Local statistics estimation for rapid frequency selective extrapolation
Nils Genser, Jürgen Seiler, André Kaup · 2017
Fast error concealment algorithms play an important role in image and video signal processing. Within this paper, a novel and highly accelerated Frequency Selective Extrapolation adaption is introduced for rapid image error concealment. The state-of-the-art complex-valued Frequency Selective Extrapolation runs with fixed parameters, e.g., a constant number of iterations, so far. In this paper, the algorithm is accelerated by determining local statistics of neighboring blocks. As a consequence, internal parameters can be estimated to match the processed image content dynamically instead of using predefined constants. Due to this, it is possible to achieve the same reconstruction quality as the state-of-the-art algorithm, while the execution time decreases on average by 41.30 % and at best by up to 57.08 %.