Subscribe Now Subscribe Today
Abstract
Fulltext PDF
References

Research Article
Inverted Pattern in Inverted Time Domain for Icon Steganography

Rengarajan Amirtharajan and John Bosco Balaguru Rayappan
 
ABSTRACT
Information technologies and communications have pervaded our homes and business places. No matter how well-organized and extensive the communication technology is there are always loop holes in the network and people who seek after the clandestine information to extract from these loop holes. The pandemic problem of security is still a raging problem as every solution compromises on one trait to heighten the other(s) according to the necessity of the hour. In this paper we suggest an algorithm where we have tried to retain all the three important banalities of secure communication: robustness, capacity and imperceptibility by using Haar Integer wavelet transform a Discrete Wavelet Transform (DWT) domain method in tandem with a cryptic scheme i.e. the direct binary or inverted binary embedding of data. Experimental results that compute the MSE and PSNR show that this algorithm caters to the need of the hour by delivering the high capacity with good imperceptibility.
Services
Related Articles in ASCI
Similar Articles in this Journal
Search in Google Scholar
View Citation
Report Citation

 
  How to cite this article:

Rengarajan Amirtharajan and John Bosco Balaguru Rayappan, 2012. Inverted Pattern in Inverted Time Domain for Icon Steganography. Information Technology Journal, 11: 587-595.

DOI: 10.3923/itj.2012.587.595

URL: http://scialert.net/abstract/?doi=itj.2012.587.595
 
Received: November 15, 2011; Accepted: January 09, 2012; Published: February 22, 2012

INTRODUCTION

In this age of booming communication and technology, everything is just one phone call or e mail or a flight away. The free flow of information has behooved the need for protecting this huge wealth of knowledge that has from time to time helped tremendously in growth of science, technologies, civilizations and mankind as a whole. With the invention of various algorithms for intelligent storing, processing and transmitting of information ranging from minutiae procedures to long exhaustive routines, many techniques have been developed for computer security, information security and information assurance. The most prominent amongst them is steganography (Stefan and Fabin, 2000, Petitcolas et al., 1999; Rabah, 2004; Cheddad et al., 2010). It involves communicating secret data in an appropriate multimedia carrier (Bender et al., 1996; Rabah, 2004), e.g., image (Hmood et al., 2010a; Amirtharajan and Balaguru, 2009; Amirtharajan et al., 2011), audio (Zhu et al., 2011), video (Al-Frajat et al., 2010) and text files (Al-Azawi and Fadhil, 2010; Yang et al., 2011; Shirali-Shahreza and Shirali-Shahreza, 2008).

Cryptography another popular technique for information security involves scrambling of data in an unintended format which would seem gibberish to any unintended user (Schneier, 2007). No extraction can be done unless the third party knows the secret code. However, it would invite tampering just because of the existence of a secret message. That’s where steganography has gained its impetus, with its basic constituents being a cover image, secret data and a key (Stefan and Fabin, 2000; Petitcolas et al., 1999; Rabah, 2004). The basic differences between cryptography and steganography is given by Zaidan et al. (2010) and Cheddad et al. (2010). When combined it gives out a stego image which has high un-detectability, is robust and also has high capacity (Hmood et al., 2010b) fulfilling requirements of the “Fridrich’s magic Triangle” in information hiding (Stefan and Fabin, 2000; Padmaa et al., 2011). Any files such as audio, video or image can be used as cover (Bender et al., 1996; Rabah, 2004), cover being the carrier without secret data in it, after embedding secret data in the cover; it is camouflaged with high level of imperceptibility and dexterity to yield the stego image. Here capacity or payload would mean the total amount of secret information that can be hidden in the stego image, without any physical notice-ability in the characteristics of the cover image while robustness defines the limit of standing modifications before an adversary can destroy the stego image. Within steganography, there are various techniques or methods to embed the data in the cover image i.e., Substitution based (Amirtharajan and Balaguru, 2009; Chan and Cheng, 2004), Transform domain based (Thanikaiselvan et al., 2011a), Spread Spectrum based (Amirtharajan and Balaguru, 2011; Thenmozhi et al., 2011; Kumar et al., 2011), Statistics based (Qin et al., 2009; Zanganeh and Ibrahim, 2011), Distortion and cover generation based (Xiang et al., 2011; Stefan and Fabin, 2000).

