基于压缩感知的欠定图像盲源分离研究

系统阐述了利用稀疏成分分析(Sparse Component Analysis,SCA)算法进行欠定图像盲源分离。首先在估计出源图像个数的基础上,利用线性聚类估计混合矩阵;其次将压缩感知(Compressed Sensing,CS)应用到恢复源图像中。为了得到自适应的过完备稀疏字典来提高分离效果,提出了利用K均值奇异值分解(K-means Singular Value Decomposition,K-SVD)算法对过完备DCT字典循环迭代训练的思想,并对图像分块处理来减少计算复杂度;最后进行了仿真测试并对分离出的图像进行了分析和进一步处理。

Study on Underdetermined Image Blind Source Separation Based on Compressed Sensing

The paper introduces the underdetermined image blind source separation using sparse component analysis （SCA）algorithm. Firstly, the mixing matrix is estimated using linear clustering based on the estimation of the number of source images. Then,the compressed sensing （CS） is used to resume the source images. To get a self-adaptive over-complete dictionary and improve theseparation efficiency,a design idea is proposed, which implements loop iteration training for overeomplete DCT dictionary based onK-means singular value decomposition （K-SVD） algorithm. The images are divided into blocks to reduce the computational complexity.The simulation test is carried out and the separated images are analyzed and further processed.