外文翻譯---基于數(shù)字水印技術(shù)的信息隱藏理論研究

1、<p><b>　　外文資料</b></p><p>　　Information Hiding Thieoretic Research Based on Digital Watermark Technology</p><p>　　School of Mathematics and Computer Science Ningxia Universit Yi

2、nchuan, China</p><p>　　Abstract-With rapidly growing interest in ways to hide information, a large number of schemes have been proposed for audio and other multimedia in digital pictures, video, audio and ot

3、her multimedia objects.In an attempt to overcome this problem, watermarking has been suggested in the literrature as the most effective means for copyright protection and authentication .The main focus of this thesis is

4、the problem of watermarking and compression of image due to bandwidth or storage constraints.</p><p>　　Keywords: information hiding; digital watermarking; transparency; robustness</p><p>　　Intro

5、duction</p><p>　　Due to the rapidly development of the Interest and the World-Wide-Web(WWW),distribution of digital multimedia data to a large population of users can be done very easily.Moreover,digital dat

6、a can be duplicated very fast and without any degration in quality-cinsider,for example,how common the copying of musical CDs has become in the last few years.Naturally,this situation has raised many concerns about possi

7、ble violations of intellectual property rights.Unauthorized duplication and distribution o</p><p>　　There have been many instances of disputes or litigations on the intellectual ownership of multimedia data.

8、A copyright violations lawsuit that received extensive publicity in the early 2000's,was that against Napster.And the Digital Video Disk,also known as Digital Versatile Disk(DVD)consortium called for proposals for a

9、copyright marking scheme to enforce serial copy management.The idea is that the DVD players sold to consumers will allow unlimited copying of home videos and time-shifted viewin</p><p>　　Due to its significa

10、nce,the watermarking field has grown tremendously in the information hiding theoretic.There are numerous articles(the references therein)and books that explain the basic of watermarking,explore its many practical applica

11、tion,and evaluate the performance of various schemes under a variety of attacks.</p><p>　　Digital Watermarking System</p><p>　　Watermarking</p><p>　　In the communication model for w

12、atermarking,the watermark embedder plays the role of the transmitter,the watermark detector plays the role of reciever and the attack represents the communication channel(Fig.1).The objective is to decode the hidden mess

13、age reliably;the original image can be construed of as side information which is always available at the encoder,but is only available at the decoder in the private scenario(not the public one).</p><p>　　In

14、this paper, a transmitter sends a signal Xn to a receive (with n transmissions), through the following channel: Y"=Xn+Sn+Zn, where Sn, Zn are independent, i.i.d.Gaussian random vectors.Here, Sn is known to the encod

15、er, while Zn acts as noise known to neither encoder or decoder.It is proved that the capacity of this channel is the same as if Sn were known to the decoderas well.</p><p>　　In general,this number grows line

16、arly with the image size,and the constant of proportionality is known as the capacity C.Practical watermarking implementation embed at a rate R bits/image dimension,where R<C.Namely,C=1/2 log（1+P/N）,where P is the ave

17、rage（per symbol）power of Xn and N is average of Zn.This result is quite surprising because one would expect Sn to act as inteference at the receiver,thus hidering the detection scenario,Sn would play the role of the cove

18、rtext（known to the watermark e</p><p>　　Characteristics of watermarking</p><p>　　In the design of watermarking schemes,we mainly think about two key requirements.</p><p>　　Transpare

19、ncy: The hidden message should not interfere perceptually with the host signal.This requirement is perfectly justified by the fact that watermarking aims at protecting multimedia data,which are sensitive,in general,to c

20、hanges.In other words,an image or a musical piece could become useless if the introduced artifacts（due to watermarking）exceeded some perceptual threshold.The quality of the watermarked data must thus be comparable to tha

21、t of the covertext,a requirement which is often exp</p><p>　　Robustness: The message must be detectable in the watermarked image（the covertext is assumed to be an image throughout this thesis,though similar

22、 techniques can be applied to other types of multimedia data）,even after degradation duo to malicious attacks or other processing（quantization,D/A conversion,etc）.Of course,detectability of the watermark is closely relat

23、ed to the maxinum amount of distortion that can be introduced by an attacker.A watermarking scheme is robust if it allows the hidden m</p><p>　　Detection scenarios</p><p>　　There are two decetio

24、n scenarios:private and public.In the private decetion scenario,the original image is available to the decetor,in the public scenario,it is not.Although public detection scheme can be more useful in practice(since it is

