LAUSR

Abstract Sparse representation is widely used in signal restoration, compression, and so on. And the admiring results got from sparse representation are based on the intelligent dictionary learned from the signals to be represented. The group sparse representation is popular these years because this kind of signal coding method make use of the self-similar information thus it can fit the signal well rather than over fitting the random noise. This paper proposes a new dictionary learning method based on splitting variables for the complex valued signals which can be used to represent the group sparse signals. The proposed method clusters the vectors which are used to train the dictionary firstly, then the object function which summarizes the fitting errors and criteria which measures the group sparsity in the codes is minimized by the variable splitting and augmented Lagrangian methods. The proposed method achieves complex valued dictionary used for group sparsity coding. Experiments are also displayed to prove the efficiency of the proposed method.