foundation of China (No. 60736016), H. P. Zhang, T. Liu, J. S. Ma and X. T. National Basic Research Program of Liao, 2005. Chinese Word Segmen- China (No.2006CB303000), and Nation- tation with Multiple Postprocessors al Natural Science Foundation of China in HIT-IRLab. Proceedings of (No.60573045). SIGHAN, pp.172-175, 2005. REFERENCES J. Walker, 1994. Steganosaurus. Circu- B. Murphy and C. Vogel, 2007. The lating on the web, December 1994. Syntax of Concealment: Reliable http://www.fourmilab.ch/stego/steg Methods for Plain Text Information o.shar.gz. Hiding. Proceedings of 9th Confe- K. Maher, 1995. Texto. Circulating on rence on Security, Steganography, the web, February 1995. and Watermarking of Multimedia http://www.ecn.org/crypto/soft/texto 19
    • .zip. ternational Conference on Informa- K. Toutanova, D. Klein, C. D. Manning tion Security, LNCS, 2200: 156-165, and Y. Singer, 2003. Feature-Rich 2001. Part-of-Speech Tagging with a Cyc- M. T. Chapman and G. I. Davida, 2002. lic Dependency Network. Proceed- Plausible Deniability Using Auto- ings of HLT-NAACL, Edmonton, mated Linguistic Steganography. pp. 173-180, 2003. Proceedings of International Con- M. T. Chapman and G. I. Davida, 1997. ference Infrastructure Security, Nicetext. Website, August 1997. LNCS, 2437: 276-287, 2002. http://www.nicetext.com/. M. Topkara, C. Taskiran and E. J. Delp, M. T. Chapman and G. I. Davida, 1997. 2004. Natural Language Water- Hiding the Hidden: A Software marking. In Delp, Edward J., Wong System for Concealing Ciphertext Ping W, editers, Proceedings of the as Innocuous Text. Proceedings of SPIE, 5681: 441-452, San Jose CA, the First International Conference 17-20 January 2005. on Information and Communica- M. Topkara, G. Riccardi, D. H. Tur, and tions Security, 1997. M. J. Atallah, 2006. Natural Lan- M. T. Chapman, G. I. Davida and M. guage Watermarking: Challenges in Rennhard, 2001. A Practical and Building a Practical System. Pro- Effective Approach to Large-scale ceedings of International Confe- Automated Linguistic Steganogra- rence on Security, Steganography phy. Proceedings of the Fourth In- and Watermarking of Multimedia 20
    • Contents, San Jose, CA, 2006, pp. nography. Proceedings of 9th Con- 106-117. ference on Security, Steganography, P. Wayner, 1992. Mimic Functions. and Watermarking of Multimedia Cryptologia, XVI(3): 193-214, Contents, San Jose, CA, 2007. 1992. S. GmbH, 2000. Sams Big Play Maker. P. Wayner, 1995. Strong Theoretical Ste- Website, 2000. ganography. Cryptologia, XIX(3): http://www.scramdisk.clara.net/play 285-299, 1995. /playmaker.html. P. Wayner, 1997. Mimicry Applet. Web- Y. L. Chiang, L. P. Chang, W. T. Hsieh, site, August 1997. and W. C. Chen, 2004. Natural http://www.wayner.org/texts/mimic. Language Watermarking using Se- P. Wayner, 1999. Mimic Functions and mantic Substitution for Chinese Tractability. Technical Report, Cor- Text. Proceedings IWDW 2003, nell University, Department of LNCS, 2939: 129-140, 2004. Computer Science, 1999. P. Wayner, 2002. Disappearing Crypto- graphy: Information Hiding: Stega- nography and Watermarking, 2nd edition. San Mateo, CA: Mor- gan-Kaufmann, pp.81-128, 2002. R. Bergmair, 2007. A Comprehensive Bibliography of Linguistic Stega- 21