Seismic Imaging with the Generalized Radon Transform: a Curvelet Transform Perspective

A key challenge in the seismic imaging of reflectors using surface reflection data is the subsurface illumination produced by a given data set and for a given complexity of the background model (of wavespeeds). The imaging is described here by the generalized Radon transform. To address the illumination challenge and enable (accurate) local parameter estimation, we develop a method for partial reconstruction. We make use of the curvelet transform, the structure of the associated matrix representation of the generalized Radon transform, which needs to be extended in the presence of caustics, and phase-linearization. We pair an image target with partial waveform reflection data, and develop a way to solve the matrix normal equations that connect their curvelet coefficients via diagonal approximation. Moreover, we develop an approximation, reminiscent of Gaussian beams, for the computation of the generalized Radon transform matrix elements only making use of multiplications and convolutions, given the underlying ray geometry; this leads to computational efficiency. Throughout, we exploit the (wavenumber) multiscale features of the dyadic parabolic decomposition underlying the curvelet transform and establish approximations that are accurate for sufficiently fine scales. The analysis we develop here has its roots in and represents a unified framework for (double) beamforming and beam-stack imaging, parsimonious pre-stack Kirchhoff migration, pre-stack plane-wave (Kirchhoff) migration, and delayed-shot pre-stack migration.