基于主成分分析的字典学习

在??奇异值字典学习方法的基础上,结合主成分分析方法提出了??主成分分析字典学习方法。该方法取代了??奇异值分解(KSVD)方法中对误差项直接进行SVD分解来更新原子,取而代之的是通过对误差项进行PCA分解,提取其主成分作为字典中原子的更新。仿真结果表明,与KSVD字典学习方法相比,所提出的方法字典学习效果更好,对训练样本的表达误差更小,学习字典更能表达训练样本的特征。

Dictionary learning based on principle component analysis

The K-principle component analysis dictionary learning method is proposed based on the K-singular value
decomposition(KSVD) method and the principle component analysis(PCA) method. Instead of the SVD decomposition to
the error in the KSVD method, the atoms of the dictionary of the method are updated by distilling the principle component
of the PCA decomposition. Simulation results show that, compared with the KSVD method, the better learning effect is
achieved, the representation error is small, and the learned dictionary reflects the features of the training data much better with the KPCA method.