Вы находитесь на странице: 1из 8

AMLAN CHAKRABORTY 0440954 amlan@cs.washington.

edu Steganography and Digital Watermarking -- Applications, Attacks and Countermeasu res Introduction Steganography is the science of hiding information in data. Normally steganograp hy is done intelligently such that it is difficult for an adversary to detect th e existence of a hidden message in the otherwise innocuous data. The piece of da ta that has the message embedded in it is visible to the world in the clear and appears as harmless and normal. This is in stark contrast with cryptography wher e the message is scrambled to make it extremely difficult or impossible for an adversary to put together. A message in ciphertext arouses some sort of suspicio n whereas invisible message embedded in clear text does not. This is the advant age of steganography. Generally, a steganographic message will appear to be something else: a picture, an audio file, a video file or a message in clear text - the covertext. Histor ically, messages were written using hidden invisible ink between the visible lin es of innocuous documents, or even written onto clothing. Other techniques used were writing messages in Morse code in knitting yarn, or marking particular word s or letters in the message, using invisible ink or pin prick that form the secr et message. During WWII Germans used the microdot technology, where an image the size of a period had the clarity of typewritten pages. In this case the period was the covertext and the image is the message. Though smart hiding and innocuou s hiding techniques are used to hide the stegotext, the algorithm itself is secu re and only known to the communicating parties and not to the world. This is in slight contrast to classical cryptography where the algorithm is well known and only the key(s) are secret. Though data is not encrypted in steganography, authe nticity of a message is normally established by using a MAC or a signature. Steganography can be used to code messages in any transport layer an image (GIF/ BMP/JPEG), a MP3 file, a communications protocol like UDP etc. Steganogrpahic in formation can also be added to richer multimedia content like DVDs. There are no rmally two motivations to send a secret message or to establish authenticity of a piece of information usually a multimedia file. The later is a major applicati on of modern steganography and known as Digital Watermarking and Fingerprinting. Watermarks establish ownership of an artifact while fingerprints or labels help to identify intellectual property violators. They are different protocol implem entation of the same basic idea. Information theory and human sensory perception Steganography is possible for the same reasons that compression is a combination of information theory and human perception of vision and audio. Digital signal contains redundancy which manifests itself as noise. Humans cannot detect all le vels of noise; in other words, humans often cannot tell an image or an audio cli p from another with slight difference in levels of noise. The larger the cover message is (in data content terms number of bits) relative to the hidden message , the easier it is to hide the latter. For example, a 24 bit bitmap image has 8 bits representing three colors Red, green and blue at each pixel 256 shades of e ach basic color. So changing the least significant bit of any of these basic col

ors would make an extremely negligible change on that pixel and possibly less on the image. So the least significant bit can be easily used to store the stegano graphic message. So, if we change the LSB of each basic color of three adjacent pixels, we get 9 bits -- enough space to store an ASCII character. This is calle d LSB manipulation and a very conventional and simple steganographic implementat ion. It can also be noted that the actual message itself can be compressed using some compression coding methodologies like run length coding. Historically, a l ot of invisible ink steganographic messages were encoded using Polybius squares or similar text to integer mapping schemes. Stated somewhat more formally, the objective for making steganographic encoding difficult to detect is to ensure that the changes to the carrier/container (the original signal) due to the injection of the payload (the signal to covertly emb ed) are visually and ideally, statistically negligible; that is to say, the chan ges are indistinguishable from the Gaussian noise of the carrier. From an inform ation theoretical point of view, this means that the channel must have more capa city than the 'surface' signal requires (entropy), i.e., there is redundancy. Fo r a digital image, this may be noise from the imaging element; for digital audio , it may be noise from recording techniques amplitude or frequency modulation. A ny system with an analog (signal) amplification stage will also introduce therm al noise, which can be exploited as a noise cover. Steganographic channel is a covert channel in Information theory terms since it transfers some kind of infor mation using a method originally not intended to transfer this kind of informati on. Steganography also supports both storage and timing covert channels. This re port primarily discusses storage covert channels where a covert message is commu nicated by manipulating a stored object like an image. Ron Rivests Chaffing and Wi nnowing protocol discussed later can be argued as an example of timing covert cha nnel. It is fairly obvious that more the data content of the cover message, the easier it is to hide the message. In case of images, bitmaps are better fits that GIFs and JPEGs because GIF is 8 bits per pixel and JPEG is a lossy compression techn ique. But on the flipside, bigger images will attract more attention than small er images as suspect stego-images. Subtlety in changes is a very important featu re and stego-images should only have subtle changes. An image with large areas o f solid colors would be a bad fit since large variances created by the embedded message would cause drastic differences easily spotted by the human eye. The spa tial frequency distribution of the image (spatio- temporal in case of audio or v ideo content) is also a determining factor in the efficiency of the hiding proce ss. As we will see later, we have techniques for both Gaussian and LaPlacian dis tribution using maximum likelihood estimators for the stego-messages. Often the embedded message is itself encrypted using a key that may or may not b e known to the adversary. Since steganography requires that communicating partie s have some prior shared information, symmetric key is a natural fit. However, p ublic steganography with steganographic key exchanges is also possible. Prisoners problem and subliminal channel The study of steganography in machine cryptography was first stated in the priso ners problem by Simmons. Two inmates Alice and Bob are accomplices in a crime and are sent to the prison. They need to communicate with each other but they have to use a public channel which is monitored by the Warden of the jail. The warden will only forward the messages if they are intelligible. The prisoners accept t his condition and find a way to communicate secretly in exchanges --- establishi ng a subliminal channel even though the messages themselves are not encrypted. T he warden will also try to deceive them, so they will authenticate each others me ssages before accepting them authentication without secrecy. Thethe situation is paradoxical because the warden demands access and the prison ers need to authenticate each other. Authentication without secrecy channels ach ieve that by placing a pre arranged condition on all messages. It is this capab

