The attack is based on redundant motion-compensated temporal wavelet transform (MC-RTWT) of watermarked frames. 98 4.9 Oracle performance in static and dynamic areas from sequences in un-.
Digital Watermarking
Applications
Broadcast monitoring: A watermark embedded in commercial advertisements can be used to monitor whether they are being broadcast in accordance with the contract. Furthermore, by analyzing the detected watermark, the exact location of the change can be determined [CMB01].
Trade-Offs
Video Watermarking
Watermarking of Uncompressed Video
DOUBLE VIDEO ANNOTATION 5If the watermark insertion force is increased to increase the consistency, the fidelity of. The reference watermark is then set to generate the watermark for the entire video frame.
Watermarking of Compressed Video
In the first approach, the watermark embedding algorithms modify the motion vectors in the compressed stream. However, the disadvantage is that the embedding and detection of the watermark is completely dependent on the structure of the compressed stream.
Attacks on Watermarking Systems
Attack Classification
But in the security attack, the attacker gains some knowledge about the watermarking system that can be effectively exploited to defeat it. Depending on the knowledge of the watermarking system that the attacker uses, security attacks can be divided into cryptographic and protocol attacks.
Collusion Issues in Video Watermarking
COLLABORATION PROBLEMS IN VIDEO WATERMARKING 10attacker gains some knowledge of the embedded watermark before the attack. MOTIVATION OF THE PRESENT WORK 11Video watermarking is vulnerable to two types of collaboration: inter-video and inter-frame or.
Motivation of the Present Work
Outline of the Thesis
Basic Attacks
As a result, for each video frame, it is possible to obtain a rough estimate of the underlying watermark in each image. However, if the watermark is temporally redundant, it is possible to obtain a refined estimation of the watermark by combining these uncorrelated individual approximations [HMY00, SKH02].
Countermeasures to Basic Attacks
In static scenes, large window widths can be used without degrading the visual quality of the attacked video. Therefore, a lower window width should be used to maintain the visual quality of the attacked video.
Frame Temporal Filtering After Registration (FTFR)
The watermark is added to the temporal wavelet frames by using the frequency masking and spatial masking features of the human visual system to increase the robustness of the watermark. The dynamic part of the watermark will prevent Type I collusion and the static part will survive Type II collusion.
Proposed Attack
Motion-Compensated Temporal Wavelet Transform
Thus, there is a direct link between ghost artifacts in low-pass time frames and energy in high-frequency time frames. In this method, the lift update steps are weighted by energy in high-pass time frames.
Motion-Compensated Frame Temporal Filtering (MC-FTF) Attack
Similar to 2D-RDWT, the MC-TWT introduced in the above subsection can also be overcompleted by removing the downsampling operation. PERFORMANCE ANALYSIS 21In the proposed MC-FTF attack, the video frames are temporally filtered using the lift-based MC-.
Performance Analysis
Performance Against Different Watermarking Schemes
Thus, the attack will have a similar effect as the FTF attack on the detectability of the watermark. So, the performance of the proposed attack against the SS-1 watermarking scheme depends on the moving areas in each frame.
Impact of Motion Modelling
On the other hand, in a dynamic scene, the success of the attack depends on the content of the scene's movement. PERFORMANCE ANALYSIS 25 the effect of these properties on the performance of MC-TWT-based video coding has been extensively investigated [BKZV05].
Effect of Changes in Motion Vectors
Independent estimation of these motion fields from the corresponding frames will result in the motion inversion error. Consider the estimation of the motion vector for the same block after embedding an additive watermark given in equation (2.1.1).
Experimental Results
Performance Against SS Watermarking
Except for the Antibes sequence, the visual quality of the affected sequence decreases with increasing level of decomposition. Finally, the performance of the attack with the adaptive update step of the MC-RTWT is presented in Table 2.3. The good visual quality of the attacked video with the adaptive update step is evident from these figures.
Performance Against SS-1 Watermarking
It is observed that increasing the accuracy of the motion vectors increases the visual quality of the attacked sequence. As expected, the adaptive update step significantly increases the PSNR performance of the attacked video with a slight decrease in NC performance. The PSNR performance of the attacked sequence is better in the case of HVSBM motion estimation compared to FSBM.
Discussion
As noted earlier, its success lies in exploiting the motion in the video frames that is not taken into account during watermark embedding. The motion compensation step in the MC-FTF attack aligns similar areas in different frames before temporal filtering. Thus, any countermeasure against the MC-FTF attack should include the motion information during watermark embedding so that the similar areas along the motion trajectories carry correlated watermarks.
Watermark Embedding Using Motion Information
Motion-Coherent Watermarking
When the camera is static, the same part of the 3D background is projected in all frames except the areas covered by moving objects. On the other hand, when the camera moves, different parts of the background are projected in different frames. The first step in the watermarking process is to generate a mosaic view of the background, which provides a snapshot of the scene.
Compressed Domain Video Watermarking
Video Coding Standards
In the 't+2D' approach, the MC-TWT is applied for the temporal decorrelation of the frames, followed by the 2D wavelet transform of the temporal subbands for spatial decorrelation. The MPEG started the exploration of 3-D wavelet transform-based video coding and the 't+2D' structure was adopted as the first working concept [R.