The most prominent being Spatial Domain based Steganographic Techniques and Transform Domain based Steganographic Techniques (Thanikaiselvan et al., 2011a; Kumar et al., 2011), Spatial domain based steganography include the Least Significant Bit (LSB) technique (Amirtharajan and Balaguru, 2009; Chan and Cheng, 2004), Pixel value differencing (Amirtharajan et al., 2010; Thanikaiselvan et al., 2011b; Padmaa et al., 2011) and more while the latter includes DCT (Provos and Honeyman, 2003), DWT and especially IWT (Thanikaiselvan et al., 2011a; EI Safy et al., 2009). A detailed survey on steganography till 1999 is available in Petitcolas et al. (1999) whereas a detailed survey on Information hiding in images, differences among cryptography, Stegano graphy (Zaidan et al., 2010) and water marking is available Cheddad et al. (2010). The counter attack called steganalysis is described by Qin et al. (2009) and Xia et al. (2009) and various steganalysis review is detailed by Qin et al. (2010).

A steganography technique is considered to be dependable when it can retain the embedded data in spite of severe modifications done to it and not displaying the payload contained in it. The most employed technique for this involves transforming the cover image into another domain and embedding data in the transformed pixels (Thanikaiselvan et al., 2011a; EI Safy et al., 2009). Before transformation the cover image exists in spatial domain, which is transformed into the frequency or time domain for the coefficients and after embedding data in the transformed pixels, it is brought back to the spatial domain. Thus, the underlying idea is, even if image is subjected to modifications and is in worst cases transformed, the data will still be hidden in the transformed pixels (Thanikaiselvan et al., 2011a).

Transformation techniques include DCT, DWT and IWT, though DCT isn’t highly preferred as it has low hiding capacity (Provos and Honeyman, 2003). Contemporary researchers use DWT, since it can be employed in compression of formats JPEG2000 and MPEG4. Techniques that use DWT i.e., wavelet transform based stego technique provides high capacity with the secret message embedded into the high frequency and low frequency coefficients of the wavelet transform, but provides less PSNR at a high hiding rate.

Cryptography in coalesce with steganography either through random walk (Luo et al., 2008) or variable bit optimal embedding (Zanganeh and Ibrahim, 2011) could be an effective solution to improve the complexity. In this study we propose a new modified version of the methodology by Thanikaiselvan et al. (2011a) which can embed a larger amount of data in Integer Wavelet Transform (IWT) domain with high PSNR while combining with the inverted pattern approach (Yang, 2008; Amirtharajan and Rayappan, 2012) to improve the complexity.

Related works: The use of Wavelet transform will mainly address the capacity and robustness of the information- hiding system features. The Haar Wavelet Transform is the simplest of all wavelet transform. In this the low frequency wavelet coefficients are generated by averaging the two pixel values and high frequency coefficients that are generated by taking half of the difference of the same two pixels (EI Safy et al., 2009). The four bands obtained are LL, LH, HL and HH which is shown in Fig. 1. The LL band is called as approximation band which consists of low frequency wavelet coefficients and contains significant part of the spatial domain image. The other bands are called as detail bands which consist of high frequency coefficients and contain the edge details of the spatial domain image.

Integer wavelet transform: Integer wavelet transform can be obtained through lifting scheme. Lifting scheme is a technique to convert DWT coefficients to Integer coefficients without losing information.

Forward Lifting scheme in IWT:

Step 1: Column wise processing to get H and L:

H = (Co-Ce)
(1)

L = (Ce-[H/2])
(2)

Fig. 1: Image and its transform domain bands

Where, Co and Ce is the odd column and even column wise pixel values

Step 2: Row wise processing to get LL, LH, HL and HH, Separate odd and even rows of H and L, Namely

LH = Lodd-Leven
(3)

LL = Leven-⌊LH/2⌋
(4)

HL = HoddHeven
(5)

HH = Heven-⌊HL/2⌋
(6)

Where:
Hodd = Odd row of H
Lodd = Odd row of L
Heven = Even row of H
Leven = Even row of L

Reverse lifting scheme in IWT: Inverse Integer wavelet transform is formed by Reverse lifting scheme. Procedure is similar to the forward lifting scheme (Thanikaiselvan et al., 2011a; EI Safy et al., 2009).

