next up previous [pdf]

Next: Acknowledgments Up: Li et al.: Robust Previous: Velocity Analysis as Inversion


An L1-type optimizer is required to retrieve a blocky model or extract information from noisy data. To avoid the troublesome parameter tuning of IRLS, we develop a hybrid-norm Conjugate Direction solver with straightforward parametrization to achieve the goal in L1 sense. Tests of this solver on three different problems, with both synthetic and field data, show that this hybrid-norm solver is robust and simple to use. These promising results encourage us to apply this solver on large-scale, real-world problems.