Computationally efficient sparse algorithms for tomographic PIV Reconstruction

“Back to physical particles” is the idea behind this paper. We introduce the physically sound Point Spread Function (PSF) model for tomographic PIV reconstruction, that aims at sparse and spiky reconstructions. Using the PSF, we show that the tracer particles are quasi-systematically in the direct vicinity of local maxima of reconstructions obtained with MLOS. Taking advantage of both the PSF model and this local maxima extraction (which we call “LocM”), we manage to dramatically reduce the dimensionality of the reconstruction that follows. For this reconstruction, we consider the SMART algorithm applied on this local maximum restriction, and also introduce a sparse algorithm, CoSaMP. We investigate the performances of LocM-SMART and LocM-CoSaMP, compared to the traditional, blob-like, MLOSSMART used in particular by [2]. LocM-XX methods combine dramatic computational gains and a very good reconstruction accuracy. LocM-CoSaMP in particular reaches equivalent or better performances than classical MLOSSMART especially if a voxel to pixel ratio equal to 1/2 is adopted.