Correntropy based robust sparse representation for erratic noise suppression

Random noise in seismic data severely corrupts the useful signal and can often lead to the misinterpretation. Conventional sparsity-based methods using the mean square error (MSE) criterion depend on the Gaussian assumption of the noise distribution and have demonstrated their effectiveness in suppressing Gaussian random noise. However, erratic noise, that has the high-amplitude feature and non-Gaussian distribution, may lead to the performance degradation. In this abstract, a novel correntropy based robust sparse representation method, is presented to improve the erratic noise suppression. Specifically, the correntropy induced metric (CIM) is utilized to adaptively assign weights for noisy data depending on the erratic noise level. And also, the sparse prior is imposed on the CIM to learn robust and sparse representations. Using the Half-Quadratic optimization technique, the correntropy based optimization can be transformed into an 1 l -constrained quadratic problem, which can be efficiently solved by a standard optimization method. Experiments on synthetic and field data show the better results of our presented method in comparison with conventional approach.