NTIT1 A chaos-based robust wavelet- domain watermarking algorithm Source: Chaos, Solitions and Fractals, Vol. 22, 2004, pp Authors: Zhao Dawei, Chen Guanrong, Liu Wenbo Speaker: Hao-Cheng Wang( 王皓正 ) Date: 2004/9/22
NTIT 2 Outline Introduction Watermarking in the wavelet domain DWT (Discrete Wavelet Transformation) Chaos and its application to watermarking The new watermarking algorithm Watermark embedding Watermark detection Results and analysis Conclusions Comment
NTIT 3 Watermarking in the wavelet domain The digital watermarking technology includes Spatial-domain Transform-domain DCT, DWT
NTIT 4 DWT (Discrete Wavelet Transformation)(1/2) 低頻 (low frequency) :像素之間的變化小,影像較平 滑,人眼的敏感度高. LL 1 高頻 (high frequency) :像素之間的差異大,影像較粗 糙、模糊,人眼的敏感度較低. HH 1 中頻:介於低頻與高頻之間. HL 1 、 LH 1 LH 1 HH 1 HL 1 LL 1
NTIT 5 DWT (Discrete Wavelet Transformation)(2/2) LH 1 HH 1 HL 1 LL 1 LH 1 HH 1 HL 1 LH 2 HH 2 HL 2 LL 2
NTIT 6 Logistic map Where When, the map is in the chaotic state. Where Chaos and its application to watermarking
NTIT 7 Logistic map(1/3) Example
NTIT 8 Logistic map(2/3) Example
NTIT 9 Logistic map(3/3) We will use the logistic map twice: To generate a label sequence To generate the watermark
NTIT 10 The new watermarking algorithm Apply the wavelet transform locally Watermark embedding I ori (256 × 256) I sub (128 × 128) DWT IDWT I’ sub (128 × 128) I’ ori (256 × 256) 8×8 block
NTIT 11 Watermark embedding(1/7) ……………….. ……………………………….. ……………………….. Original Image (256 × 256 pixels)
NTIT 12 Watermark embedding(2/7) 33, 1023, 112, 36, 77……………96, 1, 64…………………….983, 124, ……………………………… Label Sequence (Length=256) (1) (2)
NTIT 13 Watermark embedding(3/7)
NTIT 14 Watermark embedding(4/7) LH 1 HH 1 HL 1 LH 2 HH 2 HL 2 LH 3 HH 3 HL 3 LL 3 DWT
NTIT 15 Watermark embedding(5/7) Type …… Type 2 [1, -1]
NTIT 16 Watermark embedding(6/7), i=1, 2, …, N C band are the original wavelet coefficients C’ band are the watermarked wavelet coefficients αis a global parameter accounting for the watermark strength w is the watermark signal N is the element number of subband HL 1 or HH 1 or LH 1 {HL1, HH1, LH1} (1) (2) (3)
NTIT 17 Watermark embedding(7/7), i=1, 2, …, N i … i wm (i) … w(i)w(i) … C band … C’ band …
NTIT 18 Watermark detection(1/2) The detection method we use is similar to the method proposed in [1] We adopt the Neyman-Pearson criterion to determine the threshold T p [1] Barni M, Bartolini F. Improved wavelet-bsed watermarking through pixel- wise masking. IEEE Trans Image Processing 2001;10(5):
NTIT 19 Watermark detection(2/2) if ρ>T ρ : a watermark signal exists; otherwise, a watermark signal does not exist see [1] for more details (1) (2)
NTIT 20 Results and analysis(1/3) Test images: “Lena” and “Barbara”(256×256 pixels) α=6.0, i seq = and i wm =0.4123
NTIT 21 Results and analysis(2/3) PSNR=39.30
NTIT 22 Results and analysis(3/3) When we set α=1.0, or smaller, we cannot detect the watermark correctly
NTIT 23 Robustness against various attacks α=6, P f =10 -8, i seq =0.1564, i wm = Additive noise attacks Gaussian noise Salt and pepper noise JPEG compression Geometric manipulations Cropping, resizing, rotation
NTIT 24 Cropping
NTIT 25 Resizing and rotation Resizing Zoom scale m Zoom in (m>1) Zoom out (m<1) m >0.625 Rotation 25°
NTIT 26 Conclusions This scheme applies the wavelet transform locally, based on the chaotic logistic map, and embeds the watermark into the DWT domain. Introduced a blind watermarking detection technique using the Neyman-Pearson criterion. Highly robust against geometric attacks and signal processing operations and JPEG compression.
NTIT 27 Comment(1/2) 結合圖片的浮水印技術 …… LH 1 HH 1 HL 1 LH 2 HH 2 HL 2 LH 3 HH 3 HL 3 LL 3 c1c1 c 1 mod 2 = 1 or 0 ? c 1 ±1
NTIT 28 Comment(2/2) p1p1 p 1 mod 2 = 1 or 0 ? p 1 ±1