視頻編碼相關(guān)外文資料翻譯

1、<p>　　畢業(yè)設(shè)計(論文)外文資料翻譯</p><p>　　附件1：外文資料翻譯譯文</p><p><b>　　摘要</b></p><p>　　在當今的視頻傳輸與廣播網(wǎng)絡(luò)中，版權(quán)的保護問題已經(jīng)變得越來越緊迫。這是因為視頻拷貝的出現(xiàn)并沒有降低原始視頻文件的品質(zhì)。一種保護版權(quán)的方法是在視頻序列中嵌入一段數(shù)字密碼，這段數(shù)字密碼的學術(shù)名

2、稱叫做水印。</p><p>　　因此，這篇課題的目的就是研究低復雜度的壓縮域H.264視頻水印算法。視頻編碼標準決定了H.264/MPEG-4 AVC的壓縮標準。這種算法充分使用了H.264壓縮標準了明確性，原始視頻的水印也是被隨機嵌入的。這些都建立在水印算法安全的基礎(chǔ)上。</p><p>　　在這種基于壓縮域的H.264視頻水印算法方案中，水印可以在沒有原始視頻序列存在的條件下被檢測出

3、來。水印可以在所選的編碼器修正系數(shù)的幫助下提取出來。</p><p>　　被提議的視頻水印方案</p><p><b>　　導言</b></p><p>　　在這項方案中，使用的是被提議的基于壓縮域的H.264視頻水印算法，它工作在I幀的宏塊層上。在這一層上，一個宏塊中包含視頻幀的一個16×16的采樣區(qū)域。從這個包含于一個宏塊上的16

4、×16采樣區(qū)域中，我們?nèi)〕鲆粋€8×8的分塊。從這個8×8的分塊中，我們進一步提取一個4×4的分塊，用, ,和表示一個宏塊中的4種不同的4×4分塊。這個4×4分塊將幫助我們在FDCT（快速離散余弦變換）之前構(gòu)建4×4 DCT系數(shù)的輸入分塊。</p><p>　　對于水印的嵌入過程，選擇I幀可以歸結(jié)于兩個主要的理由。首先，僅僅是因為它們的存在對視頻

5、序列來說是非?；镜?。其次，P幀和B幀被運動補償高度壓縮，它們中只有較少的空間再嵌入水印。一個I幀中的宏塊是被內(nèi)部編碼好的，這就意味著在同樣的片段里，先前被編碼的采樣中，每個16×16或者4×4亮度區(qū)域和8×8色度區(qū)域是可以被預測的。正如表格3.1所示，兩種亮度預測模式被賦予了不同的職責。第一，4×4亮度預測模式適用于幀的詳細區(qū)域選擇的；第二，16×16亮度預測模式適用于幀的平滑區(qū)域選擇

6、的。內(nèi)部預測之后，殘余數(shù)據(jù)的每個4×4分塊被變換為一種整數(shù)形式。整數(shù)形式，意味著所有工作可以用避免降低解碼精確度的整數(shù)算法來完成。如果宏塊在16×16內(nèi)部預測模式中編碼，所有4×4分塊的DC系數(shù)就被改變了。這種變換是一種在經(jīng)過4×4整形變換之后的4×4哈德瑪變換而形成的，為了使這些系數(shù)能夠更加關(guān)聯(lián)。</p><p>　　用被提議的視頻水印算法，將水印嵌入到宏塊中的

7、一個量化系數(shù)中。水印嵌入過程中的量化系數(shù)，是隨機選擇的，這就保證了方案中水印算法的安全性。即使將水印嵌入宏塊的唯一一個量化系數(shù)中，也不會導致可見性。這樣，攻擊者也就不能確定究竟選擇的是哪個量化系數(shù)。如果他想要使水印檢測不能發(fā)生，他就必須改變至少一半的量化系數(shù)。如果一半量化系數(shù)被改變，視頻也就無法使用了。</p><p>　　方案中所選的是宏塊中的AC量化系數(shù)，它利用的幾個比特來進行水印的嵌入?？刂屏说趇個宏塊的A

8、C量化系數(shù)的選擇。這項方案中的用到的視頻水印算法，可以在自我聯(lián)結(jié)攻擊的情況下防止同樣的被每個視頻幀使用，即使這樣將會成為傳輸一個短的有利條件。為了解決這個問題，我們需要對每個視頻幀利用一些不同的。這能解決一個問題，但是也產(chǎn)生了另一個問題。產(chǎn)生的問題是這樣的，在這種情況下，我們需要傳送一個很長的密鑰。傳送一個很長的密鑰所帶來的問題是，我們所用的基于壓縮域的H.264視頻水印算法根本無法實施。這個問題通過生成一個普通的密鑰來解決。是來自一個

