In this article , the research of image process based on pattern recognition consists of three phases : 1 ) phase of image segmentation : a method of multi - layer and multi - character ' s segmentation in transform field based on knowledge is proposed 本文对基于模式识别的图像处理过程分三个主要阶段进行研究： 1 )图像分割阶段：提出基于知识的多层多特征变换域(特征空间)的判别分割方法。
Experimental results show the hidden information capacity of this algorithm is one time more than lsb algorithm , and much more than all kinds of transform domain algorithms , but computing complexity is much lower than transform domain 实验表明在不影响图像视觉效果的前提下，其信息隐藏量比lsb算法大一倍以上，比各种变换域算法的信息隐藏量更是大很多，而计算复杂度比变换域算法低很多。
Since two patches are used , it can detect the embedded watermarks without the original host images , which makes it a blind watermarking algorithm . finally , the paper shows the further research direction in audio digital watermarking 最后，借助图像数字水印中的patchwork算法思想，提出了一种用于dct变换域的音频数字水印方案，该技术利用两个样值的平均值和方差实现了水印信息的盲检测。
This paper is mainly devoted to the problem of how the correlation of the adjacent coefficients in the wavelets transform and the multiscale mean values are used in image - layered coding and transporting . first , the wavelet transform is studied 本文主要研究了如何利用图像小波变换域中的相邻系数间的相关性、利用均值块来实现小波系数的分级编码和传输，从而实现了图像的分级传输功能的问题。
Taking wavelet domain as transform domain , our method can deal with multi - kind watermark such as string and logo . consulting the idea of the second watermark technique , we divide wavelet domain coefficients by characters of image to decide the embedding regions and intension 该算法以小波域为变换域，采用扩频技术将水印在小波域中嵌入到载波图像中，并可处理多种形式(字符型和图标型)的有意义水印。