ility that creates a subliminal channel for the prisoners. If m redundant bits are allowed to establish authenticity, then these redundant bits create a bit by bi t subliminal channel which can be used to transmit extra information. Null ciphers and Chaffing and Winnowing A null cipher is a form of encryption where the plaintext is mixed with a large amount of non-cipher material. Null ciphers are used to hide the actual cipherte xt by introducing nulls to confuse the cryptanalyst. Classical steganography can also be thought of as an extension of this concept where the carrier / containe r data are actually the null ciphers data that create confusion and diffuse the actual payload. Ron Rivest extended this concept to an idea of Chaffing and Winnowing to create st eganographic communication channels. The concept is analogous to separating (win nowing) wheat from chaff where wheat is the actual payload and chaff is the null ciphers. In a two step process, the transmitter introduces chaff to the wheat i .e. intersperse the actual payload with meaningless data. The receiver winnows the actual payload from the non-interesting data. As with most steganographic trans fers, the transmitters add a MAC to establish authenticity of the communication to any message that is sent. MACs are calculated over the entire message and a s erial number of the message using a secret symmetric authentication key. The tra nsmitter attaches bogus MACs for the chaff packets instead of calculating it. Th is is what distinguishes the chaff from the wheat. The receiver now doesnt have to do anything special since the normal protocol of a receiver is to discard packets that do not have correct MACs. Though the adversary can see the entire communica tion, it cannot tell chaff from wheat as the MAC will look like a random functio n. However, weak MAC functions can potentially leak information in this protocol . It is also important to note that it is not possible to use digital signatures here since anyone will be then able to compare the signatures and tell chaff from wheat. However, designated verifier signature schemes where only signature designat es can verify a signature would work fine. The other key idea is that since the creation of chaff involves generation of a bad MAC and not the knowledge of a secr et key, any entity can play the role of a chaffer. Digital Watermarking and Fingerprinting Digital watermarking is the technique of adding identifying information to digit al artifacts using steganographic principles i.e. hiding the information cleverl y so that extraction is difficult by any adversary. Watermarks can be visible or invisible in the context of images. There are various techniques of placing dig ital watermarks on images but they can conceptually be divided into two categori es 1. Spatial techniques. These methods are based on hiding the messages on ge ometric characteristics of the image. These are highly susceptible to signal alt eration algorithms. Even simple signal manipulation like zooming, cropping, smoo thing would obliterate watermarks. 2. Frequency Domain techniques. These methods are used to hide messages alo ng the frequency distribution of hues, intensities, luminance etc of the images. These are comparatively robust to simple image manipulations but can fall prey to statistical steganalysis. In strict terms, visible digital watermarks are really not steganographic object they enhance information instead of hiding. Fingerprinting is a slight different implementation of digital watermarks. When an artifact is sold to an entity, information about that entity is hidden in the artifact. If illegitimate copies of the artifact are sold, the watermark inform ation would reveal the violator. A slight modification of this would be using th e canary trap protocol where unique alterations are made to each copy of artifac t sold. The illegitimate copy has a tell-a-tale that traces back to the violator .