LSB Embedding: Simple LSB embedding is detailed by Amirtharajan and Balaguru (2009), Chan and Cheng (2004) and Janakiraman et al. (2012). Consider a 8-bit gray scale image matrix consisting mxn pixels and a secret message consisting of k bits. The first bit of message is embedded into the LSB of the first pixel and the second bit of message is embedded into the second pixel and so on. The resultant stego-image which holds the secret message is also a 8-bit gray scale image and difference between the cover image and the stego-image is not visually perceptible. This can be further extended and any number of LSB’s can be modified in a pixel. The quality of the image, however degrades with the increase in number of LSB’s. Usually up to 4 LSB’s can be modified without significant degradation in the message. Mathematically, the pixel value ‘P’ of the chosen pixel for storing the k-bit message Mk is modified to form the stego-pixel ‘Ps’ as follows:

Ps = P-mod (P,2k)+Mk
(7)

The embedded message bits can be recovered by:

Mk = mod (Ps,2k)
(8)

One method to improve the quality of the LSB substitution is Optimal Pixel adjustment Process (OPAP) (Chan and Cheng, 2004).

Determination of embedding:

Direct Binary embedding of data is done and the MSE (Mean Square Error) calculated
Let it assumed to be X
Inverted Binary embedding of data is done and the MSE (Mean Square Error) calculated
Let it assumed to be Y
Depending upon values of X and Y, If X > Y
Then the value is determined to be ‘1’ and if Y > X
Then the value is determined to be ‘0’

PROPOSED METHODOLOGY

The proposed block diagram of high capacity steganography system is given in Fig. 2 and 3. Preprocessing includes R, G and B plane separation and Histogram modification. Then Integer wavelet transform is applied to the cover image to get wavelet coefficients. Wavelet coefficients are randomly selected by using key-2 for embedding the secret data. Key -2 is 8x8 binary matrix in which ‘1’ represents data embedded in the corresponding wavelet coefficients and ‘0’ represents no data present in the wavelet coefficients. Then the direct binary or inverted binary embedding of data is done in the respective co-efficient. Key-1(K1) is a decimal number varying from 1 to 4 and it will decide the number of bits to be embedded in the cover object. High capacity is achieved by varying the key-1 (K1) value.

Embedding Algorithm
Step 1: The cover image of size 256x256x3 pixels is selected
Step 2: The respective planes are separated into R, G and B constituents
Step 3: The required data file to be embedded is taken with each character taking 8 bits
Step 4: Histogram modification is done in all planes, Because, the secret data is to be embedded in all the planes, while embedding integer wavelet coefficients produce stego-image pixel values greater than 255 or lesser than 0. Then all the pixel values will be ranged from 15 to 240
Step 5: Each plane is divided into 8x8 blocks
Step 6: Apply Haar Integer wavelet transform to 8x8 blocks of all the planes, this process results in LL1, LH1, HL1 and HH1 sub bands
Step 7: Using Key-1(K1) calculate the Bit length (BL) for corresponding wavelet co-efficients (WC), here we used modified version of Bit length calculation used by Thanikaiselvan et al. (2011a). Using the following equation, we get the high capacity steganography

(9)

Step 8: Using key-2 select the position and coefficients for embedding the ‘BL’ length data using LSB substitution. Here data is embedded only in LH1, HL1and HH1 subbands. Data is not embedded in LL1 because they are highly sensitive and also to maintain good visual quality after embedding data. An example of key-2 is shown below (This is Key B)

(10)

Step 10: After doing required operations with K-1 and K-2, the direct binary embedding of required data is done in the determine position and co-efficient and MSE calculated
Step 11: Next inverted binary embedding of the same data is done in the same position and corresponding MSE calculated
Step 12: Depending upon MSE values obtained Key-3 (K-3) is obtained with value ‘1‘ when inverted binary is embedded and ‘0‘ when direct binary is embedded. Thus K-3 is also an 8*8 matrix consisting of 1s and 0s representing whether direct or inverted binary embedding of data is done
Step 13: Applying Optimal Pixel adjustment Procedure (OPAP) reduces the error caused by the LSB substitution method
Step 14: Take inverse wavelet transform to each 8x8 block and combine R,G&B plane to produce stego image

