Acceleration of PDS-Based High-Dimensional Signal Restoration

In this study, acceleration techniques for the primal-dual splitting (PDS) algorithm, one of the optimization algorithms used to restore high-dimensional signals, are proposed. In general, it is inevitable that signals observed by sensors are incompletely measured an d/or contaminated by noise. Thus, there is a demand for techniques to restore signals obtained in inadequate environments. Restoration of high-dimensional signals such as volumetric data often takes several hours or even days. Therefore, speeding up the restoration process is an important issue. This study addresses this issue in two different ways. First, acceleration methods are introduced into PDS. Second, their fixed-point implementations are introduced. In order to verify the significance of the proposed approach, the restoration performance and processing speed are evaluated and compared with previous techniques.