Generalized Tikhonov regularization method for large-scale linear inverse problems.

In this paper we propose a regularization of general Tikhonov type for large-scale ill-posed problems. We introduce the projection method of iterative bidiagonalization and show that the regularization parameter can be chosen without prior knowledge of the noise variance by using the method of balancing principle. An algorithm implicate the efficient numerical realization of the new choice rule. Numerical experiments for severely ill-show benchmark inverse problems show that new method is effective compared with other criterions. [ABSTRACT FROM AUTHOR]