9、公鑰和一個私鑰的結(jié)合，是從宏塊中提取的，它具有魯棒性，是被版權(quán)擁有者所擁有的，如圖3.1所示。從每個宏塊中提取，用一種明文形式將和輸入到一個密碼系統(tǒng)中。密碼系統(tǒng)產(chǎn)生一段密文，這就是第i個宏塊詳細的密鑰。在這種情況下，利用一種又快又簡單的視頻水印方案比密碼系統(tǒng)的安全性要高。</p><p>　　在這一過程中，我們利用了明文和的模二運算作為轉(zhuǎn)換密碼。密鑰的兩個比特決定了宏塊中8×8分塊的選擇，另兩個比特決定

10、了8×8分塊中的4×4分塊。在水印嵌入過程中，所選的4×4分塊中的AC系數(shù)是由4個比特的和所決定的。</p><p>　　宏塊的一些特征系數(shù)很敏感，它們決定了一旦有些細微的擾亂，視頻序列的品質(zhì)就會降低。應(yīng)該被提取出來，以防攻擊者想要細微的改變這些特征系數(shù)，然后視頻序列的品質(zhì)就降低了。這些特征系數(shù)在宏塊中是確實存在的，因此應(yīng)該切實得考慮它們。在這篇文章中，我們使用宏塊中分塊的水平差分直

11、流DC系數(shù)作為一種特征系數(shù)來提取。第二種方法是利用宏塊中的垂直差分系數(shù).</p><p>　　在水印嵌入過程中，我們只是利用了4×4內(nèi)部預測編碼的宏塊。不用預測編碼宏塊的理由是這樣的，內(nèi)部預測模式已經(jīng)被幀的平滑區(qū)域使用，同時水印的嵌入導致了內(nèi)部預測模式的可見性。另一個理由是，內(nèi)部預測編碼的哈德瑪變換增加了DC系數(shù)的去相關(guān)性，而且，許多去相關(guān)DC系數(shù)的值是零。所以，懂得怎樣從宏塊中提取是很重要的。<

12、/p><p><b>　　公鑰的提取</b></p><p>　　應(yīng)該用一種具有魯棒性的方法提取，這樣就能防止對視頻序列的有害影響。這意味著的提取不僅具有隨機性，而且具有魯棒性。為了完成目標，一種特性是利用了人類視覺的敏感性，還有一種是利用分塊的DC系數(shù)。雖然現(xiàn)在有兩種特性我們可以使用，但是我們必須找出一種最好的方法來防止攻擊者的攻擊。</p><p&

13、gt;　　我們可以用DC系數(shù)本身來提取公鑰。但是在這種方法中存在一種威脅，攻擊者已經(jīng)有能力用精確的方法改變每個宏塊的DC系數(shù)，而不影響宏塊的總數(shù)。這樣做的后果是，水印的檢測會被完全阻礙，也就是說水印將不可能被提取。對這種行為的后果是可以預見的，盡管這種后果不會造成視頻幀品質(zhì)的改變。</p><p>　　另一方面，也可以用圖3.1所表示的分塊的差分DC系數(shù)來提取。使用這種方法后，攻擊者將很難讓版權(quán)擁有者對水印的檢測

14、不能發(fā)生。這是因為，他必須完成一個或多個宏塊中DC系數(shù)的增加或減少，使的提取完全不可能。一旦攻擊者完成這項操作，將會造成可見性被辨認出的后果。用表示分塊中的差分DC系數(shù)來提取，如圖3.1所示，1≤α≤M,1≤β≤N ，M × N是視頻幀的總共大小。</p><p>　　另一個提取的特征系數(shù)是垂直差分系數(shù)，</p><p><b>　　當時，變化為</b>&l

15、t;/p><p><b>　　公鑰的決定</b></p><p>　　首先，我們提取第i個宏塊中的分塊的系數(shù)；第二，我們希望所有這些提取的系數(shù)將呈現(xiàn)一種混亂形式。為了達到這個目的，所有提取的系數(shù)將呈現(xiàn)為DC系數(shù)的矢量形式。矢量形式的量化DC系數(shù)是在第i個宏塊中的，它是由24個塊構(gòu)成的。亮度分量Y提供了16個塊給，而色度分量和各提供了4個塊給，如下圖3.3所示。</p

16、><p>　　用第i個比特位表示出構(gòu)成公鑰的所在第i個宏塊中的位置。公鑰在宏塊中的位置，是從矢量形式的量化DC系數(shù)中獲得的。下面的式子就是獲得的方法：</p><p><b>　　水印的嵌入過程</b></p><p>　　在壓縮域的水印嵌入過程進行之前，首先要進行壓縮視頻比特流的解碼，這是為了更接近熵編碼標準。一旦熵編碼標準與水印嵌入過程更接近以