Prior Work
PROPOSED APPROACH 45problem addressed by the authors is the increase in bit rate due to adding watermark.
Proposed Approach
MC Watermarking of MPEG-2 Streams
In this scheme, the watermark needs to be embedded only in the I-frame of each GOP. 3 Compute the watermark that will propagate to the last P-frame of the current GOP. 4 Decode the I-frame and the motion vectors in the next GOP. In the unconnected points, add the watermark corresponding to the same location in the I-frame of the first GOP.
MC-TWT Domain Watermarking
PROPOSED APPROACH 51 Similarly, if the watermark is added to low-pass temporal frames at the second level of MC-TWT decomposition, the resulting watermarked frames are . Under the assumption of the composition of motion operators, i.e. Mk1→k2(Mk2→k3(n)) =Mk1→k3(n), it can be shown that watermarking in low-pass frames of L-level MC-TWT decomposition results in. It can be further shown that the first relation in equation ( 3.3.11) holds regardless of the choice of wavelet filter.
Performance Analysis
- Coding Parameters
- Motion-Coherency in the Watermark
- Improved Robustness to Compression
- Robustness to Known Inter-Frame Collusion Attacks
Because of the precise quantization in I frames, its effect on the watermark can be neglected. The quantized watermark will extend to the intercoded frames and thus the motion consistency in the watermark. The effect of recoding on the embedded watermark therefore depends on the contribution of the watermark in terms of prediction errors.
Experimental Results
Performance Against MC-FTF Attack
This explains the decrease in NC value with the increasing degradation rate of MC-RTWT. Comparing the results presented in Table 3.1, we can see that changing the motion estimation algorithm slightly reduces the performance against the MC-FTF attack. This is due to the difference in the motion vectors estimated using the two techniques.
Performance Against WER and FTF Attacks
The motion vectors for the attack are estimated using the FSBM from (a) the host sequence and (b) the watermark sequence. The motion vectors for the attack are estimated using the HVSBM from (a) the host sequence and (b) the watermark sequence. The motion vectors for the attack are estimated from the watermark sequence using the HVSBM.
Discussion
Today, this assessment relies on a computationally expensive procedure: (1) a series is watermarked, (2) the watermarked series is subjected to different interframe collusion attacks and (3) the detector checks the presence of the watermark in the attacked series. As shown in the previous chapter, the motion coherence in the watermark is a sufficient condition to guarantee the resistance to interframe collusion. Thus, the assessment of the motion-coherence in the watermark is a sufficient test for inter-frame collusion resistance.
Proposed System
- Motion-Compensated Prediction
- Statistical Modelling of Prediction Error Frames
- Estimation of Distribution Parameters
- Oracle Design
We suggest that such an approximation is valid for the local variance estimates of the host PEFs. The local variance histograms of the PEFs are studied for the sequences encoded using the MPEG-2 standard. The gamma distribution is a reasonably good approximation to the histograms of the local variance estimated from the PEFs of a compressed sequence.
Experimental Results
Experiment I: Comparison of Feature Vectors
Similarly, the null blocks in the PEFs obtained from the compressed sequences are excluded from the calculation of the Gaussianity features. Due to better motion compensation, both the oracles perform better in the case of the B-PEFs compared to the P-PEFs. The performance of the oracles is better in the case of B-PEFs compared to the P-PEFs.
Experiment II: Effect of Compression
EXPERIMENTAL RESULTS At 90 bit rate, the oracle detects more number of P-PEFs compared to B-PEFs. However, no direct correlation can be established between oracle performance and NC values. For example, the oracle detects 76.3% of P-PEFs from sequences with ID5 encoded at 2Mbps, where the NC value is 0.58.
Experiment III: Hybrid MC watermarking
Discussion
The first problem addressed by the thesis is how to improve the performance of inter-frame collusion attacks by exploiting motion information in video frames. The reported experimental results confirmed the robustness of the proposed MC-TWT-based watermarking against MC-FTF attacks and other collusion between frames. The oracle is based on statistical analysis of motion-compensated prediction error (PEF) frames.
Tracks for Future Work
I Security and Watermarking of Multimedia Contents II, bind 3971 af Proceedings of SPIE, side 346-357, januar 2000. In Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing, side 605-611, juni 1995. In Proceedings af den internationale IEEE-konference om akustik, tale og signalbehandling, bind 6, sider marts 1999.
Generic watermark embedding process
Generic watermark detection process
A watermarked frame from the Stefan sequence and the corresponding MC-FTF at-
A watermarked frame from the Foreman sequence and the corresponding MC-FTF
Detector performance of the SS and SS-1 watermarking schemes after subjected to the
Mosaicing-based watermarking
Propagation of the watermark in a GOP
Propagation of watermark across GOP boundary
A sample frame from the Foreman sequence marked using the MPEG-2 watermarking
A sample frame from the Antibes sequence marked using the MPEG-2 watermarking
Sample frames from the Coastguard and Foreman sequences after the FTF attack
Local variances of a P-PEF from the Mobile sequence and the plot of the sorted array
Distribution of the local variance of a P-PEF
Scale-Shape plots corresponding to the P-PEFs
Local variance histogram of a P-PEF from the Foreman sequence coded at different
Local variance histogram of a B-PEF from the Foreman sequence coded at different
Block diagram of the proposed oracle
