H.264视频中基于速率失真理论的高效水印算法

116 浏览量更新于2024-08-26 收藏 450KB PDF 举报

本文主要探讨了一种基于速率失真理论的视频水印算法，由刘连山、郑永国和王思博三位来自中国山东科技大学的研究者提出。该研究关注于如何在H.264/AVC这一下一代视频编码标准下设计一个高效且鲁棒的水印嵌入策略。首先，作者们注意到在编码序列中，相邻的块被分为一组进行处理。这种方法的关键在于，他们将每个组中的运动向量（Motion Vector, MV）作为嵌入水印的重要载体。运动向量在视频压缩过程中扮演着关键角色，因为它用于描述像素之间的空间移动，对图像质量的影响相对较小，适合隐藏秘密信息。接着，他们提出了一个量化过程。通过对每个成员宏块（Macroblock）的运动向量进行单独修改，并计算修改后带来的速率失真变化。速率失真理论在这里起到了关键作用，它衡量的是编码效率和解码质量之间的权衡。选择改变运动向量后速率失真变化较小的宏块作为候选块，这有助于在保持视频质量的同时尽可能地隐藏水印。然后，候选块的选择不仅考虑了当前帧，还与前一帧的运动向量进行比较，通过调整宏块的运动矢量分量来嵌入水印。这种策略保证了水印的嵌入不引起明显的视觉干扰，同时还能抵抗帧删除攻击，即在视频剪辑过程中，由于某些帧被删除，传统的水印可能会丢失，而这种基于速率失真理论的方法则能够更好地维持水印的完整性。实验结果显示，该算法在面对帧删除等常见的篡改攻击时表现出良好的鲁棒性，确保了视频数据的完整性和水印的安全性。因此，这项研究对于视频版权保护和防伪有着重要的实际应用价值，尤其是在H.264/AVC这样的高效率编码环境下。总结来说，这篇文章的核心贡献是提出了一种利用H.264/AVC编码特性，结合速率失真理论，实现视频水印嵌入的创新算法。这种算法在提高视频编码效率的同时，提供了有效的水印保护，对于视频内容的版权管理和安全验证具有重要意义。

The video watermark algorithm based on rate distortion theory

Lianshan Liu

Shandong University of Science and

Technology, Qingdao, China

e-mail: lshliu6042@163.com

YongGuo Zheng

Shandong University of Science and

Technology, Qingdao, China

e-mail: zhengyg206@163.com

Sibo Wang

Shandong University of Science and

Technology, Qingdao, China

e-mail: wang_sibo@live.cn

Abstract—A novel watermarking algorithm based on rate-

distortion theory was proposed. Firstly, every two adjacent

blocks in coding sequence were divided into one group. The next

step was to separately modify the motion vector of member

macroblock and compute the change of rate distortion. Then the

macroblock with smaller change of rate distortion was chosen as

the candidate macroblock. The third was to seslect the candidate

macroblock from previous and change the motion vector

component of macroblock in order to embed the watermark.

Experimental results show that the proposed scheme was robust

to frames deletion.

Keywords- video watermark; motion vector; rate-distortion;

H.264

I. INTRODUCTION

Next-generation video coding standard H.264/AVC is the

best video coding standard

[1]

, which not only has excellent

compression performance, but also has a good network affinity.

The applications of H.264 become more and more abundance,

so that the video watermark algorithms based on H.264 are

required. Due to the high compression ratio of H.264, most of

the watermark information has been removed as redundant

data, so that previously proposed methods can not be directly

applied to the watermarking algorithms using the H.264 coding

standard.

A number of video watermark schemes based on H.264

coding standard have been proposed. For the intra-prediction

mode, Dong-Wook Kim et al.

[2]

considered that the proposed

algorithms at present can not extract the original watermark

after the videos have been attracted. For the inter-prediction

mode, Wang Pei et al.

[3]

recently proposed a scheme using to

change the position of the optimal matching block in the

motion estimation to embed the watermark. In this scheme, the

search area was divided into two parts. If the current watermark

bit information was 1, then the optimal matching block was

searched in the left region; If the current watermark bit

information was 0, the optimal matching block was searched in

the right region. In the extraction, the watermarks were

extracted according to the motion vectors. This method is

simple, but the optimal matching block may be searched the

other region not the original region, and the motion vectors

may be modified too much. Najla Mohaghegh et al.

[4]

managed

the embedding process using a threshold. If the amplitude of

motion vector was greater than the threshold, the watermark

can be embedded, then modify the larger motion vector

component to embed the watermark. Otherwise the macroblock

was ignored. The scheme complexity is low, but the choice of

threshold tends to experience value. The watermark is

embedded only according to the amplitude of motion vector, so

that video quality may be affected more seriously. Cong-Van

Nguyen et al.

[5]

firstly selected the macroblocks meeting certain

conditions, and embeded the watermark with modifying the

larger component of motion vector, then computed the number

of same pixels between original macroblock and the

macroblock embedded watermark in all macroblocks of a

frame, and the next was to select the smaller value of

macroblock to embed the watermark. The algorithm takes into

account that the modification will increase the residual between

original macroblock and prediction macroblock. However, the

description of motion vectors, reference frames, B frame

direction and macroblocks partition mode also need storage

space. The algorithm does not take into account the alteration

of the above storage space after the motion vector has been

modified.

There are a lot of macroblocks in a video, which means

there are sufficient modification options. If we only consider

the single macroblock, it is difficult to take full advantage of

the richness and the video quality may be affected more

seriously.

The algorithm proposed in this paper choosed two adjacent

macroblocks in one video sequence as a group. In order to

reduce the influence to video quality, the change rate of rate-

distortion cost of two adjacent macroblocks should be

computed separately, and the smaller macroblock to embed

watermark should be selected.

The remainder of this paper is organized as follows.

Information rate distortion theory are presented in Section 2.

Section 3 is devoted to the watermark algorithm. Section 4

analysises the simulation and results. Section 5 gives a brief

conclusion of the work.

II. INFORMATION

RATE-DISTORATION

Information rate distortion theory is the theoretical basis of

the data encoding. Information rate distortion function can be

expressed as follows:

(|)

() min{(;)}

jj D

Pv u B

DIUV

∈

(1)

where R is transinformation rate; D is the distortion

permission; I(U;V) denotes average mutual information; B

denotes the test channel set with D; and P(v|u) represents

channels.

In the case that information source and D are fixed, rate

distortion function R(D) can choose a test channel to promise