Regularization parameter trimming for iterative image reconstruction

Conventional automatic parameter choosing involves testing many parameter values, increasing computing time for iterative image reconstructions. The proposed approach first measures the image quality after each iteration and then predicts the convergence trend corresponding to each value of the parameter. Values unlikely to achieve the best quality upon convergence are trimmed from successive iterations to save time. Experimental results show that our parameter trimming method could reduce the running time of total variation parameter selection solved by Split Bregman iteration by more than 50% when the numbers of iterations and parameter candidates are large.