基于自适应稀疏字典学习的图像盲分离技术研究

Blind separation of image source based on adaptive sparse dictionary learning

稀疏性对多通道观测的信号源分离具有重要作用,现有的算法只能有效的分离稀疏域已知的信号源。为了解决这个问题,可以将字典学习与信号源分离相结合。定义一个代价函数,采用ELad等人提出的降噪方法使其最小化。由于直接采用降噪方法使代价函数最小化难于实现,可以采用一种分级的字典学习方法,根据每一个信号源建立一个自适应的局部字典。该方法能够在噪声条件下有效的提高信号源分离的效果,仿真结果表明了该方法的有效性。

Sparsity is very useful in source separation of multichannel observations. If the underlying sparse domain of the sources is not available, then the current algorithms will fail to successfully recover the sources. To address this problem, a solution via fusing the dictionary learning into the source separation is proposed. Firstly, a cost function is defined based on this idea and an extension of the denoising method is proposed in the work of Elad to minimize it. Due to impracticality of such direct extension, a hierarchical approach is used. A local dictionary is adaptively learned for each source along with separation. The quality of source separation is improved even in noisy situations. The results of our experiments confirm the strength of the proposed approach.