17、后，計算量將更少，因此也更適合于實時應(yīng)用。此外，水印嵌入過程越接近DCT變換過程，對壓縮視頻比特流的品質(zhì)影響就越小。在量化之后進行水印的嵌入過程是非常值得的，這是為了確保水印沒有被消除，量化過程也沒有產(chǎn)生損耗。因此，在這篇文章中，我們準備在重新排序的量化AC系數(shù)中嵌入水印。這是因為這樣將避免量化過程產(chǎn)生損耗，熵編碼與解碼是一個快速的過程，水印的嵌入與提取也可以在實時的狀態(tài)下完成。</p><p>　　我們可以發(fā)現(xiàn)

18、，實際上是由24個比特構(gòu)成的，這些比特沒有完全被利用。只有其中的幾個比特，在系數(shù)中被選擇出來用作水印嵌入。從第i個宏塊中選取的系數(shù)是被修正過了，為了在宏塊中嵌入水印信息。如下所示的表格3.2揭示了所選系數(shù)的修改方法。</p><p>　　V是原始視頻序列，是水印嵌入后的視頻序列，S是比例系數(shù)，是嵌入第i個宏塊中的水印信號，是第i個宏塊中隨機選取的用作水印嵌入的修正系數(shù)。從下面的式子中，我們可以得到嵌入水印的視頻序

19、列。</p><p><b>　?。?）</b></p><p>　　當 時 </p><p><b>　?。?）式變化為</b></p><p>　　這只是最大限度改變水印嵌入的量化系數(shù)的一步，因此，一個宏塊修正的

20、量化步長值和同一個宏塊的非量化DCT系數(shù)一樣大。最終，水印嵌入導致的量化誤差造成了視頻的品質(zhì)降低了。</p><p>　　附件2：外文原文（復印件）</p><p><b>　　ABSTRACT </b></p><p>　　In today's video delivery and broadcast networks, issues

21、 of copyright protection have become more urgent than in analog times. This is simply due to copying of digital video does not result in the decrease in quality that occurs when analog video is copied. One of the methods

22、 of copyright protection is to embed a digital code into the video sequence. The digital code is termed as watermark.. </p><p>　　Therefore, the objective of this project is to propose a digital video waterm

23、arking algorithm for H.264 in the compressed, domain whose complexity is low. The video coding standard decided to use was the H.264/MPEG4 AVC compression standard. The proposed algorithm exploits the specific characteri

24、stics of the compression standard H.264. The watermark to be embedded into the original video is randomly localized. It is on this foundation that our proposed video watermarking algorithm bases its secu</p><p

25、>　　In this proposed video watermarking algorithm for H.264 in the compressed domain, the watermark can be detected without the presence of the original video sequence. The watermark is blindly extracted by the help of

26、 the selected modified coefficient, at the decoder. </p><p>　　PROPOSED VIDEO WATERMARKNG SCHEME</p><p>　　Introduction</p><p>　　The video watermarking algorithm for H.264 in the com

27、pressed domain proposed in this project works at the macroblock level of I-frames. At this level, a 16×16 sample region of a video frame is contained in a macroblock. From the 16×16 sample region of a video fra

28、me contained in a macroblock, we extract an 8×8 block. From the 8×8 block of the 16×16 sample region of a video frame contained in a macroblock, we further extract a 4×4 block. The , , and represent t

29、he four different 4×4 blocks in the 8</p><p>　　For watermark embedding process, I-frames are chosen due to two major reasons. First,simply because their existence is very fundamental for the video seque

30、nce. Secondly, the two frames P and B are highly compressed by motion compensation and there is less room in them to embed a watermark.. Macroblocks in an I-frame are intra-coded. This implies that each 16×16 or 4&#

31、215;4 luma region and each 8×8 chroma region is predicted from samples coded previously in the same slice as indicated in the Table 3.1</p><p>　　The watermark is embedded in one quantized coefficient of

32、 a macroblock by the proposed video watermarking algorithm. The quantized coefficient, for watermark embedding process is randomly selected. This is what guarantees the security of the proposed algorithm in this project

33、. Visible artifacts are not induced by embedding the watermark in only one quantized coefficient in a macroblock. Even though, the attacker cannot identify which quantized coefficient has been chosen .If he wants to the

34、mak</p><p>　　is the quantized AC coefficient selected in the MB for watermark embedding process by using several bits of the . The controls the selection of the quantized AC coefficient in the macroblock fo

35、r watermark embedding process. A proposed video watermarking algorithm in this project can be under self-collusion attack in case the sameis used for every video frame, although this would be of an advantage due to trans

