Classification With Dictionary Learning and A Distance Barrier Promoting Incoherence

We present a new approach to the incoherent dictionary learning problem using a barrier function that promotes incoherence. This function has a context-dependent quadratic term and a distance barrier term that can be used in both local and global structures. This strategy achieves better results in terms of error representation and incoherence of the dictionary, compared with the standard problem. We demonstrate on several datasets that this function can improve the performance of dictionaries in classification problems.