A COMPUTATIONAL APPROACH TO FOURIER SYNTHESIS FROM NON-UNIFORM SAMPLES

Abstract

Gaussian Process Regression Image Synthesis is a proposed numerical method intended for image reconstruction via Inverse Fourier Transform directly from a partial non-uniform sampling of the Fourier Space. The motivation for this comes from the field of radio interferometry, where the measurements are obtained within the Fourier space and part of the algorithm for the reconstruction of the images relies on altering the samples in order to meet the restrictions of the Inverse Fast Fourier Transform. In this work a proposed GPR based algorithm is described, a prototype of it is implemented and then empirically evaluated against the current FFT-based approach. The results of these experiments are positive enough to expect that, with further research and development, this method can become a competitive contender in this field