Extraction Algorithm
Step 1: The corresponding stego image of 256*256 pixels is selected
Step 2: The respective planes are separated into R, G and B constituents
Step 3: Each plane is divided into 8x8 blocks
Step 4: Apply Haar Integer wavelet transform to 8x8 blocks of all the planes, This process results LL1,LH1, HL1 and HH1 sub-bands
Step 5: Using Key-1 calculate the Bit length(BL) for corresponding wavelet co-efficients(WC), using the ‘BL’ equation used in Embedding procedure
Step 6: Using key-2 select the position and coefficients for extracting the ‘BL’ length data
Step 7: Then using key-3, determine whether direct or inverted binary of the data has been embedded and extract it then depending upon it
Step 8: Combine all the bits and divide it in to 8 bits to get the text message

Fig. 2: Block diagram for Embedding

Fig. 3: Block diagram for extraction

ERROR METRICS

A performance measure in the stego image is measured by means of two parameters namely, Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR). The MSE is calculated by using the equation:

(11)

where, M and N denote the total number of pixels in the horizontal and the vertical dimensions of the image Xi, j represents the pixels in the original image and Yi, j, represents the pixels of the stego-image. The Peak Signal to Noise Ratio (PSNR) is expressed as:

(12)

RESULTS AND DISCUSSION

In this present implementation, Lena and baboon 256x256x3 color digital images have been taken as cover images, as shown in Fig. 4a and 5a, Stego lena (Thanikaiselvan et al., 2011a) in Fig. 4b and 5b and Proposed Stego images in Fig. 4c and 5c and tested with key-1(key-J) and various key-2s.

Fig. 4(a-c): (a) Cover Images Lena; (b) Stego lena (Thanikaiselvan et al., 2011a); (c) Proposed Stego lena

The effectiveness of the stego process proposed has been studied by calculating MSE and PSNR for the Lena and Baboon digital image in RGB planes and tabulated.

In this analysis, Key-2 is used for random selection of coefficients for embedding data (in this analysis Key-1 has been set as K1 = 1) and the results are tabulated in Table 1 for various Key-2 using the proposed method.

Fig. 5:(a-c): (a) Cover Images Baboon; (b) Stego Baboon (Thanikaiselvan et al., 2011a); (c) Proposed Stego Baboon

Table 1: Comparative Analysis for the proposed method key - 1(K1) = 1

It’s evident that from Table 1, Key-J provides high capacity and Key-A provides low capacity, moreover proposed methodology gives High PSNR over previous method (Thanikaiselvan et al., 2011a). By keeping Key J constant various key 1 are tabulated in Table 2.

Table 2: Key J constant by varying Key 1 for Lena and Baboon
Method* -Thanikaiselvan et al. (2011a)

Complexity level estimation: For 8x8 pixels block case IWT, total number of blocks (N) = 1024 Number of PRNG output for randomizing the 1024 blocks:

= NpR = N!/(N-r)! = 1024p1024 = 1024!

As per DES, the complexity of each block = 2^64.

Either binary or inverted binary would be embedded hence complexity increases by 2 and 3 out 4 sub bands are used and based on key 2 either 0 or 1 then 0.5.
Therefore the total complexity:

= (2^64)*(1024)! * (2) *3/4 *(.5)

These embedding are carried out in transform domain and the security level estimation reveals the firmness of the proposed stego against hackers.

CONCLUSION

It has been observed that steganography provides excellent avenue for high payload combined with imperceptibility. Literature suggests that in several techniques robustness may have been compromised, however the proposed method gives high payload (capacity) in the cover image with very little error. The algorithm is robust, as there is an option of embedding data in the transformed domain and also the option of reducing the error by determining whether direct or inverted binary maybe embedded. Key-1 and Key-2 not only provide high security but also increased capacity with the wavelet transform. The drawback of the proposed method is the computational overhead. This can be reduced by high speed computers. Thus, it can be summarized as:

PSNR is increased in this system with intelligent use of Key-1, Key-2 and Key-3
Because data is embedded in imperceptible areas that too in the transformed domain, the stego image can hardly raise suspicion
Even if the stego image is transformed it would be difficult to determine what data has been embedded without Key-3 as either direct or inverted binary would be embedded. Thus capacity, imperceptibility and robustness, all the three requirements are catered to in this innovative algorithm
REFERENCES
Al-Azawi, A.F. and M.A. Fadhil, 2010. Arabic text steganography using kashida extensions with huffman code. J. Applied Sci., 10: 436-439.
CrossRef  |  Direct Link  |  

