bartorch.prox.rof

Contents

bartorch.prox.rof#

bartorch.prox.rof(input, lamda, axes)#

Total-variation (Rudin-Osher-Fatemi) denoising along axes.

Approximates argmin_x 0.5 ||x - y||_2^2 + lamda * sum_r ||(D x)_r||_2, where D stacks periodic first differences along axes and the inner norm is over the derivative directions (isotropic TV of complex values). BART runs 50 ADMM iterations with rho = 0.1, so a large lamda does not reach the constant minimizer; the mean of input is preserved.

Parameters:

  • lamda (float) – Weight of the TV term relative to the halved squared data error, in the units of input’s values.

  • axes (int or tuple of int) – Axes along which differences are taken.

Contents