25、not always possible to have the original image availableduring the decetion),private schemes usually offer more robustness.</p><p>　　Watermarking Rates In The Presence Of Memoryless Attacks And Gaussian Att

26、acks</p><p>　　We choose that the memoryless attack is fixed (in terms of the conditional probability distribution of the attack channel), for both discrete and continuous alphabets.We establised the region a

27、chievable rates (Rq, Rw) under memoryless attacks.</p><p>　　Discrete Alphabets.</p><p>　　The general form of the watermarking system under consideration is shown in Fig.2.The watermark index W i

28、s uniformly distributed over a set of size 2nRw; Yn is the watermarked image which can be found in a source codebook of size 2nRw, the attack channel is memoryless.</p><p>　　The watermark decode outputs W

29、9;.We consider a private scenario here,so we assume that In is known at the decoder.</p><p>　　A.Attacking Procedure</p><p>　　The watermarker (who designs the encoder pair) tries to maximize the

30、achievable rate region while the attacker tries to minimize it [12].The procedure of attack are expressed as follows:</p><p>　　1. The encoder (watermark embedder) is designed without any knowledge of the att

31、ack conditional distribution p.Firstly; the watermark designs an encoding function f. The watermark must ensure that the rates are chosen such that the watermark is detected reliably for any attackd diatribution p chosen

32、 by the attacker.</p><p>　　2. The attack, who knows k, plays second and chooses a conditional distribution p for the attack such that the distortion constraint is met.</p><p>　　3. The watermark

33、plays next, and designs his watermark decoding algorithm g with respect to the distribution p chosen by the attacker.</p><p>　　Conclution</p><p>　　In this paper, we established the region of ach

34、ievable rates under memory attacks.The paper is discussed the following case:</p><p>　　1. The memoryless attack is chosen independently of the embedding strategy and is known to both encoder and decoder.Resu

35、lts are obtained for two statistical models:the general discrete alphabet case for arbitrary image distributions and distortion constraints,as well as the Gaussian case where the original image and the attack channel are

36、 i.i.d.Gaussian,and the distortion metric is quadratic.</p><p>　　2. The information and the attacker play a game which is the attacking procedure. Specifically, the attacker knows the encoding function used

37、by the watermarker, while the decoder knows the attack distribution.</p><p>　　Finally, there are a number of direction for future research.All results derived in this thesis assume a private scenario.We were

38、 not able to astablised the region of achievable rates in a public scenario, even for the simple Gaussian case.Futhermore, it would be interesting to establish rate regions for joint watermarking and compression systems

39、under distortion constraints that do not involve averaging of distortion measures.</p><p><b>　　中文翻譯</b></p><p>　　基于數(shù)字水印技術(shù)的信息隱藏理論研究</p><p><b>　　宋麗娟 </b></

40、p><p>　?。▽幭拇髮W(xué) 數(shù)學(xué)與計(jì)算機(jī)科學(xué)，銀川 中國，slj@nxu.edu.cn）</p><p>　　摘要：隨著信息隱藏技術(shù)的迅速發(fā)展，水印和其它一些在數(shù)碼照片、視頻、音頻以及其它的多媒體設(shè)備等信息方面相繼也提出了很多方案。為了解決版權(quán)保護(hù)盒認(rèn)證的相關(guān)問題，有人提出了使用水印作為最有效的手段。鑒于帶寬的限制和壓縮存儲(chǔ)的問題，本文的重點(diǎn)是研究水印和壓縮方面的問題。</p>

41、<p>　　關(guān)鍵字：信息隱藏；數(shù)字水??；透明度；魯棒性</p><p><b>　　I. 引言</b></p><p>　　由于互聯(lián)網(wǎng)和計(jì)算機(jī)的迅速發(fā)展，數(shù)字多媒體迅速發(fā)展到數(shù)據(jù)同時(shí)共享于眾多用戶是很容易實(shí)現(xiàn)的。此外，數(shù)據(jù)可以被快速復(fù)試而不產(chǎn)生任何質(zhì)量損耗。例如，最近幾年里，音樂和唱片的拷貝變得非常普遍。理所當(dāng)然，這種情形喚起了人們對(duì)知識(shí)產(chǎn)權(quán)侵害的關(guān)注。沒有

