Blind dual watermarking for color images’ authentication and copyright protection Source : IEEE Transactions on Circuits and Systems for Video Technology Authors : Xiao-Long Liu, Chia-Chen Lin, and Shyan-Ming Yuan Speaker : Jun-Yong Chen Date : 4/22 題目、來源、作者。
Outline Related Works Proposed Scheme Experimental Results Conclusions YCbCr Color Space DWT Proposed Scheme Embedding Robust Watermark Embedding Fragile Watermark Extracting Fragile Watermark Extracting Robust Watermark Experimental Results Conclusions Comments 目錄。
Related Work- YCbCr Color Space The transform from RGB to JPEG-YCbCr is as follow: The backward transform from JPEG-YCbCr to RGB is as follow: 這邊是RGB 轉成 YCbCr 的公式 右邊的圖是一張彩圖轉成 YCbCr 後的樣子
Related Work- YCbCr Color Space RGB The transform from RGB to YCbCr Y Y = 0.2989*R + 0.5866*G + 0.1145*B + 0 187 239 131 187 ≅ 0.2989 * 185 + 0.5866 * 211 +0.1145 * 68 + 0 = 55.2965 + 123.7726 + 7.786 + 0 利用 R Channel轉成 Y Channel的例子 239 ≅ 0.2989 * 255 + 0.5866 * 251 + 0.1145 * 140 + 0 = 76.2195 + 147.2366 + 16.03 + 0 131 ≅ 0.2989 * 0 + 0.5866 * 174 + 0.1145 * 249 + 0 = 0 + 102.0684 + 28.5105 + 0
Related Work- YCbCr Color Space RGB The transform from RGB to YCbCr Cb Cb = (-0.16874)*R + (-0.33126)*G + 0.5*B + 0.5 -66 -55 68 -66 ≅ (-0.16874)*185 + (-0.33126)*211 + 0.5*68 + 0.5 = (-31.2169) + (-69.89586) + 34 + 0.5 利用 G Channel轉成 Cb Channel的例子 -55 ≅ (-0.16874)*255 + (-0.33126)*251 + 0.5*140 + 0.5 = (-43.0287) + (-83.14626) + 70 + 0.5 68 ≅ (-0.16874)*0 + (-0.33126)*174 + 0.5*249 + 0.5 = 0 + (57.63924) + 124.5 + 0.5
Related Work- YCbCr Color Space RGB The transform from RGB to YCbCr Cr Cr = 0.5*R + (-0.41869)*G + (-0.08131)*B + 0.5 -1 11 -93 -1 ≅ 0.5*185 + (-0.41869)*211 + (-0.08131)*68 + 0.5 = 92.5 + (-88.34359) + (-5.52908) + 0.5 11 ≅ 0.5*255 + (-0.41869)*251 + (-0.08131)*140 + 0.5 = 127.5 + (-105.09119) + (-11.3834) + 0.5 利用 B Channel轉成 Cr Channel的例子 -93 ≅ 0.5*0 + (-0.41869)*174 + (-0.08131)*249 + 0.5 = 0 + (-72.85206) + (-20.24619) + 0.5
Related Work- YCbCr Color Space YCbCr The transform from YCbCr to RGB R R = 1*Y + 1.402*Cr 186 254 1 187 -66 -1 239 -55 11 131 68 -93 186 ≅ 1*187 + 1.402 * (-1) = 187 + (-1.402) 把 Y Channel轉成 R Channel的例子 254 ≅ 1*239 + 1.402 * 11 = 239 + 15.422) 1 ≅ 1*131 + 1.402 * (-93) = 131 + (-130.386)
Related Work- YCbCr Color Space YCbCr The transform from YCbCr to RGB G G = 1*Y+ (-0.34414)*Cb + (-0.71414)*Cr – 0.5 210 250 174 187 -66 -1 239 -55 11 131 68 -93 210 ≅ 1*187 + (-0.34414) * (-66) + (-0.71414) * (-1) – 0.5 = 187 + 22.71324 + 0.71414 – 0.5 把 Cb Channel轉成 G Channel的例子 250 ≅ 1*239 + (-0.34414) * (-55) + (-0.71414) * (11) – 0.5 = 239 + 18.9277 + (-7.85554) – 0.5 174 ≅ 1*131 + (-0.34414) * (68) + (-0.71414) * (-93) – 0.5 = 131 + (-23.40152) + 66.41502 – 0.5
Related Work- YCbCr Color Space YCbCr The transform from YCbCr to RGB B B = 1*Y+ 1.772*Cb – 0.5 187 -66 -1 239 -55 11 131 68 -93 70 141 251 70 ≅ 1*187 + 1.772 * (-66) – 0.5 = 187 + (-116.952) – 0.5 把 Cr Channel轉成 B Channel的例子 141 ≅ 1*239 + 1.772 * (-55) – 0.5 = 239 + (-97.49) – 0.5 251 ≅ 1*131 + 1.772 * 68 – 0.5 = 131 + 120.496 – 0.5
Related Works- DWT 163 161 160 159 324 320 2 322 -4 321 1 L H 324 320 2 322 -4 321 1 644 642 4 -4 2 -2 LL HL H 把一張圖做DWT壓縮的例子 先是對它做垂直的壓縮後,再對它做水平的壓縮 這樣一來一張圖的壓縮LL區塊是最重要的 L LH HH
Proposed Scheme- Embedding Robust Watermark Color image 分別轉成R,G,B R G B
Proposed Scheme- Embedding Robust Watermark 再分別對R,G,B轉成Y,Cb,Cr,由於使用YCbCr 可以有效抵抗JPEG 壓縮以及高斯噪聲攻擊,因此使用YCbCr會比RGB的抗攻擊效果來的好。 Y Cb Cr
Proposed Scheme- Embedding Robust Watermark 186 182 177 168 24 11 9 4 184 179 172 2 1 3 187 183 180 188 22 23 6 5 8 7 LL HL DWT LH HH 首先對Y進行DWT,取得DWT後的頻率域LL,HL,LH,HH Y
Quantization 這是等等要進行量化的量化表 [11] Int. Telecommunication Union, “Information technology-digital compression and coding of continuous-tone still images-requirements and guidelines,” CCITT Recommendation T.81, 1992.
