A Novel Online Dictionary Learning Method from Compressed Signals

Dictionary learning algorithm facilitates a sparse representation of a given set of training signals, which has significant impact on signal reconstruction error in compressive sensing. To reduce the recovery error caused by environmental noise, in this paper, a novel structured dictionary learning method for sparse signal representation is presented. The training signals are collected from compressive data gathering methods. And the self-coherence of the dictionary is punished. In comparison with the DCT basis and the K-SVD method, experimental results verify that the proposed dictionary is more effective to alleviate the recovery error caused by environmental noise.