42、給予版權(quán)持有人適當(dāng)?shù)难a(bǔ)償，而進(jìn)行未經(jīng)許可的復(fù)制以及傳播享有著作權(quán)的材料（照片，音樂，電影等），正變得越來越嚴(yán)重。為了打擊盜版，許多公司（尤其是娛樂和新聞行業(yè)）相當(dāng)重視信息隱藏（或數(shù)字水印）技術(shù)的發(fā)展。簡(jiǎn)單來說，水印作為一種信號(hào)隱藏技術(shù)在一些多媒體數(shù)據(jù)中，并攜帶有關(guān)數(shù)據(jù)（例如，所有者、標(biāo)題、制作日期等）的信息。因此，水印就能唯一的標(biāo)記受保護(hù)的作品，幫助解決有關(guān)數(shù)據(jù)所有權(quán)的紛爭(zhēng)的問。</p><p>　　有很多的關(guān)于

43、多媒體知識(shí)產(chǎn)權(quán)的糾紛或所有權(quán)的訴訟案件方面的例子。在2000年初一，關(guān)于Napster軟件版權(quán)侵犯的訴訟案得到了普遍關(guān)注。數(shù)字視頻光盤協(xié)會(huì)（即數(shù)字通用光盤協(xié)會(huì)）號(hào)召提出一些版權(quán)保護(hù)措施來加強(qiáng)資源的管理。該想法是，賣給消費(fèi)者的DVD播放器將允許不受權(quán)限的復(fù)制家庭錄影以及觀看的是移電視節(jié)目，不能輕易將此行為定位商業(yè)侵權(quán)。像Imprimature和CITED的歐洲項(xiàng)目和Working Group 提議的關(guān)于知識(shí)產(chǎn)權(quán)的美國項(xiàng)目都提到了電子版權(quán)的

44、管理。</p><p>　　由于其重要性，水印領(lǐng)域有了大量關(guān)于信息隱藏技術(shù)的研究。有很多文章（其中的參考文獻(xiàn)）和書籍都解釋了數(shù)字水印的基本知識(shí)，探索了它的許多實(shí)際應(yīng)用，并評(píng)估在各種攻擊下的性能。</p><p>　　II. 數(shù)字水印系統(tǒng)</p><p><b>　　A. 水印</b></p><p>　　在水印的通信模型

45、中，水印嵌入扮演的是發(fā)射機(jī)，水印檢測(cè)器扮演的是接收機(jī)。攻擊代表通信系統(tǒng)信道（圖1）。目標(biāo)是可靠檢測(cè)出隱藏信息；原始圖像可以被理解為一直用于編碼器的隱藏信息。但是，僅僅是在隱藏（不公開）的情況下解碼器可用。</p><p>　　在本文中，發(fā)射機(jī)發(fā)出了一個(gè)信號(hào)Xn到一個(gè)接收機(jī)（有n個(gè)傳輸），通過以下途徑：Yn=Xn+Sn+Zn,其中，Sn，Zn是獨(dú)立分布高斯隨機(jī)向量，這里Sn屬于編碼器，Zn既不屬于編碼器也不屬于解碼

46、器的噪聲。實(shí)驗(yàn)證明這里的信道容量相當(dāng)于Sn在解碼器中的角色。</p><p>　　一般而言，這個(gè)數(shù)字隨圖像尺寸的大小線性變化，其中C即比例常數(shù)，實(shí)際的水印以每幅圖片Rbit的速度嵌入，其中R<C，即C=1/2log(1+P/N),P是Xn的功率的平均值。結(jié)果相當(dāng)出乎人們的意料是因?yàn)槠谕鸖n扮演接收器的干擾的角色，這樣就能阻礙信息的監(jiān)測(cè)。這個(gè)對(duì)水印的比喻很明顯：在公眾的檢測(cè)方，Sn是看不見的（僅僅是隱藏的水印

47、），Xn代表嵌入的水印，Zn將是由攻擊者添加的噪聲。</p><p><b>　　水印特征</b></p><p>　　在數(shù)字水印的設(shè)計(jì)方案中，我們主要考慮兩個(gè)關(guān)鍵因素：</p><p>　　可見性：隱藏的信息不能隨意干擾主要的信號(hào)。因?yàn)樗〉哪康氖潜Ｗo(hù)那些在一般情況下很容易改變的多媒體數(shù)據(jù)，所以前面的要求是合理的。換句話說，如果做了改動(dòng)一些作

48、品允許的范圍（由于水印的原因），那么這些圖片和音樂作品將變得沒用了，嵌入水印必須具有隱蔽性，這一要求通常出現(xiàn)在對(duì)失真具有的限制的情況中。</p><p>　　魯棒性：即使在因?yàn)閻阂夤艋蚱渌幚恚炕?，D/A轉(zhuǎn)換等）導(dǎo)致降質(zhì)之后，信息必須能在水印圖像中被檢測(cè)出來（在這篇論文中演示文本被假定是一幅圖像，同樣的技術(shù)能夠被應(yīng)用在多媒體數(shù)據(jù)的其它類型方面）。當(dāng)然，水印的可檢測(cè)性與由攻擊者導(dǎo)致的最大程度的失真緊密相關(guān)。一個(gè)