Al-Frajat, A.K., H.A. Jalab, Z.M. Kasirun, A.A. Zaidan and B.B. Zaidan, 2010. Hiding data in video file: An overview. J. Applied Sci., 10: 1644-1649.
CrossRef  |  Direct Link  |  

Amirtharajan, R. and J.B.B. Rayappan, 2012. An intelligent chaotic embedding approach to enhance stego-image quality. Inform. Sci., 193: 115-124.
CrossRef  |  Direct Link  |  

Amirtharajan, R. and R.J.B. Balaguru, 2009. Tri-layer stego for enhanced security-a keyless random approach. Proceedings of the IEEE International Conference on Internet Multimedia Services Architecture and Applications, December 9-11, 2009, Bangalore, India, pp: 1-6.

Amirtharajan, R. and R.J.B. Balaguru, 2011. Covered CDMA multi-user writing on spatially divided image. Proceedings of the 2nd International Conference on Vehicular Technology, Information Theory and Aerospace and Electronic Systems Technology, February 28-March 3, 2011, Chennai, India, pp: 1-5.

Amirtharajan, R. and R.J.B. Balaguru, 2011. Data embedding system. WIPO Patent Application WO/2011/114196. http://v3.espacenet.com/textdoc?DB=EPODOC&IDX=WO2011114196&F=0

Amirtharajan, R., D. Adharsh, V. Vignesh and R.J.B. Balaguru, 2010. PVD blend with pixel indicator-OPAP composite for high fidelity steganography. Int. J. Comput. Applic., 7: 31-37.
CrossRef  |  Direct Link  |  

Amirtharajan, R., R.R. Subrahmanyam, P.J.S. Prabhakar, R. Kavitha and J.B.B. Rayappan, 2011. MSB over hides LSB: A dark communication with integrity. Proceedings of the IEEE 5th International Conference on Internet Multimedia Systems Architecture and Application, December 12-14, 2011, Bangalore, Karnataka, India, pp: 1-6.

Bender, W., D. Gruhl, N. Morimoto and A. Lu, 1996. Techniques for data hiding. IBM Syst. J., 35: 313-336.
CrossRef  |  Direct Link  |  

Chan, C.K. and L.M. Cheng, 2004. Hiding data in images by simple LSB substitution. Pattern Recognit., 37: 469-474.
CrossRef  |  Direct Link  |  

Cheddad, A., J. Condell, K. Curran and P. Mc Kevitt, 2010. Digital image steganography: Survey and analysis of current methods. Signal Process., 90: 727-752.
CrossRef  |  Direct Link  |  

El-Safy, R.O., H.H. Zayed and A. EI-Dessouki, 2009. An adaptive steganographic technique based on integer wavelet transform. Proceedings of the International Conference on Networking and Media Convergence, March 24-25, 2009, Cairo, Egypt, pp: 111-117.

Hmood, A.K., B.B. Zaidan, A.A. Zaidan and H.A. Jalab, 2010. An overview on hiding information technique in images. J. Applied Sci., 10: 2094-2100.
CrossRef  |  Direct Link  |  

Hmood, A.K., H.A. Jalab, Z.M. Kasirun, B.B. Zaidan and A.A. Zaidan, 2010. On the capacity and security of steganography approaches: An overview. J. Applied Sci., 10: 1825-1833.
CrossRef  |  Direct Link  |  

Janakiraman, S., R. Amirtharajan, K. Thenmozhi and J.B.B. Rayappan, 2012. Pixel forefinger for gray in color: A layer by layer stego. Inform. Technol. J., 11: 9-19.
CrossRef  |  Direct Link  |  

Kumar, P.P., R. Amirtharajan, K. Thenmozhi and J.B.B. Rayappan, 2011. Steg-OFDM blend for highly secure multi-user communication. Proceedings of the 2nd International Conference on Vehicular Technology, Information Theory and Aerospace and Electronic Systems Technology, February 28-March 3, 2011, Chennai, India, pp: 1-5.

