Image recovery using improved total variation regularization

We introduce generalized regularization functionals to overcome the practical problems associated with current total variation (TV) penalty. Specifically, we extend the TV scheme to higher order derivatives to improve the representation of smoothly varying image regions. In addition, we introduce a rotation invariant anisotropic TV penalty to improve the regularity of the edge contours. The validation of the scheme demonstrates the significantly improved performance of the proposed methods in the context of compressed sensing and denoising.