49、水印方案如果允許準(zhǔn)確檢測(cè)跟嵌入水印的質(zhì)量很相近的失真圖像隱藏的信息，那這個(gè)方案將變得很完善。（這項(xiàng)規(guī)定同樣出現(xiàn)在失真限制中）</p><p><b>　　檢測(cè)方案</b></p><p>　　這里有兩個(gè)檢測(cè)方案：盲檢測(cè)和明檢測(cè)。在盲檢測(cè)方案中，檢測(cè)器需要原始圖像，在明檢測(cè)中不需要。盡管明檢測(cè)方案在實(shí)際中更有用（在檢測(cè)過程中不總需要原始圖像），盲檢測(cè)的魯棒性要求更多。&

50、lt;/p><p>　　III. 數(shù)字水印技術(shù)在無記憶攻擊和搞死攻擊的出現(xiàn)</p><p>　　對(duì)于離散和連續(xù)的字母，我們選擇的攻擊固定的記憶頻道（攻擊范圍分布的條件概率）。我們建立在無記憶攻擊下的可達(dá)到的速率區(qū)域。</p><p><b>　　離散字母</b></p><p>　　圖2顯示了可考慮到的常規(guī)水印系統(tǒng)模型。水印

51、指數(shù)W均勻分布在一個(gè)大小2nRw的集合中，Yn是受水印標(biāo)記的圖像，它可以在一個(gè)2nRw大小的代碼源中找到，攻擊信道是無記憶的。</p><p>　　水印解碼器的輸出是W'。因?yàn)椋颂幰旾n在檢測(cè)器中為已知，所以僅考慮的是隱藏情況。</p><p><b>　　攻擊程序</b></p><p>　　水印制造者（設(shè)計(jì)了編碼器對(duì)）試圖將能夠獲

52、得的速率域達(dá)到最大，同時(shí)攻擊者試圖將此速率降到最小，攻擊過程如下所示：</p><p>　　1.將編碼器（嵌入水?。┰O(shè)計(jì)成沒有任何攻擊的條件分布p.首先，將水印設(shè)計(jì)成具有編碼功能的f。水印必須確保速率的選擇能使得在任何由攻擊者選擇的p分布的攻擊下能夠準(zhǔn)確檢測(cè)出來；</p><p>　　2.知道p的攻擊，第二次選擇了p的條件分布，例如限失真；</p><p>　　3.

53、水印的下一步，設(shè)計(jì)水印編碼的g算法，同時(shí)考慮攻擊者p的分布。</p><p><b>　　IV. 結(jié)論</b></p><p>　　在這篇論文中，我們建立了在無記憶攻擊下所能夠達(dá)到的速率域。本篇論文討論了一下的這些東西：</p><p>　　1.無記憶攻擊時(shí)嵌入策略的自主選擇，對(duì)編碼器和解碼器同樣適用。從兩個(gè)統(tǒng)計(jì)學(xué)模型獲得證明：一般對(duì)任意圖像失

54、真約束的離散分布和字母的情況，以及高斯情況下原始圖像和攻擊通道是獨(dú)立同分布高斯，而失真度量是二次的。</p><p>　　2.信息攻擊者所操作的是攻擊程序。特別是，攻擊者知道了水印編碼者所使用的編碼功能，同時(shí)，解碼者知道了攻擊分布。</p><p>　　最后，這里有一些未來的研究方向。在本論文中所產(chǎn)生的所有結(jié)果在隱藏情況下發(fā)生。我們無法建立在公眾情況下所能達(dá)到的可能速率，即使是簡(jiǎn)單的高斯情

55、況。此外，在限失真，不涉及平均的失真對(duì)策的情況下聯(lián)合水印和壓縮系統(tǒng)去建立速率域的研究將會(huì)很有趣。</p><p><b>　　謝 辭</b></p><p>　　本論文的工作是在我的導(dǎo)師潘崧的悉心指導(dǎo)下完成的，潘崧老師嚴(yán)謹(jǐn)?shù)闹螌W(xué)態(tài)度和科學(xué)的工作方法給了我極大的幫助和影響。在此衷心感謝潘崧老師對(duì)我的關(guān)心和指導(dǎo)。</p><p>　　潘崧老師