Some digital watermarking algorithms It is not too difficult to formulate algorithms that can cleverly hide informati on in images. The key idea to avoid detection is to hide the message in such a w ay that statistically it comes across like normal distribution making pattern de tection very difficult. Masking and filtering These are some basic techniques to create visible watermarks by altering the lum inance or colors of certain regions in the image. These can be detected very eas ily by simple statistical analysis but these are fairly resistant to lossy compr ession and image cropping. It doesnt hide the data in noise but embed it in signi ficant areas just the reverse of LSB manipulation. LSB Manipulation This is the manipulation described in the Introduction that is susceptible to ev en slight image modification. It is very efficient in hiding a GIF or BMP image in another but a linear analysis is enough to figure this out. It is fairly easy to hide an image in 3 or even 4 least significant bits of another image without causing major noticeable change. The motivation for steganography is important here. If the intention is to covertly pass messages, this can still work unless all artifacts are sniffed for steganographic information. But if this is meant f or digital watermarks, it is very easy to extract and /or get rid of the info. Spread Spectrum methods In spread spectrum methods, the message is scattered across the image making it harder for cropping, rotation and other basic image manipulation techniques to o bliterate the watermark. This is also somewhat resistant to statistical steganal ysis because it gives it the impression of noise in an image. Patchwork is a too l from IBM uses this technique to scatter hidden information based on statistica l distribution of luminance in the image. It iteratively selects two patches on the image, brightens one and darkens one. It then calculates the standard deviat ion, S between light and dark patches over the sample patches. To encode, it pic ks up two patches up in random and then brightens one by S and darkens one by S. This process is iterated and the whole image palette is laid in a mosaic of bri ght and dark patches one of which is used to hide data. This patch information i s vital to decode the hidden message later. This is clearly a frequency distribu tion method. Patchwork makes the assumption that the image has a Gaussian distri bution. Texture Block coding In this method, pairs of areas of similar texture are found and one area is copi ed over the other. Thus we have identical blocks of texture in the image. Iterat ing a few times, we can get two large blocks of identical textures. These two bl ocks would get altered identically for all non-geometric alterations of the imag e. These two blocks can then contain information about these images. M-Sequences using linear shift registers M-sequences are based on starting vectors of a Fibonacci recursion relation whic h form a Galois field of finite cardinality. Mathematically and statistically th ese numbers are known to have desirable autocorrelation functions; the distribut ion of Galois field numbers is known to be of normal distribution thus resemblin g Gaussian noise in an image. So images encoded using m-sequences are statistica lly impossible to distinguish from the original as they are similar to noise in a normal distribution. If the stego message is encoded using m-sequences, it can easily be embedded in the image by a LSB substitution. A more secure implement ation would be to use LSB addition instead to embed the watermark. So it will re quire the examination of the complete bit pattern and the current linear shift r egister implementation. This is more secure because to crack this, the adversary would have to do the same computations without any apriori knowledge.

Frequency hopping In this method scattering of the message is done on the basis of rules that chan ge cumulatively. The idea is similar to DES block encryption; bits are swapped a ccording to rules that are dictated by the stego-key and random data from the pr evious round. White noise storm, an implementation of this methodology, creates a message space of 8 channels where each channel has a window of W bytes, where W is a random number. Each channel however carry only one bit of the message and a lot of unused bits. The bits inside a window permutate and rotate according t o an algorithm that is regulated by the previous windows operations and the stego -key. Finally this encoded message is embedded in the image using LSB substitut ion. The idea again is to simulate a distribution that is similar to a Gaussian distribution.

