Block Sparse Dictionary Learning Based on Recursive Least Squares

Based on the over-complete dictionary, the signal can be described as sparse linear combination of atoms. Traditionally, the dictionary learning methods are mostly based on a single atom unit. In our framework, sparse subspace clustering is used to categorize the atoms that have the same sparse expressions into groups to form block structure of the dictionary, and then encode the training signal with the sparse coding algorithm, finally applying the recursive least squares method to update the dictionary. Experiments show that our method converges faster with the same iterations, and the signal reconstruction error rate is better than the traditional methods.