Luo, G., X. Sun and L. Xiang, 2008. Multi-blogs steganographic algorithm based on directed hamiltonian path selection. Inform. Technol. J., 7: 450-457.
CrossRef  |  Direct Link  |  

Padmaa, M., Y. Venkataramani and R. Amirtharajan, 2011. Stego on 2n: 1 Platform for users and embedding. Inform. Technol. J., 10: 1896-1907.
CrossRef  |  Direct Link  |  

Petitcolas, F.A.P., R.J. Anderson and M.G. Kuhn, 1999. Information hiding-a survey. Proc. IEEE, 87: 1062-1078.
CrossRef  |  Direct Link  |  

Provos, N. and P. Honeyman, 2003. Hide and seek: An introduction to steganography. IEEE Secur. Privacy, 1: 32-44.
CrossRef  |  

Qin, J., X. Sun, X. Xiang and Z. Xia, 2009. Steganalysis based on difference statistics for LSB matching steganography. Inform. Technol. J., 8: 1281-1286.
CrossRef  |  Direct Link  |  

Qin, J., X. Xiang and M.X. Wang, 2010. A review on detection of LSB matching steganography. Inform. Technol. J., 9: 1725-1738.
CrossRef  |  Direct Link  |  

Rabah, K., 2004. Steganography-the art of hiding data. Inform. Technol. J., 3: 245-269.
CrossRef  |  Direct Link  |  

Schneier, B., 2007. Applied Cryptography: Protocols, Algorithm and Source Code in C. 2nd Edn., John Wiley and Sons, New Delhi, India.

Shirali-Shahreza, M. and S. Shirali-Shahreza, 2008. High capacity persian/arabic text steganography. J. Applied Sci., 8: 4173-4179.
CrossRef  |  Direct Link  |  

Stefan, K. and A. Fabin, 2000. Information Hiding Techniques for Steganography and Digital Watermarking. Artech House, London, UK.

Thanikaiselvan, V., P. Arulmozhivarman, R. Amirtharajan and J.B.B. Rayappan, 2011. Wave (let) decide choosy pixel embedding for stego. Proceedings of the International Conference on Computer, Communication and Electrical Technology, March 18-19, 2011, India, pp: 157-162.

Thanikaiselvan, V., S. Kumar, N. Neelima and R. Amirtharajan, 2011. Data battle on the digital field between horse cavalry and interlopers. J. Theor. Applied Inform. Technol., 29: 85-91.
Direct Link  |  

Thenmozhi, K., P. Praveenkumar, R. Amirtharajan, V. Prithiviraj, R. Varadarajan and J.B.B. Rayappan, 2012. OFDM+CDMA+Stego = Secure communication: A review. Res. J. Inform. Technol., 4: 31-46.
CrossRef  |  Direct Link  |  

Xia, Z., X. Sun, J. Qin and C. Niu, 2009. Feature selection for image steganalysis using hybrid genetic algorithm. Inform. Technol. J., 8: 811-820.
CrossRef  |  Direct Link  |  

Xiang, L., X. Sun, Y. Liu and H. Yang, 2011. A secure steganographic method via multiple choice questions. Inform. Technol. J., 10: 992-1000.
CrossRef  |  Direct Link  |  

Yang, B., X. Sun, L. Xiang, Z. Ruan and R. Wu, 2011. Steganography in Ms Excel document using text-rotation technique. Inform. Technol. J., 10: 889-893.
CrossRef  |  Direct Link  |  

Yang, C.H., 2008. Inverted pattern approach to improve image quality of information hiding by LSB substitution. Pattern Recognit., 41: 2674-2683.
CrossRef  |  

Zaidan, B.B., A.A. Zaidan, A.K. Al-Frajat and H.A. Jalab, 2010. On the differences between hiding information and cryptography techniques: An overview. J. Applied Sci., 10: 1650-1655.
CrossRef  |  Direct Link  |  

Zanganeh, O. and S. Ibrahim, 2011. Adaptive image steganography based on optimal embedding and robust against chi-square attack. Inform. Technol. J., 10: 1285-1294.
CrossRef  |  Direct Link  |  

Zhu, J., R.D. Wang, J. Li and D.Q. Yan, 2011. A huffman coding section-based steganography for AAC audio. Inform. Technol. J., 10: 1983-1988.
CrossRef  |  Direct Link  |  

©  2014 Science Alert. All Rights Reserved
Fulltext PDF References Abstract