Proposed Scheme- Embedding Robust Watermark 17 18 11 16 15 13 9 14 8 6 Quantization Q-LL 對LL區塊進行量化,分別除以量化表,得到量化後的LL區塊,這邊稱作Q-LL
Proposed Scheme- Embedding Robust Watermark 186 182 177 168 24 11 9 4 184 179 172 2 1 3 187 183 180 188 22 23 17 18 16 15 13 14 8 7 5 6 LL HL 17 18 11 16 15 13 9 14 8 6 Replace Q-LL LH Q-LL 然後把Q-LL直接替換掉HH區塊
Proposed Scheme- Embedding Robust Watermark Y’ 186 182 177 168 24 11 9 4 184 179 172 2 1 3 187 183 180 188 22 23 17 18 15 14 10 7 5 13 12 8 LL HL Robust Watermark bit 𝑊 𝑅𝑛 01001011… LH WQ-LL 然後利用Robust Watermark 藏在Q-LL,得到了WQ-LL。這邊的k是一個固定的參數,關係到水印的強度。 然後再進行Inverse DWT 轉換成空間域,就是Pixels. IDWT
Proposed Scheme- Embedding Robust Watermark Y’ Cb Cr 藏完後 Y’CbCr 轉成 RGB
Proposed Scheme- Embedding Robust Watermark R G B 即完成Robust Watermarked Image。
Proposed Scheme- Embedding Fragile Watermark 這邊是進行Fragile的水印嵌入,把Robust Watermarked Image轉成R,G,B, R G B
Proposed Scheme- Embedding Fragile Watermark 44 45 37 31 88 70 78 106 25 5 R 其中對R做嵌入的動作,假設Watermark bit 是 63 轉成二進制 是 111111,我們把它轉成3的n次方,這邊的例子我們設n為2,因此我們轉成9進制,可以得到 70 3 𝑛 −𝑏𝑎𝑠𝑒 𝑛𝑜𝑡𝑎𝑡𝑖𝑜𝑛𝑎𝑙 𝑠𝑦𝑠𝑡𝑒𝑚 𝑊 𝐹 = 111111 2 𝑊 𝐹 ′= 70 9 𝑛=2
Proposed Scheme- Embedding Fragile Watermark 44 45 𝑠=7 𝐸= 3 0 ×44 𝑚𝑜𝑑 9+ 3 1 ×45 𝑚𝑜𝑑 9=8 𝑡= 7−8+ 3 2 −1 2 𝑚𝑜𝑑 3 2 =3 然後我們選擇44以及45這兩個pixel進行嵌入,取得了digit E,再利用Watermark bit 7去取得 temporary value t,並且把它轉成二進制後分別-1,生成subtracted sequence t’’。 𝑡 ′ =𝑡(3)= 10 3 𝑡 ′′ = 𝑡 ′ (1−1,0−1)=(0,−1) 𝑊 𝐹 ′= 70 9
Proposed Scheme- Embedding Fragile Watermark 44 45 𝑝 1 ′ = 𝑝 1 + 𝑑 2 =44−1=43 𝑝 2 ′ = 𝑝 2 + 𝑑 1 =45+0=45 再利用subtracted sequence t’’ 進行嵌入。 43 45 𝑡 ′′ = 𝑑 1 ,…, 𝑑 𝑛 =(0,−1)
Proposed Scheme- Embedding Fragile Watermark 37 31 𝑠=0 𝐸= 3 0 ×37 𝑚𝑜𝑑 9+ 3 1 ×31 𝑚𝑜𝑑 9=4 𝑡= 0−4+ 3 2 −1 2 𝑚𝑜𝑑 3 2 =0 接下來我們利用一樣的方法,選擇37以及31這兩個pixel進行嵌入。 𝑡 ′ =𝑡(0)= 00 3 𝑡 ′′ = 𝑡 ′ (0−1,0−1)=(−1,−1) 𝑊 𝐹 ′= 70 9
Proposed Scheme- Embedding Fragile Watermark 37 31 𝑝 1 ′ = 𝑝 1 + 𝑑 2 =37−1=36 𝑝 2 ′ = 𝑝 2 + 𝑑 1 =31−1=30 43 45 36 30 88 70 78 106 25 5 最終,我們取得了嵌入後的Block 。 36 30 𝑡 ′′ = 𝑑 1 ,…, 𝑑 𝑛 =(−1,−1)
Proposed Scheme- Embedding Fragile Watermark 43 45 36 30 88 70 78 106 25 5 然後表示為R’ R’
Proposed Scheme- Embedding Fragile Watermark R’ G’ B’ 並且對G,B分別也進行了嵌入,然後得到了最終的Watermarked Image.