Steganalysis and Digital Watermarking Attacks Steganalysis is analogous to cryptanalysis in the context of steganography. Ste ganalysis is composed of three steps:1. Detection of hidden message (Passive Steganalysis) 2. Extracting of hidden message (Active Steganalysis) 3. Disabling/ Destruction of hidden message. It is important to note here that it is not necessary to extract a message to di sable or destruct a message. It is often very difficult to extract a hidden mes sage and at times even to detect one because they are scattered and show up as n oise. The case of visible watermarks is obviously different. But the problem lie s in the fact detection is also not important if we have a suspicious attitude. We can run algorithms that are known to destruct digital watermarks in messages. On top of that there are algorithms that instead of disabling watermarks, either overwrite watermarks or create exact replicas rendering the watermark useless e ither way. Luis Von Ahn et al formulates and proposed universal robustness for ste ganographic information. They prove that robust steganography is as secure as the underlying crypto used to encrypt the message that is hidden in the clear. But t his just ensures extraction is hard and likens it to cryptography. But their alg orithm doesnt prove that obliteration of the steganographic secret is not possibl e. Some common types of attacks on Digital Watermarking are 1. Removal attacks Denoising, Remodulation, Lossy Compression These attacks attempt at completely removing watermark from the data. Since a lo t of steganography algorithms try to hide data as noise, removal of noise should obliterate the watermark. These algorithms they try to estimate the cover data using a given statistic for the noise in it. It assumes the noise to be the wate rmark. Langelaar et al proposes a sequence of filtering operations( median filt ering, highpass filtering) on the image to denoise the image that will likely g et rid of the digital watermark. There are several other watermark estimator alg orithms that uses either Maximum Aposteriori Probability (MAP) if we know the im age statistics or Maximum likelihood (ML) Classifier algorithms if we do not kno w anything about the images, to find an estimate of the digital watermark. Volos hoynovisky proposed an algorithm where he used the MAP estimator and then remodu lates the image to find the least favorable noise distribution. This is guessed to be the watermark. Often lossy compression of uncompressed image data like JPEG, would completely w ipe out the watermark since the raw data would be replaced by Direct Cosine Tran sforms of the data. However, this is mitigated by algorithms that can hide infor mation directly in compressed data. 2. Geometric Attacks Warping, transforming, jittering etc.

These attacks are the easiest to implement and often very effective. Instead of removing the watermark, these stress on distortion of embedded data by spatial o r temporal alterations (in case of audio and video data). The result of these at tacks is to scatter and alter the way the watermark is laid out in the image. Fo r a simple attack, if an image is rotated by a slight angle, say 1 degree and th e edges filled by the texture of the average of adjacent pixels, there is a high likelihood that the watermark would fall out of sync with the watermark detecto r. The key idea here is though the digital watermark data exists in the artifact , it has moved in such a way that the watermark detector can no longer detect th e data. Jittering is another effective attack that works extremely well for audio data. An audio signal is chunked up into n chunks and then either one chunk is deleted o r a copy is made and then assembled back together ending up in either (n -1) or (n +1) samples. This introduces a jitter in the signal that is not detectable by humans. Digital watermarks would totally get destroyed in this attack. Unzign implements a pixel jittering algorithm that works well on spatial domain waterma rks. Another important observation is that though some algorithms survive basic geome tric attacks like rotation, shearing, resizing etc., they succumb to a combinati on of different attacks. StirMark is an implementation based on these principles that simulates an iterative resampling process where the image is slightly resi zed, sheared and rotated by a random small amount. However, repeated iterations of StirMark degrade the image to the point that humans can detect the differenc e between the original and the processed. 3. Crypto attacks Exhaustive key search, Collusion, Averaging, Oracle attac k These are similar to normal cryptographic attacks where the steganographic key i s searched exhaustively. Statistical averaging attacks involve taking the same d ata set with different instances of watermarks and then averaging them to find t he attacked data set. A modification of the averaging algorithm is the collusion attack where smaller portions of the data set are taken and attacked data set f ound using averaging algorithms. These smaller datasets are then combined to get a new attacked data set. 4. Protocol attacks Watermark inversion, Copy Attack. These attacks do not aim to detect, destroy or disable the watermark, but to att ack the basic tenets of watermarking e.g. watermarks cannot be extracted from no n watermarked data. The Watermark inversion attack uses the feature of overmarking that is the abili ty to mark an image more than once. Bob gets an image from Alice that has her wa termark. Bob subsequently generates his own watermark and subtracts his watermar k from the image he got from Alice. Due to overmarking, Alices signature would st ill be readable from this image making it almost identical to the image Alice ci rculated. Bob can now argue that Alice has removed his signature and added hers to generate this image. This will establish that Bob was the actual owner of the image. The Copy Attack gets an estimate of the watermark using a MAP or a ML estimator. It then processes this watermark using the least favorable noise function (ment ioned in replacement attacks) to smoothen the watermark. It then adds the waterm ark to a new document. Copy attack allows anyone to identify his own document as being watermarked by a well known entity by placing a watermark copied from a d ocument published by that entity on it. This is a very serious attack that Kutte r et al experimentally succeeded to accomplish. Defenses against Steganalysis We noticed that most steganographic algorithms pretty cleverly hide data to avoi d detection by --- blending in as Gaussian noise, embedding in significant areas , scattering across the frequency spectrum etc. It has also been seen that it is often not easy to extract a digital watermark. Luis von Ahn et al propose robus