36、mission of a very short. To resolve this kind of dispute, we need to have several diff</p><p>　　At this point we use a shift cipher with modulus 2 that is to say the plaintext , and the . There are 2 bits of

37、 generated general key that determine the selected 8×8 block in the macroblock. There are also other 2 bits that determine the selected 4×4 block in the 8×8 block. For watermark embedding process, the wh

38、ich is the selected AC coefficient in that 4×4 block is determined by a sum of 4 bits.</p><p>　　Some characteristics in the MB are so delicate to the point that once they are tempered with slightly in a

39、 disorganized manner, the quality perception of the video sequence must degrade. It is these characteristics then that the should be extracted. In case the attacker tempers just slightly to change that exploited charact

40、eristic, then the video sequence quality perception must degrade. Since these kinds of characteristics do exist in the MB, then it is therefore strongly advised to utilize th</p><p>　　For watermark embedding

41、 process, it is only the 4×4 intra predicted MBs that are utilized. There are reasons as to why the intra predicted MBs are not utilized. The intra prediction mode is being utilized for the smooth regions of the fr

42、ames. And also at the same time the embedding of the watermark causing visible artifacts in the intra prediction mode. The other reason is that, the extra Hadamard transform for B16x16 intra predicted MBs decorrelates t

43、he DC coefficients even more. Furthermore,</p><p>　　Public Key, Pk Extraction</p><p>　　The should be made in such a way that it is robust. This is to prevent some harmful consequences to the vi

44、deo sequence. This means that the extraction of the is not automatic but also robust. To fulfill the target, one of the MB characteristic is utilized to which the human visualization is sensitive. One of that characteri

45、stic is the DC coefficients of the blocks. But then there two approaches in which we can exploit that characteristic. However we are supposed to exploit the best approach to</p><p>　　It is possible to utiliz

46、e the DC coefficient itself to extract the public key, . But there is danger in this kind of approach. The danger is that the attacker has got the ability of carrying out the changes to the DC coefficient of each and eve

47、ry block by exactly the same amount. By so doing, the detection of the watermark is completely blocked and hence watermark detection impossible. The consequence for such an action is the visualization of brighter frames

48、or darker frames. Despite these cons</p><p>　　On the other hand, it is also possible to utilize the relative difference of the DC coefficients of the blocks as shown in Figure 3.1 above to extract the from

49、 the MB. Applying this kind of approach, it becomes so hard for the attacker to make the extraction impossible for the copyright owner. This is simply because he must carry out the decrement or increment of the DC coeff

50、icient of either one block or more to completely disable the extraction. Once the attacker carries out this kind of op</p><p>　　Another characteristic from which the is extracted is the vertical relative d

51、ifference of the coefficients in a MB, .</p><p>　　For becomes</p><p>　　Public Key, Pk Determination</p><p>　　First of all we extract the coefficients of blocks of the macroblock

52、. Secondly, all those extracted coefficients are supposed to be presented in a key scrambled style. To fulfill this, all the extracted coefficients are presented in a vector form of the DC coefficients, . The quantized

53、 DC coefficient of the MB in the vector form, can be made up of twenty four elements. The luma component, Y contributes 16 elements to the whereas the chroma component contributes 4 elements from each of its </p>

54、<p>　　Let represent the position of the bits that make up a public key in an macroblock, . The bits of the public key in a macroblock, are obtained from the quantized DC coefficients that are presented in a vect

55、or style, . Below is how the bits are obtained:</p><p>　　VIDEO WATERMARK EMBEDDING PROCESS</p><p>　　In a compressed domain before carrying out the watermarking process first decode the compress

56、ed video bitstream. This is to bring nearer the entropy coding level. Once the entropy coding level is brought nearer to the watermark embedding process, then the computation complex becomes less and hence more suitable

57、for real time applications. Furthermore, the nearer the watermark embedding process is to the DCT transform operations, the less induced degradation to the compressed video bitstream. It </p><p>　　We saw tha

58、t the is actually made up of several bits that total up to 24. All these bits are not utilized. It is only several of those bits that are utilized in the selection of the coefficient in the MB for watermark embedding.

59、The selected coefficient from the MB is modified in order to embed the watermark in the MB. The table 3.2 shows how to modify the selected coefficient .</p><p>　　If V is the original video sequence,

60、is the watermarked video sequence, S is the scaling factor, is the watermark to be embedded in the MB, and is the modified randomly selected coefficient from the MB for watermark embedding, then we obtain the watermar

61、ked video sequence from the expression below </p><p><b>　?。?）</b></p><p>　　Let （9）</p><p>　　and hence the expression (8) becom

62、es </p><p>　　It is just one level that is the maximum change made to the quantized coefficient selected for watermarking embedding. Therefore, the quantization step size modification of a block is as large a