Spectral Constrained Frequency Selective Extrapolation for Rapid Image Error Concealment
Nils Genser, Jürgen Seiler, André Kaup · 2018
High quality error concealment plays a crucial role in image and video processing. In general, those algorithms introduce a high computational load as they calculate complex models to estimate the missing samples. In the following, a highspeed algorithm is introduced, which succeeds the state-of-the-art Frequency Selective Extrapolation. Therefore, two novel spectral constraints are introduced in this paper. Firstly, processing the DC part of the distorted spectrum separately allows to decrease computational complexity and to increase the reconstruction quality. Secondly, by exploiting spectral properties, only a subset of basis functions has to be processed. Moreover, complex-conjugated pairs of basis functions are selected to generate a symmetric Fourier spectrum and an according real-valued output signal. Taking these constraints into account, a PSNR gain of up to 0.36 dB is achieved compared to the state-of-the-art algorithm, while the execution speed is approximately doubled.