Optimized Compressed Sensing for Curvelet-based Seismic Data Reconstruction

Compressed sensing (CS) or compressive sampling provides a new sampling theory to reduce data acquisition, which says that compressible signals can be exactly reconstructed from highly incomplete sets of measurements. Very recently, the CS has been applied for seismic exploration and started to compact the traditional data acquisition. In this paper, we present an optimized sampling strategy for the CS data acquisition, which leads to better performance by the curvelet sparsity-promoting inversion in comparison with random sampling and jittered sampling scheme. One of motivation is to reduce the mutual coherence between measurement sampling schemes and curvelet sparse transform in the CS framework. The basic idea of our optimization is to directly redistribute the energy in frequency domain making original spectrum easily discriminated from the random noise induced by random undersampling, while offering control on the maximum gap size. Numerical experiments on synthetic and real seismic data show good performances of the proposed optimized CS for seismic data reconstruction.