Real-Time Frequency Selective Reconstruction through Register-Based Argmax Calculation

TL;DR

This paper introduces a GPU-accelerated, highly parallelized version of Frequency Selective Reconstruction (FSR), significantly reducing computation time and enabling real-time image reconstruction applications.

Contribution

The paper presents a novel parallelized formulation of FSR and a fast argmax calculation method, achieving a 100-fold speed-up for real-time image reconstruction.

Findings

Achieved 100-fold speed-up in FSR computation.

Enabled real-time image reconstruction using GPU parallelization.

Significantly reduced reconstruction time from minutes to seconds.

Abstract

Frequency Selective Reconstruction (FSR) is a state-of-the-art algorithm for solving diverse image reconstruction tasks, where a subset of pixel values in the image is missing. However, it entails a high computational complexity due to its iterative, blockwise procedure to reconstruct the missing pixel values. Although the complexity of FSR can be considerably decreased by performing its computations in the frequency domain, the reconstruction procedure still takes multiple seconds up to multiple minutes depending on the parameterization. However, FSR has the potential for a massive parallelization greatly improving its reconstruction time. In this paper, we introduce a novel highly parallelized formulation of FSR adapted to the capabilities of modern GPUs and propose a considerably accelerated calculation of the inherent argmax calculation. Altogether, we achieve a 100-fold speed-up, which enables the usage of FSR for real-time applications.