基于稀疏表示的图像分离.pdf

摘 要
图像盲源分离是图像处理的重要课题之一，
像盲源分离是指，在关于图像源及图像传输方式等先验知识缺乏或知之甚少的前
提下，仅由来自传输系统的观测图像来估计未知图像源的过程.
基于形态成分分析和稀疏表示的盲源分离方法的主要思想是：利用混合图像
中各组成成分的形态差异性，，
介绍了图像稀疏表示的基本理论，其中包括图像稀疏表示模型、字典选择算法、
稀疏分解算法，，本文
根据图像多种形态结构的特征，
表示的模型表示多成分的信源信号的方法，得到一种比传统模型更稀疏的表示形
，由于缺乏源图像的先验信息采用较为传统的L1范数刻画不同成分在固定
，新模型
，本文给出了一种新的迭代算法求解
，本文的模型及其算法能够有效地处理含多成分的图像
盲源分离问题.


关键字：图像分离 稀疏表示 形态成分分析 多成分字典
Abstract
Blind separation of images is an important task in image processing, and it has
been a hot topic in this field nowadays. Blind separation of images is to estimate the
unknown sources by using the observed images only from the transmission system, in
which one knows just a little prior information of the transmission way, or even knows
nothing at all.
The main idea of the Morphological Composition Analysis (MCA) and the Sparse
Representation (SR) based blind source separation method is to sparsely represent an
image under different dictionaries according to the morphological diversity of an
image’s components. Firstly, the basic theory of sparse representation theory is
discussed, which includes the sparse representation model, the design of over-complete
dictionary, and the sparse decomposition algorithms. The application of the sparse
representation in multi-channel morphological composition analysis is also discussed.
Secondly, multi-component dictionary is constructed according to different morphology
features of images. We also use multi-component dictionaries to represent an image, and
then obtain the sparse representation of the multi-morphology sources. Because of lack
of prior information of the source images, the sparse measurement by using the
traditional L1 norm can not separate the different components from mixed images. In
order to solve this problem, we use