本文提出了一种具有几何变换撸棒性的动态图像水印技术。一个沿时间轴上的模板嵌入到动态图像序列中的每个水印最小段（Watermark Minimum Segment）内。利用预先嵌入的时间轴模板，可以恢复受几何仿射（affine）变换攻击后的含水印动态图像帧，仿真试验结果表明本文提出的动态图像水印技术，经仿射（affine）变换攻击后仍是撸棒的。

本文提出一种将彩色数字水印镶嵌到原始图像中的方法。利用静态图像的压缩编码技术，将彩色数字水印编码为一系列二值ID 数字，以实现水印的镶嵌。由于水印的嵌入过程是基于原始图像的数字余玄变换（DCT）系数之间的关系，所以水印的提取不需要原始图像。实验结果表明本文提出的数字水印技术，经一些图像处理操作和有损压缩后仍是健壮的。

A video watermarking with robustness against frame's geometrical distortions (rotation, aspect ratio, scaling, translation shearing, and bending) is proposed. The watermark information is embedded into pixels along the temporal axis within a Watermark Minimum Segment (WMS). Since the geometrical distortions operations for every frame along the time xis in a video sequence are the same at a very short interval, the watermark information can be detected from watermarked frames in each WMS subjected to the distortions. Furthermore, adaptive embedding method is proposed for gaining a good quality of the watermarked video. Experimental results show that the proposed technique is robust against common attacks such as rotation, aspect ratio, scaling, translation shearing, and bending of frames, MPEG-2 lossy compression, and color-space conversion.

A multiresolution watermarking method with grey-level watermarking is presented. Both the watermarking and the original image are decomposed into three-level multiresolution structures using different mechanism. The decomposed watermarking information is further coded by error correction coding. Each level of watermark information is embedded into the corresponding level of the transformed image. Experimental results show that the proposed algorithm is robust against common image manipulations.

This paper proposes a wavelet analysis method of non-uniformly parallel offset. Compared with traditional Fourier method, the method is characterized by higher accuracy of non-uniform estimation, less sampling points of the testing signal, and higher precision of error calibration. The results of simulation show that, for an 8-bit quantizer, the residual timing error is reduced to about 0.03 percent of the sampling period.