t steganographic algorithms as well as new advances to public key steganography etc. But this often doesnt safeguard against attacks to destroy or replace waterm arks on images, audio files etc. We have also notices that watermarks are partic ularly susceptible to attacks that are combination of more than one attack. Ther e have been mitigations suggested to particular types of attacks e.g. error cor rection of coding theory using hamming distance (or some other distance measurin g algorithm like Euclidean algorithm ) for statistical steganalysis. But the pr oblem is that attacks are preceding mitigations. Barr et al from DigiMarc are s uggesting the concept of image signature to mitigate the copy attack where perpe tually similar images would produce the same signature whereas perpetually diffe rent image would produce very different signatures. While this would successful ly mitigate the copy attack, one can still launch a geometric attack and obliter ate the watermark. Since the image would be perpetually similar the signature wo uld be the same, and image signature would not mitigate the attack. Additionally the problem here is that there is an additional burden on the watermark detecto r to verify the signature of the image. This is double verification and needs ad ditional security. The key idea is to make the digital watermark such that destruction of the water mark would destroy the image itself. One idea proposed by Neil Johnson is to us e a gradual mask instead of a sharp mask for the visible watermark, so that the watermark is not visible until the luminance of the image is significantly incre ased. This makes it more robust against changes of lower bits. Though extensive image processing and spatially selective alteration of luminance based on the lu minance distribution may make the digital watermark vulnerable, the image would also be distorted enough by so much processing. Conclusion The challenges in digital watermarking stem from the fact that the attacks deriv e from the same phenomenon as the watermarking technology itself -- small noise insertion doesnt create humanly noticeable changes to an artifact. Clearly right now the watermarking technology is not robust enough to mitigate combination of attacks. Introduction of new authentication schemes as proposed by public key s teganography would attach another layer of security but does not in itself guara ntee universal absolute robustness of watermarks. I think the solution may very well lie in better statistical models based on information theory. We can mitiga te some attacks using authentication and authorization but pattern detection and obfuscation should be mitigated by better scattering algorithms. References 1. Neil Johnson and Sushil Jajodia, Exploring Steganography: Seeing the Uns een. 2. Gustavus J Simmons, The Prisoners Problem and the Sublimimal channel 3. Ronald L. Rivest, Chaffing and Winnowing: Confidentiality without Encryp tion 4. Fabien A Petitcolas, Ross J Anderson and Markus Kuhn, Information Hiding a Survey. 5. Pierre Moulin and Joseph O Sullivan, Information-Theoritic Analysis of in formation Hiding 6. Nicholas J Hopper, John Langford and Luis Von Ahn, Provably Secure Stega nography 7. Neil Johnson and Sushil Jajodia, Steganalysis : The investigation of hid den information 8. Bender, Gruhl, Morimoto and Lu, Techniques for data hiding. 9. Neil F Johnson, An Introduction to Watermark recovery from Images 10. Fabien A Petitcolas, Ross J Anderson and Markus Kuhn, Attacks on copyrig ht marking systems. 11. Martin Kutter and Sviatoslav Voloshynoviskiy, The Watermark Copy attack 12. Niels Provos, Defending against Statistical Steganlysis 13. Barr, Bradley and Hannigan, Using Digital watermarks to mitigate the thr

eat of copy attacks. 14. Karen Su, Deepa Kundur and Dmitrios Hatzinakoa, A novel approach to coll usion resistant Video watermarking. 15. Stefan Katzenbeiser and Helmut Beith, Securing symmetric watermarking sc hemes against protocol attacks.

Вам также может понравиться