Abstract: I'll discuss two recent research directions in total variation (TV) image restoration. The first is the use of the L1 norm (instead of L2) to measure data fidelity. It turns out that the L1 model has some very interesting and subtle properties which lead to better contrast preservation and multiscale decomposition, as well as the ability to find global optimum of certain geometric optimization problems. The second is the use of duality in deriving fast and robust iterative algorithms for solving TV image restoration problems