24小时热门版块排行榜    

查看: 419  |  回复: 1
本帖产生 1 个 翻译EPI ,点击这里进行查看

yanglanying

金虫 (正式写手)

[求助] 翻译

The time dimension of the video medium makes imperceptibility
of data embedding more challenging compared to
still images. Maintaining the distortion or peak signal-to-noiseratio
(PSNR) at a consistent level across all frames of a
video stream to avoid flickering is a concern when video data embedding is considered. To that respect, we recall that t he MSE distortion per block to the nth frame isDn.We understand that when we minimize the average distortion over all N framesthe distortion pertinent to each frame may vary.To address the above technical concern, we consider a sub-optimal embedding approach that exploits the strongsimilarity of adjoint frames. For such frames, the transformdomain hosts have similar statistical properties. Therefore,
instead of using Rx, we calculate (si, ci), i = 1, . . . , K,
based on Rx,1, the autocorrelation of the host vectors of
the first frame, and embed data in all frames using (si, ci),
i = 1, . . . , K. Thus, real-time operation can be supported
since data embedding starts right after collecting the first
frame and data-embedded video can be delivered immediately.
To provide consistent distortion across frames, the embedding
amplitudes Ai,n for the ith message in the nth frame is
scaled/adjusted by Ai,n = 1, . . . , N, which assures equal distortion per-block to
each frame. We note that, if frames are similar in second-order
block statistics, then Rx,1 ≈ Rx,2 ≈ . . . ≈ Rx,N ≈ Rx and
sub-optimal embedding approaches global video optimal
embedding.

» 猜你喜欢

下一个目标
已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖

genhunter

至尊木虫 (著名写手)

【答案】应助回帖

★ ★ ★ ★ ★ ★ ★
yanglanying: 金币+5, 翻译EPI+1 2015-05-29 21:41:17
yanglanying(RXMCDM代发): 金币+2 2015-06-06 18:12:53
视频中的时间维度使数据的隐蔽性嵌入比静止图像更具挑战性。如何在视频所有画面中保持失真或峰值信噪比(PSNR)在一致的水平,同时避免闪烁是视频数据嵌入过程中值得关注的问题。在这方面,我们记得,每个片段到第n画面为止的MSE失真是Dn。我们知道,当我们将平均失真在整个N 个画面中降低到最小时,每个画面相对的失真度可能会有所不同。 为了解决上述技术问题,我们利用相邻画面有较强相似性,考虑一个仅次于最优嵌入方法。由于这样的画面在变换域主机有类似的统计特性, 因此,基于RX,1,我们计算(Si,Ci),i = 1,...,K,主机的向量的自相关第一画面,并在所有画面中嵌入数据使用(Si,Ci),i= 1,...,K,而不是用Rx. 这样可以支持实时操作,因为数据嵌入在收集第一画面
之后就开始, 而且数据嵌入的视频可以立即交付使用。为了提供在所有画面中保持一样的失真,为第n画面中的第i个信息而嵌入振幅Ai,n经Ai,n = 1,...,N,作为调整, 以保证每段中每幅画面都有相同的等失真度。我们注意到,如果画面在二阶分组统计中相似,那么RX,1≈RX,2≈...≈RX,N≈RX仅次于最优嵌入法与全球视频优化方法嵌入法接近。
离成功还差一点
2楼2015-05-29 16:56:34
已阅   回复此楼   关注TA 给TA发消息 送TA红花 TA的回帖
相关版块跳转 我要订阅楼主 yanglanying 的主题更新
信息提示
请填处理意见