Proposed Scheme- Extracting Fragile Watermark 43 45 36 30 88 70 78 106 25 5 這邊是對R’進行提取,先轉成Block。 R’
Proposed Scheme- Extracting Fragile Watermark 43 45 43 45 36 30 88 70 78 106 25 5 𝐸= 3 0 ×43 𝑚𝑜𝑑 9+ 3 1 ×45 𝑚𝑜𝑑 9=7 36 30 𝐸= 3 0 ×36 𝑚𝑜𝑑 9+ 3 1 ×30 𝑚𝑜𝑑 9=0 再分別對兩兩Pixel進行提取Watermark然後再轉成 二進制,就是我們要的Fragile Watermark Bit,然後再對G’,B’進行一樣的運算後,總共可以取得3個Fragile Watermark Bit 再跟 Original Fragile Watermark進行比較,就可以偵測出被竄改的部分。 𝑊 𝐹 ′= 70 9 𝑊 𝐹 = (111111) 2
Proposed Scheme- Extracting Robust Watermark Y’ 186 182 177 168 24 11 9 4 184 179 172 2 1 3 187 183 180 188 22 23 17 18 15 14 10 7 5 13 12 8 LL HL DWT LH WQ-LL 這邊是Robust Watermark提取的部分,首先把Color image 轉成 Y’CbCr,然後對Y’進行DWT,因此我們可以得到WQ-LL,然後對LL區塊進行量化,得到Q-LL區塊,然後WQ-LL 減掉 Q-LL之後 再除以之前乘上去的固定參數k,就可以得到我們要的Robust Watermark. Y’ Robust Watermark 𝑊 𝑅𝑛 = (𝑊𝑄𝐿𝐿 𝑛 − 𝑄𝐿𝐿 𝑛 )/𝑘= 01001011…
Experimental Results 這個實驗結果用了八張的結果圖,並且得到了最高的視覺質量。
Experimental Results AR : Accurate Rate MP : Modified Pixels DP : Detected Pixels 這邊是檢測被竄改後的圖,它的準確率高達99點多百分比。 在這篇的方法中,第二個水印主要是應用於認證機制,因此並沒有在圖象修復的方面進行運算,因此只有檢測的功能。
Experimental Results NC : Normal Correlation 這邊是抗各種攻擊的實驗,例如拼貼、JPEG壓縮、胡椒鹽等攻擊。
Experimental Results [7] C. S. Lu, H. Y. M. Liao, “Multipurpose watermarking for image authentication and protection,” IEEE Transactions on Image Processing, vol. 10, no. 10, pp. 1579-1592, 2001. [8] P. Y. Lin, J. S. Lee, C. C. Chang, “Dual digital watermarking for internet media based on hybrid strategies,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 19, no. 8, pp. 1169-1177, 2009. [9] F. Lusson, K. Bailey, M. Leeney, K. Curran, “A novel approach to digital watermarking, exploiting colour spaces,” Signal Processing, vol. 93, no. 5, pp. 1268-1294, 2013. 這邊是與其他方法的分別比較,像是有沒有辦法檢測,藏在頻率域或是空間域,是否不可視的水印…等等。 Blind在這篇文章中的意思是,在取出水印時,不需要原圖就可以還原水印,同時也是這個文章的重點之一,因此在我的舉例當中,兩張水印皆不需要原圖即可復原。
Conclusions In this paper, author uses the property of YCbCr to against attacks. The author uses duel watermarks to achieve copyright protection and image authentication. Extraction of watermarks can be processed blindly without the original host image and watermarks. 結論,說明利用了甚麼技術來抵抗攻擊,並且運用了兩種水印,同時又可以進行盲提取,不用原圖或者原水印就可以從中提取出來。
Comments The method can detect the tampered pixels, but the method can repair the image whether this method will be better or not. Based on security consideration, whether this method is security or not. 雖然可以檢測出被竄改的部分,但是如果有修復的功能是否會更好 基於安全性的考量,這個方法是否安全