日常生活中，我们经常会接收到各种各样的混合信号，而我们可能只需要混合信号中的部分源信号或其中某一个源信号，所以对盲源分离的研究是很有必要的。盲源分离是研究如何从混合信号中分离出需要的源信号；当从混合盲信号中成功分离出各个独立的源信号后，如何更高效快捷地传输这些源信号成为我们面临的问题。 本文针对图像盲源分离问题，首先研究了盲信号分离在图像处理领域中的应用；接着以超宽带图像通信为例，提出一种新的二阶段欠定盲源分离方法——阶梯图-最小角度法。该方法在第一阶段即混叠矩阵估计阶段，提出用阶梯图对均匀抽样的数据点进行聚类计算，建立了一种估计混合矩阵的工程化模型，它可以直接估计出源信号个数，并且得到精度较高的混合矩阵估计值；在第二阶段借助最小角度分离准则估计出源信号，从而获得较好的分离性能。最后在成功地分离出各源图像信号后，把各图像信号归类研究；重点研究了图像分类原则，以及归类后在特定的压缩算法下给每类图像选择一个合适的母小波进行压缩，达到在通信系统中更高效快速地传输图像。实验结果表明所提方法具有一定的优越性。

In everyday life,we often receive a variety of mixed signals,and we may just need one or parts of the source signal. So it is necessary to study BSS (blind source separation). The blind source separation is how to isolate the original signal from the mixed signal. When we separate successfully the signals from the observation data and recover the source signals in the end, the problem we face is how to quickly and efficiently transfer the source signal. In order to deal with image BSS issue,this thesis firstly studies the applications of blind signal separation in the field of image signal processing. Secondly taking the UWB image communication as an example, it presents a novel two-stage underdetermined blind source separation method named stairstep graph-minimum angle. In the first stage of estimating matrix estimation stage, cluster calculation is performed on uniform sample of the mixed data point by using stairstep graph,thus one engineering model for estimating the mixing matrix has been established,which can directly estimate the number of source signal and get high precision estimate of mixing matrix. In the second stage,the source signal is estimated by way of the minimum angle criterion for better separation performance. In the end,after the source image signals are seperated successfully,the images are divided into specific categories, focusing on the classification criteria and a associate mother wavelet for a given compression algorithm. So that more rapid and efficient transmission of images can be obtained in the communication system. Experiments results demonstrate the superiority of the proposed technique.