5.3 Experimental Results
5.3.3 Discussion
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Hamming Distance
Proposed scheme Lin’s scheme Kim’s scheme Lin’s scheme*
Franco’s scheme Rana’s scheme
4x4 8x8
16x16 4x4 8x8
16x16 block a Different schemes
block b block c block d
(a) Aloe image
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Hamming Distance
Proposed scheme Lin’s scheme Kim’s scheme Lin’s scheme*
Franco’s scheme Rana’s scheme
4x4 8x8
16x16 4x4 8x8
16x16 block a Different schemes
block b block c block d
(b) Cloth1 image
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Hamming Distance
Proposed scheme Lin’s scheme Kim’s scheme Lin’s scheme*
Franco’s scheme Rana’s scheme
4x4 8x8
16x16 4x4 8x8
16x16 block a Different schemes
block b block c block d
(c) Plastic image
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Proposed scheme Lin’s scheme Kim’s scheme Lin’s scheme*
Franco’s scheme Rana’s scheme
4x4 8x8
16x16 4x4 8x8
16x16 blo
ck a Different schemes
blo ck b
blo ck c
blo ck d
(d) Average of Middlebury Stereo 2006 Datasets of 21 image
Figure 5.10: Hamming distance comparison of proposed scheme with existing schemes [55,62,76,101] for synthesized center view at Stereo-JPEG compression at quality 75.(blocka = (Iw/8)×(Ih/8), blockb = (Iw/32)×(Ih/32), blockc = linearly embed and blockd= 16×256)
original left view. In this collusion attack, the independent view regions of the synthesized left view (generated usingDIBR technique) is filled using hole filling technique [9].
5.3 Experimental Results
1 1.4 1.8 2.2 2.6 3 3.4 3.8 4.2 4.6 5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Camera view number
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(a) Aloe image
1 1.4 1.8 2.2 2.6 3 3.4 3.8 4.2 4.6 5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Camera view number
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(b) Cloth1 image
1 1.4 1.8 2.2 2.6 3 3.4 3.8 4.2 4.6 5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Camera view number
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(c) Plastic image
1 1.4 1.8 2.2 2.6 3 3.4 3.8 4.2 4.6 5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Camera view number
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(d) Average of Middlebury Stereo 2006 Datasets of 21 image
Figure 5.11: Hamming distance comparison of proposed scheme with existing schemes [55,62,76,101] against view synthesis attack at Stereo-JPEG compression at quality 75 (where 1 & 5 defines the left and the right views and others are the interme- diate synthesized views).
existing literature, it has been observed that if the embedding view and the extraction view are not same, the watermark signal is degraded. Intuitively, if the size of the embedding block is less than the disparity, the watermark signal when extracted other than embedding view may be degraded. This may be the cause that existing schemes are not performing well for the relatively large disparity values. In the proposed scheme, since the block selection depends on the disparity value (refer to §5.1.4 and 5.2.3), the extracted watermark signal is not degraded even in case of large disparity values. This is the intuitive reason that the proposed scheme outperforms other existing schemes [55,62,76,101] when embedding and extraction views are different as shown in Fig. 8, 9, 10 and 11.
15 20 30 40 50 60 70 80 90 100 0
0.1 0.2 0.3 0.4 0.5 0.6
JPEG quality
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(a) Aloe image
15 20 30 40 50 60 70 80 90 100
0 0.1 0.2 0.3 0.4 0.5 0.6
JPEG quality
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(b) Cloth1 image
15 20 30 40 50 60 70 80 90 100
0 0.1 0.2 0.3 0.4 0.5 0.6
JPEG quality
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(c) Plastic image
15 20 30 40 50 60 70 80 90 100
0 0.1 0.2 0.3 0.4 0.5 0.6
JPEG quality
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(d) Average of Middlebury Stereo 2006 Datasets of 21 image
Figure 5.12: Hamming distance comparison of proposed scheme with existing schemes [55,62,76,101] against Stereo-JPEG compression for right view.
In other words, the selection of the similar dependent view regions from the left and right views and embedding identical watermarks in those similar dependent regions makes the scheme more robust than the existing literature against view synthesis process as well as collusion attacks. Additionally, careful selection of the watermark embedding strength and visual quality threshold makes the scheme robust against Stereo JPEG compression and any other noise addition attacks.
Time Complexity Analysis: In this proposed scheme, the watermark is embedded with the DC coefficients of the dependent regions of both the views.
To extract dependent view,DIBRis applied in pixels of the 3D image to generate the synthesis left and right view from the given view. The time complexity for dependent view generation isO(Ih×Iw), whereIwandIh represent the image with
5.3 Experimental Results
0 25 50 75 100 125 150 175 200
0 0.1 0.2 0.3 0.4 0.5 0.6
variance
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(a) Aloe image
0 25 50 75 100 125 150 175 200
0 0.1 0.2 0.3 0.4 0.5 0.6
variance
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(b) Cloth1 image
0 25 50 75 100 125 150 175 200
0 0.1 0.2 0.3 0.4 0.5 0.6
variance
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(c) Plastic image
0 25 50 75 100 125 150 175 200
0 0.1 0.2 0.3 0.4 0.5 0.6
variance
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(d) Average of Middlebury Stereo 2006 Datasets of 21 image
Figure 5.13: Hamming distance comparison of proposed scheme with existing schemes [55,62,76,101] against Gaussian noise attack at Stereo-JPEG compression at quality 75 for right view.
and height respectively. For embedding, n×nblock wise 2D-DCTis used and the DC coefficients are altered as the embedding rule. Using butterfly architecture, 1D-DCTcan be computed with complexity ofO(nlogn) for n coefficients. So for a n×nblock, time complexity of 2D-DCTisO(n2logn) [102]. In this watermarking scheme, some blocks are selected for watermarking (refer to §5.1.5). In extreme case, if all the blocks are selected for watermarking, (Ihn×I2w) times of 2D-DCTneed to be done for embedding of watermark in each view image. As a result, the total time complexity can be written as O(Ih×Iw+Ihn×I2wn2logn)'O((Ih×Iw) logn) for embedding of watermark with both the views. In this watermarking method, the watermarking block size is limited to 4×4, 8×8 and 16×16. So the final
0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2 0
0.1 0.2 0.3 0.4 0.5 0.6
Noise density
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(a) Aloe image
0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2
0 0.1 0.2 0.3 0.4 0.5 0.6
Noise density
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(b) Cloth1 image
0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2
0 0.1 0.2 0.3 0.4 0.5 0.6
Noise density
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(c) Plastic image
0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2
0 0.1 0.2 0.3 0.4 0.5 0.6
Noise density
Hamming Distance
Proposed 4X4 Proposed 8x8 Proposed 16x16 Lin 4X4 Lin 8X8 Lin 16X16 Kim Kim*
Franco Rana
(d) Average of Middlebury Stereo 2006 Datasets of 21 image
Figure 5.14: Hamming distance comparison of proposed scheme with existing schemes [55,62,76,101] against salt & pepper noise attack at Stereo-JPEG compression at quality 75 for right view.
time complexity can be written asO(Ih×Iw).